Welcome back to AI in the Real World for another look at AI applications creating real-world value for businesses and consumers.

In this iteration of the blog, I am doing a deep dive into AI usage in the healthcare industry as I’ve seen a lot of progress over the years. I’ve gathered some studies that explore whether generative AI can be used to help doctors daily and whether AI can be more empathetic than doctors. We have known for a while now that AI-driven analysis can reduce cost, reduce errors, and enhance patient outcomes in medical imaging use cases, but here we are focused on the type of communication.

I’d be the first to acknowledge that when it comes to bedside manner, not all doctors are created equal – compare, for example, the affected irritability of Bones McCoy, the (potentially alcohol-driven) geniality of Hawkeye Pierce, and the manifest narcissism of the prototypical “mad scientist”, Victor Frankenstein. But putting that variability aside for a moment, let’s ask the question: “Can AI be more empathetic than doctors?” Without wanting to make this sound like a “clickbait” article, the answer might shock you!

AI assistant vs. Physician responses to patient questions

A recent study that measured physicians’ responses in terms of empathy showed that patients’ messages that were generated by ChatGPT were preferred over those generated by qualified physicians.

I imagine that this raises as many questions for you as it did for me. Who conducted the study? Who preferred these responses? How often did they prefer these responses? Well, it was published by JAMA Internal Medicine (Journal of the American Medical Association), and it doesn’t get much more prestigious or respected than that. The study presented patients’ questions and randomized responses to a team of licensed healthcare professionals who evaluated the quality of the empathy or bedside manner provided. They found an almost 10 times higher prevalence of empathetic or very empathetic responses from the chatbot.

At this point, armed with the knowledge that generative AI has earned a reputation for bias and hallucination, you are probably thinking that this turned out to be a triumph of “style over substance”, of “form over function”. Me too. But no, it turns out that the panel of expert evaluators also indicated that chatbot responses were rated of significantly higher quality than physician responses – they preferred the ChatGPT responses in 78% of cases.

Does this indicate that we should all be ditching expensive medical appointments in favor of online solutions? I think that is a resounding “no”. While the study tells us that the AI responses were consistently (with statistical significance) preferred both for accuracy and empathy, it doesn’t say anything about the quality in absolute terms of the instances where the physician responses were better. Would any of the less accurate ChatGPT responses, for example, have been life-threatening?

If my new digital doctor is getting it wrong, what I want to know is just how wrong that is. The stakes are high in this domain, and few among us would be willing to play medical roulette as a trade-off for a more empathetic Doctor-Patient experience. But this potential for errors does not render the technology useless – in their conclusion, the authors suggest that AI assistance in the generation of draft responses that physicians can edit might be a practical way in which the technology can be applied. They suggest that pending the results of careful clinical studies, this could improve the quality of responses, reduce the levels of clinician burnout and improve patient outcomes and experiences.

Empathetic, sincere, and considerate scripts for clinicians

In a Forbes article from last summer, Robert Pearl, M.D. also explored the topic of whether doctors or ChatGPT were more empathetic and the results aligned with those of the JAMA study. One of the examples shared came from a New York Times article that reported on the University of Texas in Austin’s experience with generative AI.

The Chair of Internal Medicine needed a script that clinicians could use to speak more compassionately and better engage with patients who are part of a behavioral therapy treatment for alcoholism.  At the time, no one on the team took the assignment seriously. So, the department head turned to ChatGPT for help. And the results amazed him. The app created an excellent letter that was considered “sincere, considerate, even touching.”

Following this creation, others at the university continued to use generative AI to create additional versions that were written for a fifth-grade reading level and translated into Spanish. This produced scripts in both languages that were characterized by greater clarity and appropriateness.

Clinical notes on par with those written by senior internal medicine residents

Referencing his recent study, Ashwin Nayak of Stanford University told MedPage Today that “Large language models like ChatGPT seem to be advanced enough to draft clinical notes at a level that we would want as a clinician reviewing the charts and interpreting the clinical situation. That is pretty exciting because it opens up a whole lot of doors for ways to automate some of the more menial tasks and the documentation tasks that clinicians don’t love to do.”

As was the case for the JAMA study, Nayak is not expecting Generative AI to replace doctors, but did he report that ChatGPT could generate clinical notes comparable to those written by senior internal medical residents. Although the study found minimal qualitative differences in ‘history of present illness’ (HPI) reporting between residents and ChatGPT, attending physicians could only identify whether the source was human or AI with 61% accuracy.

So, does generative AI have a promising future with healthcare professionals? Can it be more empathetic than doctors themselves?

Looking at these studies and use cases, I’d say that the deck is stacked heavily in favor of YES. We are still in the infancy of the technology, and the experiments reported here were carried out using general-purpose models – it is pretty much inevitable that once more specialized models become available the results will be even more compelling.

The post AI in the Real World: Generative AI’s role in empathy-driven Healthcare appeared first on High-Quality AI Data to Power Innovation | LXT.

Introducing the founder and managing director of Freshriver.ai, Joti Balani

PHIL:

My guest today began her career as a software engineer before moving through product management, product development, marketing, and consulting with some of the biggest organizations on the planet on their strategies for the implementation of conversational AI.  These include Google, Citigroup, Johnson & Johnson, and American Express. She’s the co-founder of  Women in Voice New Jersey. She’s a member of the All Ladies League. And within G100, a global organization of women leaders whose purpose is the creation of massive opportunities for women worldwide, she is the USA country chair for robotics and automation.

Please welcome the founder and managing director of Freshriver.ai, Joti Balani. Hi, Joti.

JOTI:

Hi Phil, thank you. Thank you for having me.

What does providing AI with emotional intelligence mean for your clients?

PHIL:

It’s great to have you here today. Joti, your company, Freshriver.ai, provides AI with emotional intelligence. What does that mean for your clients and in turn for their clients? How far removed is AI with emotional intelligence from what’s happening in mainstream contemporary AI?

JOTI:

That’s a great, great opening question. So, you know, I began this journey into conversational AI, I five years ago, after exiting a 22-year corporate career in helping build the world’s largest wireless and wireline networks. When I exited and I started this journey, I started to see, you know, where AI was beginning to come in, you know, as you know, AI was coined, created, thought of in 1950s and it’s gone through its winters and it’s come in. It’s now here and it’s here to stay.

And I’ve seen the journey from the front lines actually working with a deploying it for some of the large organizations that you mentioned earlier. Part of what I’m seeing is an evolution where it’s getting better and better, but we still have ways to go. The emotional intelligence part came when I wrote down the manifest or the mission for Freshriver.ai when I first began, is to say AI cannot stand on its own as us humans. We need to take a look at the emotional, ethical, and economic intelligence if we are to have these machines actually do good for humanity. So it was that thought process alongside the fact that if you’re going to have emotional intelligence in these machines, it’s got to be created by humans that have that as their end goal while we still make money. I’m a capitalist, I tell folks, I’m a capitalist through and through, but it cannot be at the cost of emotional and ethical intelligence. It’s really that balance is what I created as the formula and it continues today as the foundation.

And as you can see in the media, as you pointed out, a lot of the fallouts that are coming from this remarkable technology, which by the way, I’m now dividing up into two parts. There’s a pre-GPT era and now a current GPT era, which is what began with the floodgates that opened November 2022, which OpenAI did, right? So, what we’re seeing is fallouts with lawsuits, largely coming from the hallucinations of these models, for example, is a lack of thinking and thought and consideration on the parts of the developers and these large companies, the technology companies in the emotional intelligence is how do you think about the humans that you’re going to serve, right? So it takes humans to make these machines emotionally intelligent in order to serve, humanity, if that makes sense as the framework.

How is mainstream, contemporary AI handling emotional and ethical intelligence?

PHIL:

Yeah, how do you think the mainstream is doing on that? Are we there?  Are they doing the right things?

JOTI:

So, I think this is a journey of a thousand miles and we’ve taken the first step. And the reason that is, and if you look at history, you know, as much as I’ve been deep in the space for the last five years, I’ve also studied anthropology in equal measure, because we’ve seen this happen before with the advent of as early as, you know the invention of electricity, for example, right?

Which took a lot of learnings on, you know, somebody must have put their finger in, you know, a socket and said, oh, we need fuses now, right? To just to use that as an analogy. When you take the first step and you say, wow, this is a remarkable tool. And I know, you for your audience purposes, we need to look at AI as the tooling, not a hammer looking for a nail, which is a mindset issue, right?

So, if you begin with thinking about that, a lot is going to happen because every time human beings find a new tool, they just get at it, right? It’s like, let’s go make some money. It’s the economic intelligence that first kicks in. After which when there’s these major fallouts, and by the way, you know, generative AI is an unprecedented, I’m going to call it technology in this case, or a tool that human beings have not had before. If you think about the web, you think about mobile, you think about cloud, you know, the last 25 years, if I were to frame that from a digital transformation standpoint, you know, it’s all been deterministic software. You know, as a software engineer, I clearly understood that, is that when you do software development, you know that the machines will do exactly what you want it to do.

We are now in the era with these generative pre-trained models where they’re appearing to think for themselves, but they’re not, right? This is still the data that’s being used to train these deep learning models, which by the way, scientists themselves don’t really understand how they are working with it, which is the other risk part that comes in when you don’t understand something. But the bigger news, for everybody, this is what we share with our clients, everybody from the C -suite down to developers, to designers, to product managers, look, this is non-deterministic technology. So, you have to think about it in a very, very different way. So, what we’re seeing happen is people are using it, they’re plugging it, and all sorts of issues are now falling out because they were never considered.

Going back to the framework of the emotional and ethical intelligence, it needs to be done what we believe in a crawl, walk, run. You can’t just plug it and go all out, which is what’s happening right now, right? Everybody’s got access to this tech and unfortunately, so do the bad actors. We just saw a report come out from Microsoft and Open AI where they see hacking activity from, you know, nation state like Russia and China coming in. And that’s a double-edged sword. If you actually look at it, it’s like, oh, wow, that’s a bad thing, But, oh wow! You’re actually monitoring usage, which is now a privacy issue. So, there’s this, all sorts of things are falling out. I’m still an optimist. I do believe they will work themselves out. But the political, social, business institutions, regulatory institutions, and legal institutions have to step up faster. I don’t think those frameworks are moving fast enough. Right now it’s the wild, wild west. So, what you’re seeing really is the wild west emerging first.

PHIL:

Yeah, and I guess that’s not surprising. As you said, that’s human nature that you see this, you see the gold fields opening up and you grab a shovel and go.

JOTI:

That’s right. It’s a FOMO effect as well, right? Everybody wants to be the very first that does it. And, you know, there’s a downside to being the first you could, you could fall, fall over yourself in a very big way. So we advise our clients not to put their brands at risk. What we tell them is you it’s very hard for you to recover once you put something out there. So, you, you know, look at the examples, Air Canada, got fined because their chatbot responded with hallucinatory policies that did not exist, right?

So, you know, this is where it gets really real, is when businesses and enterprises and governments, by the way, start to play with this tech, put it out there, but the 20% of the time that it will not work, it will hallucinate can really cause some really, really big issues. So that’s what we’re worried about on behalf of our clients, right Making sure we’re taking into account the risk management. We have a risk management first framework that says, look, just like the medical, the healthcare, which says do no harm. Those are the first philosophies that we work with. It’s, we make sure whatever we do, whether we’re experimenting with a pilot or going all out with, you know, from a product launch and a roadmap is everybody tests for do no harm.

Data collection and data training in the pre-GPT era vs GPT era

PHIL:

That’s great, thank you. You have said in the past, I’m not sure how far in the past, but you’ve said “Facebook is littered with 100,000 bots that have mostly failed.” Now, the first part of my question is how and why did they fail? And the second part of the question is given the rapid and accelerating rate of change in AI, is this still true and how long will it remain true?

JOTI:

So, I want to go back to the context that I said earlier, right? There was a pre-GPT era and then there’s a current GPT era. I actually had that comment, I made that comment in the first two years that I was in this space. This was the pre-GPT era, right? These language models, these large language models, deep learning, machine learning has been around for a while, right? The problem has been the access to data and content to train them. So, in the pre-GPT era, by the way, there were 2000 vendors in this space, you know, and I have, with my own hands worked with a lot of these technologies early when I was beginning into this, into this field. And I would see that same, I would observe the same thing, right? They’re not working very well. They seem like they were dumb. And if you think about a lot of the virtual assistants that are out there, right? You look at Siri, you look at Alexa, they are not very smart in today’s terms, right?

But you look at these GPT models, the reason for that is it all depended on how much data was available.  And most of those models depended on two things. One, on the clients, if it’s an enterprise deployment or if Facebook was putting it on, they needed to be able to use their user data to go train those models, right? And that was not quite available yet, because these generative pre-trained models really are gas guzzlers, right? If you think about Sam Altman saying, hey, we’re going to need $7 trillion. He’s actually not too wrong because it takes a lot to get access to that data as well as train it and alongside the compute and storage costs, right?

So, in the early days as Facebook being obviously Facebook, you know, early adopters took the tech in the pre -GPT era and put it out there. It didn’t get too far. And they also said to, you know, businesses, hey, look, we have this tool and you can go put it on messenger so that you can work with it with your clients. Well, didn’t work. And so those bots didn’t get used because they were not productive or efficient for people to, for example, go order from the small business. So that was at that point in time. And that is true in the pre -GPT era, the tech was not that great, but what happened over the last five years. Every time we’ve worked in that course, you know, our teams expanded, we’re working with more of these technologies. We saw a change in how good these bots were getting just from the data that was being fed. But the second part of the issue we had first was not enough data.

Second is requiring too many humans because you got to counter the fact that these things don’t work really well. You got to have to hire literally thousands of human beings. If you’re somebody like an Amazon for Alexa, to keep them running and they’re still not very smart. So if you think about the ROI of saving costs upfront when you’re deploying these technologies, it was getting washed away because, the cost was being transferred and hiring human beings, right? So you say, well, how is this going to work? Now enter the whole GPT era.

So, in 2017, there were papers being written about these generative pre-trained models, right? And it takes time to train on the wide swath of what’s available on the web. So think about Reddit posts or the sub-Reddit categories underneath them, right? Entire encyclopedias, dare I say New York Times content, right, which now there’s a lawsuit. But to sweep all the content and data that humanity has created, good, bad, or ugly, took that much time, so, in parallel, there were these models being trained on everything under the sun. And when they got released in November 2022 with a conversational AI interface of chat, you know, a hundred million users later in a month. Now there’s data being generated by more human beings, right? So now we’re off to a start where it looks remarkable.

I’m sure you’ve tried chat GPT or entropic or any one of these, It’s just exploding. People’s minds are blown. But there’s a reason for that. It’s because of the wide swath of data content that it was fed. Great. We’ve made movement as humanity from a technology standpoint. Now you ask me that question, you say, well, where are we right now? Wow, these things are really, really good, but they hallucinate. With such confidence, they will tell you facts that are completely wrong, right? Um, I think it was a Gemini Pro that was just released that it was asked the question: give me, you know, images of senators from the 1800s. It responds, Sure!

Here are the diverse senators in 1800s. And it’s got four senators that are, you know, not white that are on there, which is not true, right? We never had that kind. So when people look at that and you go, well, could you put that, you know, the genie back in the bottle, well you can’t, and you shouldn’t, because there are such remarkable, problems that we can solve.

So, you know, the pandemic’s a great example. The COVID vaccine came to us as fast as it did. And I’m not going to go into, you know, all the health issues controversy right now, but it came to be because of this technology, they were able to find, right, the right protein structure in order to create the vaccine. So we, human beings, have to be cognizant and creative and innovative in finding what kind of problems you want to solve. Don’t go throwing this tech at everything. It’s not meant to solve everything. There are cases where you do need humans handling certain things, but where is the collaboration going to be? That’s really the question. And so you will see a lot of failures. You will see remarkable, spectacular failures coming out, but the march towards innovation is not going to stop. And hopefully we can solve some real problems that humanity been facing that we haven’t been able to tackle so far.

Through this rapid rate of change to generative AI technology, is there any aspect of it that surprised you?

PHIL:

Great. Now we’ve already talked about this rapid rate of change. Through this rapid rate of change, is there any aspect of it that has taken you by surprise? Something you really didn’t see coming?

JOTI:

I think it was how quickly the government jumped in where now the FTC, for example, is starting to move very quickly and starting to put things into place. For example, the deep fake videos, right. And the other thing that surprised me is how quickly other countries jumped on the bandwagon and released their models, which means everybody’s been working on it for this long. And it’s now began this huge war and how quickly that happened.

Have you ever seen the government respond so fast to something in the past? If you think about web, think about mobile, right? It took a while, right? We’re 25 years into the web, we’re 15 years into mobile, 10 years on the cloud journey. You know, everybody had their time to figure out how to go. Right now I’m shocked. The question is, will it move fast enough to stop some of the damage that can be done, while still supporting record. If they release, and I think there’s some laws that are being invoked that could potentially just kill innovation itself. Like for example, tech companies being responsible if a user uses their technology to do something that causes harm. It’s similar to the self-driving car question is when that unfortunate death happens who’s responsible? So, I think we have a lot of gray areas to think about. I just hope there’s not knee jerk reactions. And of course you’ve got all the lobbyists, right? That the tech vendors have seats at the table there.

So, their reactions surprised me, but what’s not surprising me is the way they’re going about it is, for example, the White House got only a portion of the tech leaders and the generative AI piece to come in. I don’t know whether we will get enough. And in contrast, you’ve got countries like India that are saying, we’re not going to regulate gen AI.

Right, and then you’ve got Europe, that’s, you know, the first mover to actually do something about it. So it’s going to, it’s going to find its way there. That’s the, uh, it’s like every day I wake up, I tell people it’s like a soap opera, the next episodes already drop every time you wake up, you know, in your time zone, you wake up, you say, oh my God, let me see what’s happening. The speed at which the governments are responding is, is actually spectacular, I have to say.

PHIL:

Yeah, yeah, you raise some good points there. And it is interesting reconciling the US government, the EU, both moving very fast, not that it appears that they’re not landing in the same place. And it’s quite easy to contrast those two. But when you bring Russia, China, India into the picture, the range of possibilities in the contrast really broadens.

And, you know, there’s definitely, and of course, each of them will have an effect on the other. So I think the, the US’s ideal policies or the EU’s ideal policies will not exist in a vacuum, they will be, their center will be shifted by, by what happens in Russia, China, India other, other large powers.

JOTI:

Absolutely. No one wants to get left behind and nobody wants another country to have the leg up. And also defense, right? I feel if the US doesn’t move faster, we’re sitting ducks. The fact that combining generative AI with the quantum compute capabilities, which 11 countries already have access to, our elections, our infrastructure for the country. You know, you can see all of the, you know, what happened with MGM and Caesar bringing those networks to their knees is quite scary.

And I do think that us as a country, we do need to think about the U.S. as a country first. I think we need to look at it. And I’m not, you know, I’m not going to talk about nationalistic agendas and stuff. I don’t want to get into the politics of it. But there is an impact, to making decisions on this technology combined with technologies like quantum compute, our hands are gonna be forced, but with bickering in DC and with the elections coming up this year, I just feel like  we’re losing ground and we need to move faster. So that’s what worries me the most.

PHIL:

Fair enough too. I think it’s worrying if you’re not worrying about it then you’re not informed.

JOTI:

Correct, you don’t know what you don’t know. That actually, Phil, is the biggest problem. Most people who have not been in this space don’t know what they don’t know. And so there is this sense of fear that they don’t understand. And so you know what happens when people make decisions with something they don’t understand? It’s like, what are you doing?

As we celebrate Women’s History Month, who are some women who have inspired you?

PHIL:

Yeah, yeah, exactly. So Joti, March is Women’s History Month and if everything goes to plan, we’ll be releasing this interview on March 8th for International Women’s Day. Who are some of the women that have been particularly inspiring to you?

JOTI:

I love that question. So, you know, I look as far back in history, Ada Lovelace, right, who effectively was the mother of computers, right? Although she may not be getting as much credit for it, all the way to folks like Timnit Gebru, who was courageous enough to stand up when she saw things not working too well at Google when it came to ethics. And you know, she started her nonprofit called DAIR to say, you know, at least what warms my heart is to understand that there are women out there that are finding a way to do the right things. It’s not, you know, saying it’s a woman-led initiative, for the sake of it.

But women do think, you know, women are wired differently than men, it’s just a human, that’s how nature created, you know, beings and all the genders in between, right? To take the best of those thinkers and leaders and creators and innovators, I draw inspiration from all of them. And it’s not just folks in the technology sector, but, you know, folks who’ve done human rights work, because as you know, there is a huge impact to jobs in this space. So, you know, looking at what women have done, over the course of history is what inspires me to say, okay, this new era that we’ve entered, how do we want the role of AI to be played in our society, in our business, in our government institutions,  education systems, it’s impacting everything.

So, I’m looking, actually, I need that inspiration from those women that thought differently, right? We already know the men that have made the change. And you know, I’m… I am not a feminist. I actually do believe that, you know, for women to win, men don’t have to lose. We just have to make sure we’re looking at things in a human manner. But, you know, I do try to draw up and make sure that I am, like I mentioned, you know, Ada Lovelace and I mentioned, you know, Timnat Gibru, even, you know, political reformers, social reformers, I think they all need to have a seat at the table, in order for us to imagine what this new world is going to look like.

So I read a lot on anthropology and history to really understand that because this is bigger than just technology, right? This is bigger than just the capitalism of deploying these systems into government and enterprises. Who are the women that are speaking? And I can tell you there are, when people tell me, oh, I can’t find… women to hire in, you know, technology fields, I say, I can introduce you to 200 or they’ll say, well, there’s not been enough women. I said, here’s more that I can show you, but you have to be aware, right? To your point around that awareness and education around it. So, we got to look back in order to look forward as well, with women who’ve done this before

How do we address gender bias in AI?

PHIL:

Okay, my next question, my last substantive question. It’s a fairly complex setup, so just bear with me. For this question, I’d like to focus on AI bias. And here are some things that have been written on that subject. From the Harvard Business Review, “AI systems learn to make decisions based on training data, which can include human bias decisions or reflect historical or social inequities.” From the UX collective, “bias in AI is a mirror of our culture.” From AI researcher, Brandon Lewowski, “the reflection of societal biases in AI-generated content is not just a technical issue, but a societal one.”

Now, as I mentioned in the introduction, you are the G100’s USA country chair for robotics and automation. And the G100’s vision is to create an equal, progressive, and inclusive environment for women worldwide. To my mind, this puts you in quite a unique position. You’re right at the intersection of what Brandon Lewowski described as not just a technical issue, but a societal one. And with all of this as context, how do you think we address gender bias in AI? Is it urgent to address gender bias in AI specifically as a high priority? Or do you think that perhaps achievement of G100’s broader societal aims might make it possible to solve the problem of AI gender bias indirectly?

JOTI:

I think it’s going to take work at both ends and here’s why. So, in the trenches of folks that are working with these generative AI systems, who are dealing with this dirty bias data, which is what’s driving, you know, it’s fuel for these models. We are going to need human intervention to make sure how do we take the bias out?

And thankfully, I believe there’s also technical solutions to that, but it will take humans to actually ask the question saying, hey, before you use this data for training, is it biased? I always draw the example to illustrate for folks to say, you know, not only is it the right thing to do to remove the bias, but economically it’s the most profitable thing to do. And here’s why. If you want this technology to serve consumers and consumers who’ve got green in their wallet or digital, you know, digital wallets, you’re going to have to make sure that you appeal to them. And you can only appeal to them if the experience that they’re getting from these AI systems reflects who they are.

If you have bias in the system, for example, I give this example that everybody gets it. If you have a young white man in Silicon Valley training one of these GPTs on working with an African American woman; a 50-year-old African American woman who’s going through menopause; you know that there’s no way that young white male would understand what it’s like to deal with. So, if they don’t actually look at the data to make sure it’s not a turnoff. They’re not going to adopt, you know, that a hundred thousand bots that died on Facebook. That’s what we’re going to see happen here.

So, I always draw from the win-win, right? From a capitalistic standpoint, it’s like, if you don’t do this, your service, it will actually not generate your revenue. Everybody gets that. And so they say, okay, what do we do about it? But I said, well, you have to think through how you are gathering the data, how you’re cleaning it, the folks that are assigned, like if you, you’ve probably seen this, how OpenAI trained ChatGPT initially, right? It was a blend of scraping, you know, whatever they could get access to, on the web, but also a human expert. There’s actually an article I read that came out on Wired Magazine today about the human experts that were paid as low as $30 an hour. They had nuclear physicists and they had linguists that were told, and they didn’t know what they were doing because they were these middleman companies that said, you know, go online, answer these questions.

There was actually a mathematician who was correcting, you know, calculus formulas on there. And so humans were being used to make these things more intelligent. But if the creators of these models don’t have the intention of removing bias, the bias won’t get removed. Which means these beautiful systems that are being released to the world contain all that bias. And what will happen is it will perpetuate racism and hate and all the bad things that humanity has created so far because they reside in it. And by the way, there is technically a challenge for them to guardrail against those things. There is a site called jailbreakchatgpt.io that’s making these models reveal their racist sides. But you know, these models are not human. They’re being trained with that.

So, you’re going to see that fallout and the way, you know, we’re positioning this with the companies and organizations we work with. And I said, look, you have to think about this. Don’t blindly go plug, you know, these closed models into your systems because you really don’t know what they will do in the field. And there’s not enough guardrails in this world and on the planet that technically can stop it from doing it. So, um, from a societal problem, yes, it is going to make sure, for example, loans, right? We know that historically mortgage loans, uh, they’ve been folks from the, you know, the diverse community, the folks that are non-white, not getting enough, African-Americans, for example, in particular. That would perpetuate for sure.

But at some point at the same time, those wallets where folks have the spending power, if they see something come out that is egregious to their sense of ethnicity, they will not use those products and services. It’s a double-edged sword. Does that make sense in sort of the complex question and the setup?

PHIL:

Yeah, absolutely. And I think the core of that is that we do need to address it from both directions.

JOTI:

Correct. Oh, I actually do want to add one more thing because you talked about the G100.

The massive opportunity, green field opportunity that this era of generative AI has brought upon us, and I’ve talked about this for the last four years as I started to see the massive impact, is for women and for minority communities to come into this field, for a couple of reasons. From the highest level of statistics standpoint, the World Economic Forum came out with a report in 2020 that they updated in 2022 saying there are 97 million new jobs that are coming in this era of AI that we’ve stepped into and 85 million existing jobs that will be eliminated. Fast forward, we’re in 2024. We’re already seeing the start of that, right? With the layoffs that have been occurring.

So, what I’ve been telling folks, and I’ve actually been doing a mentorship and training program for the last, since 2020, the summer of 2020, when everything was in lockdown, is could Freshriver train women and minorities in this technology? You know, you can’t take existing developers and UX, UI designers and put them into roles to deal with generative AI, because it’s a very different technology they’re dealing with. And so, I tell women, I tell folks of different minorities, say, if you feel that you’ve been left behind in the previous generations of all the technological advancements that we’ve had, this is your time to start to enter a field and gain a foothold because those 97 million jobs, guess what? Everybody’s got to figure out what role they go to play. And those roles have not even been defined yet.

When I first started in this industry five years ago, as an independent consultant, the job descriptions that would come in would have three lines on them. And when I would talk to the hiring manager, they like, we’re not quite sure what, you know, what we’re looking for in, but you’re who we want. This happened over and over and over again. And so, I tell folks it’s an open season for everybody. And so, imagine a world, if we fast forward five years from now, we’ve got equity in the jobs. It’s a great opportunity, I think.

And so, the G100, that is what we’re aiming for across the world. You know, women leaders from every different country you can think of that’s part of it is putting programs into place to train more women, young girls, all the way down from K through 12 to universities. And if we push harder and we build momentum, I think we will land because there are 97 million jobs that need to be filled.

Where are the 97 million jobs coming from? What are going to be the key roles that this is going to produce?

PHIL:

Okay, that was going to be my very next question. The layoffs are so easy to see. They’ve been coming in huge waves. The new jobs are a little harder to measure. Where are the 97 million jobs coming? What are going to be the key roles that this is going to produce?

JOTI:

So, it’s actually what we’ve just been talking about, right? Who’s going to clean the data of bias? Who’s going to train these systems to make sure they don’t hallucinate? Who’s going to continually test these systems for their lifetime to make sure that they are performing the job that they were meant to perform and they haven’t gone sideways? Product managers who need to dream up how to work with these technologies, right? Where does it make sense to deploy them? Developers who have, we’ve talked about deterministic software, right? Decision trees, which these systems are not. How are we going to have developers that understand how to do that?

You know, the layoffs that have come have been technical roles and all those roles have to be retrained. There is no university or a place of education, I guarantee it, on this planet that can teach you how to do this. AI, generative AI is about doing, right? So they’re literally emerging based on the problems we’re seeing, whether it’s a bias or whether we should deploy this technology.

Just because we can doesn’t mean we should. So, I’ve always said this, the last five years I said… We need to have as many non-technical roles at the table as technical roles. So, we need philosophers, we need social scientists, we need linguists. By the way, those are the folks that are on our teams. When we’re working on these projects, we’re making sure that it’s not just the engineers or the MLOps folks or the data analysts that are making these decisions. We’re making sure the pods that work together to bring these systems to life for our clients.

There is an equal voice at the table of folks that are considering social impact. And our job is to advise our clients to make sure, by the way, we know you want to do this, but here’s the repercussions of it. The decision is yours, but it’s our job to uncover that impact back to emotional intelligence, back to ethical intelligence, right? Should you deny, you know, healthcare benefits for a 80 year old? Because now they’re looking at the full history and the risk, right? Just because they could gather up all the data from the fact that they’ve been ill and they’re denied. Should you? Right?

PHIL:

I think the answer to that should be very clear to anyone but of course it isn’t.

JOTI:

A lot of gray areas when you’re trying to be, you’re being pulled in three different directions. Economics, ethics, and emotional intelligence. Not easy, not easy to do.

PHIL:

Yeah, but I’m glad that you’re approaching this with a perspective of optimism because I think that’s going to be important to achieving these goals. Otherwise, we’ll be defeated before we start.

JOTI:

Exactly.

How will AI and technological advancements impact the world over the next five years?

PHIL:

So, I have just one last question for you. If you were running this interview, what’s the big question that I’ve forgotten to ask and should have asked and what’s the answer to it?

JOTI:

I think the question would be what do we see the world be like in five years, given the trajectory we’re on? Is it dystopian? Is it utopian? Because we’re hearing that from, you know, as much as, you know, Bill Gates and folks that are sounding alarm bells to… You know, Sam Altman saying, this is the most beautiful thing ever, and Elon Musk, right? What’s the kind of world do we imagine we’re going to be in? The answer to that is again, back in history is how human behavior, which has not changed over the last 70,000 years of, you know, when we were still Neanderthals, right? I think we will find a place of balance.

I think it will be violent, I think there’ll be moments of beauty. There’ll be moments of thrill when we solve, for example, health care issues, you know, developing cures for diseases that have played this, you know, Parkinson’s. I’ve got, you know, my mom and my mom-in-law both have Parkinson’s. Like in our lifetimes, even in five years, because I do believe this is how fast it’s going to move, is we’re going to see the best of and the worst of.

And my optimism is that because we cannot kill this technology, we shouldn’t. And not that that will happen because the trains left the station here. But I do believe that worlds will be a little bit better than where we’re at solving those problems. But there will also be a lot of angst that we’re going to see. So, the five-year view is still more optimistically heavy, but we’ve got a lot of work to do before we get there.

PHIL:

Yeah, well, I sincerely hope that the optimistic part of that is absolutely true. And I would love to see a little less violence and turmoil globally. It’s a sobering thought that it might be worse in five years than it is today. But I sincerely hope that that’s that part of it is not the case.

Joti, thank you so much for spending time together today. It’s been great talking to you, great hearing your insights and your perspective on this. As I said earlier, you really do sit in a very unique position at the intersection of technology and society with that capacity to talk about that. And as you’ve said, the intersection of society and technology is probably where the successes for this technology are going to be driven. So really appreciate this. I’m sure people will enjoy hearing your perspectives and thanks again for being here.

JOTI:

Thank you for having me and it’s been my pleasure.

The post LXT podcast – episode 8: Joti Balani – Founder of Freshriver.AI appeared first on High-Quality AI Data to Power Innovation | LXT.

Introducing Pradnya Desh, CEO at Advocat

PHIL:

It would be an understatement of mammoth proportions to say that today’s guest has an interesting work history. After completing her Master of Science in Economics at Georgia Tech, she went into 10-plus years working for the US government. Initially as an economic analyst and then later as a frontline representative of the government in trade attache, negotiator, and diplomat roles. In parallel with her government career, she completed a Doctor of Law degree, which set her up for the next phase of her career as a legal practitioner and educator. Fast forward to January 2021, when she founded Advocat AI as a generative legal nAI company, with the lofty goal of making contracting accessible to all, while striving to make legal work joyful. It gives me enormous pleasure to welcome today’s guest, economist, negotiator, diplomat, visionary entrepreneur in legal AI, CEO, and World Cup finalist, Pradnya Desh.

PRADNYA:

Thank you, I’m glad to be here.

How did your unique background and experiences in international diplomacy and international law lead you into generative legal AI?

PHIL:

It’s great to have you here. So Pradnya, we are, I believe, all products of our experiences. How did your quite unique background and experiences in international diplomacy and international law lead you into generative legal AI and the foundation of Advocat AI?

PRADNYA:

I think it led me, the reason that I have taken the path that I have is because I’m really inspired by solving problems. And when I was a diplomat, I saw the economy on the whole and saw the world as a whole and saw different parts of… actions that I could take and policies that I could be a part of to make a difference in bringing about prosperity. And then after being a diplomat, I switched into a law firm and was pretty surprised by what I saw. I thought that being on the front lines, in working directly with people, I thought that the private sector, unlike the government, really had it all figured out. I thought, so the government is slow, the government is known for being slow, but the private sector on the other hand probably moves at light speed and it’s not that way in the legal industry. I found that the legal industry; it’s very expensive for people and companies to access. It’s very slow and because it’s expensive and slow access is very limited. There’s really a system of scarcity, rather a system of abundance and as a diplomat I wanted to bring about abundance and as a legal practitioner I wanted to do the same. And I saw, really, legal tech  as a way to solve that problem

Do you see major ethical or legal barriers in the way of AI reaching its full potential?

PHIL:

Great. And I think you’ve opened my eyes to what being a diplomat is in a way that nobody has before. And your references to prosperity and abundance. I hadn’t really thought of that as part of the role of a diplomat. But yeah, that’s quite inspiring. In discussions of AI in general, and generative AI in particular, technological limitations, ethical concerns, and legal frameworks to manage those ethical concerns seem to be inextricably linked. Do you see major ethical or legal barriers in the way of AI reaching its full potential?

PRADNYA:

It’s a tough question really because the technology itself is not good or bad. The technology is. And so it’s really our role as practitioners to make sure that it is used for good. And therein really lies the challenge is that it’s important to do so. And there’s not really a framework for doing so. And one thing I really worry about is a bad actor making either an accidental, making a mistake or intentionally doing something harmful and then a reactive regulatory framework coming about. And that will really stifle the potential benefits that we would see in AI. And so we need to act quickly to figure out the correct regulatory framework so that we don’t have this catastrophic event from happening and then the resulting boomerang regulations.

Is it possible that we’re already in a reactionary phase in terms of regulation?

PHIL:

Do you think it’s possible that we’re already in a reactionary phase in terms of regulation?

PRADNYA:

I think we are. We’re seeing that in the EU. We’re seeing that we’re seeing the US regulations trying to get a handle on it. So we’re certainly seeing that. I think it could be worse. I think that the issue is that the industry is moving so fast and the technology is moving so fast that it at the moment may be outpacing regulations. Now over broad regulations can put a damper on that. And so while we are in that mode already, it’s not too late, yet. I think that we can get a handle around it, but it’ll be important for industry and government to have a say and to come together rather than it just being government dictating what should and should not be done. And I say that as you know is coming from government.

Is reconciling different regulations across countries a difficulty?

PHIL:

Yeah, yeah, absolutely. And when I asked that question, I was specifically thinking of EU where it feels to me like maybe we’re in a reactionary phase already that and the EU’s orientation to many things is more restrictive, perhaps than… maybe restrictive is not the best word, protective. They seem to be very driven by protective, being protective of individual rights. Do you see an issue or a difficulty in reconciling different regulatory perspectives globally? Certainly, if you look at it, there is a contrast, I think, between the US and the EU in how they perceive the role of regulation, the importance of regulation, and the extent to which regulation should apply. But if you throw, say, Russia, China, India into the mix as well, then you get very broad differences, very broad range of possibilities there. Is reconciling that a difficulty?

PRADNYA:

So, I think that question is different for an individual company than it is really for governments as a whole. Because what I’ve seen at least recently for tech companies, at least in the United States, is that while they are not certain as to how the EU regulatory framework will apply to them, they’re afraid of offering their most cutting-edge products in the EU. So, I’ve been seeing European consumers complaining about notices they’ve been getting on various software that they use saying not yet available in the EU.

And that’s the reason why is regulations there’s no technical reason for that. It’s simply that as their lawyers are trying to figure out what risks there are in offering their product to the EU, I think that one effect is that legal systems are selecting necessarily what products their people can and cannot get by how restrictive, you didn’t want to say restrictive but I will say it, how restrictive the regulations are.

And so your question really was though about harmonizing regulatory systems. And so if everybody had the same system as the EU, certainly tech companies would have to figure out how to comply. But at least right now, the reactionary, well, rather the reaction from these companies are, let’s not deal with it now, let’s simply block access. And then maybe we won’t have to deal with it while they figure out the right level.

PHIL:

Yeah, I think that’s a great answer to that question. So I guess it means yes. Yes, it is a difficulty and it would appear that rather than attempting to adapt the product to multiple legislations, they’re simply saying, well, you just can’t have the product. At least right now. Which will have an interesting run-on effect. If people feel that they’re being shortchanged, that will put a lot of pressure on governments.

PRADNYA:

Certainly.

As a former government insider, do you think governments will have the necessary agility and capacity to keep up with the rate of change?

PHIL:

In my experience, governments and government-funded agencies, including DARPA and IARPA in the US, have a great track record of supporting and driving innovation.

Great and very long track record of that. But right now we’re seeing massive acceleration in the rate of progress in the development of generative AI. And it seems that the center of gravity in this is very much within the corporate sector. As a former government insider, do you think governments will have the necessary agility and capacity to keep up with the rate of change?

PRADNYA:

So there’s really two parts of that, is because I think that the government and the agencies that you named did a great job in creating grants and creating funding mechanisms to spur innovation. And that seems like an excellent role of government, in fact. But keeping up with the details of the technological innovations, government just, governments cannot move that quickly. Even on issuing these grants, it takes months to maybe, six months to 12 months to get these grants out.

And by then the technology has so significantly advanced that they are funding whatever their funding is not really what was applied for in the first place. But, so I don’t think that the governments are yet able to do that. But a partnership between industry and governments is the only way to keep up.

Is 2024 the year of AI accuracy?

PHIL:

Oh, that’s a great answer. Thank you.  At this year’s World Economic Forum in Davos, executives from some of the top companies discussed how 2024 will be the year of AI accuracy. I can see this from more than one perspective. Is it wishful thinking? What’s your perspective on it?

PRADNYA:

I don’t think it’s wishful thinking because when I mentioned the fear of a bad actor coming in and taking down the system or causing very different regulations, it’s all because of accuracy and I actually call it safety. Is that we’re in the legal industry, is that we have a legal AI product and in having a legal AI product, I consider my number one concern, my number one ethical concern, but my number one concern at all to be safety for those that use our product.

And AI accuracy is an important part of what leads to that safety. And I think that application companies like mine are working very hard to figure out how to take an LLM which while it has quite a bit of material in it, some of it accurate, some of it not, some of it coming from a source that knows what it’s talking about, some of it coming from a source that has no idea what it’s talking about or can hallucinate, as well as draw on information that is not the best source. And so application companies like mine have been working hard to tackle that problem.  Is that to bring a layer of accuracy to the wild and moving changes that are in LLMs.

Which female historical figures have been particularly inspiring to you?

PHIL:

So March is Women’s History Month. And so I’d like to ask a couple of questions related to this. I’ll start with asking, which female historical figures have been particularly inspiring to you?

PRADNYA:

Let’s see, so I have two very different ones. Is that the first I would say Amelia Earhart because female innovators inspire me enormously and in fact innovators somebody who is attempting to do something that has not done before that should be done inspire me and so that’s my first and then the other is Mother Teresa and the reason why is her spirit of service also inspires me and I feel like innovation with the spirit of service is really what drives me.

What do you think is needed to ensure that the representation of women in technology continues?

PHIL:

Wow, that’s great. What do you think is needed to ensure that the representation of women in technology continues?  Do you think this requires proactive affirmative action or do you believe that this will ultimately take care of itself?

PRADNYA:

Well, at least among the US venture capital scene, it’s not taking care of itself yet. And at the moment, we’re not seeing on the horizon that problem being improved. And I say that because only 2.2% of venture capital, goes to company, in the United States goes to companies that are founded by women. And that is a shocking number to me because it’s not that we have a dearth of female innovators, it’s that those female innovators are getting funded at a much lower rate. And if they’re getting funded at a much lower rate, we’re necessarily choosing, the market rather, is necessarily choosing, what gets built and who leads that charge. And I think that in AI in particular, that’s a problem because having representation among all types of diversity is important in AI because an AI can only look like the humans that are working on it and developing it and adding expertise to it.

PHIL:

What do you see as a solution for that? How are we going to move that agenda forward quickly?

PRADNYA:

Yeah, well… for starters, invest in more female founders that can do so, because I think that investing in female founders means that those founders will hire more women. And so while I think it’s important to bring in, we’re talking gender I know, but bring in all types of voices, because as we invest in diverse voices and diverse founders, we’re able to bring more diverse products to market. And so, I would start with simply, say, talk with your investment dollars that’s my answer.

What’s next for Pradnya Desh?

PHIL:

Okay, that’s great. And you’ve once again shifted my thinking. I was perhaps, I think I probably asked that question with quite a lot of optimism and you’ve shaken my optimism a little bit, but I do hope we can actually collectively do the work to, to achieve that diversity, as you said, not just gender diversity, but overall diversity in order to ensure that we actually have a much stronger product development field.

Economic analyst, negotiator, diplomat, legal practitioner, legal educator, AI entrepeneur. What’s next for Pradnya Desh?

PRADNYA:

Well, I really love what I’m doing right now. So I’m the CEO of Advocat AI. I’m not planning to go anywhere, at least for quite a while, because we’re in such an important moment for the world. I love being at the center of… I would call it a revolution. The AI is bringing about a revolution that’s similar to what we saw in the industrial revolution.

I mean, I consider it that important. The Industrial Revolution changed the world by making manufacturing by hand and manufacturing one item at a time and switched that to large-scale manufacturing in which a human and a machine were working together to bring widespread prosperity to the world.

And we did see that happening. And AI is causing no less of a shift, except that this is for knowledge, work, and expertise. We’re seeing, we’ll again see human and machine work together to bring widespread prosperity.  So I love leading an AI company and I’m excited to see what’s next in AI for the world.

AI optimism and perspective changes

PHIL:

Wow. Well, I wish you well with your continued work with Advocat. That’s a very exciting prospect. And I share your view that this is somewhat parallel, analogous to the Industrial Revolution. So yeah, it’s an exciting time to be in the business. It certainly is. Okay, what’s the burning question that I should have asked you but I didn’t? And what’s the answer to it?

PRADNYA:

Interesting. So, I don’t have an answer to that, but I do want to know this. I’d like to turn that around and ask you a question.

PHIL:

Sure thing.

PRADNYA:

So you do this podcast and I think you talk to some very interesting guests and come upon some very interesting insights. And so I’d like to know an interesting insight that you’ve gained on AI in the last several months in the conversations that you’ve had. Where has your perspective shifted?

PHIL:

Um, I, talking to, um, some remarkable people over an extended period. Um, I think, unlike my question regarding, uh, women in technology where you’ve shaken my optimism, I would say that my optimism has been, enhanced, with regard to AI through the conversations that I’ve had. I think that the people I’ve been talking to are consumed by the power for good. No one is unaware of the risks. No one is unaware of the potential for bad, but I think everyone is very much focused on the power to… change things for the better, change society for the better, and to extract the positive potential out of this technology. I think that’s the biggest single takeaway that I’ve had from the last few months.

PRADNYA:

I’m glad to hear that. That’s my experience as well.

PHIL:

I don’t know, maybe if I took an hour or two to think about it, I might come up with something else.  But yeah, multiple people that I’ve spoken to have taken this angle.

And I have spoken to some very interesting people from very different ends of the AI spectrum.  And it’s great to have your legally oriented perspective on this today. I’ve thoroughly enjoyed this. So Pradnya Desh, thank you so much for participating in this. Thank you so much for your insights. I hope that we’ll have the opportunity to perhaps do this again at some point in the future and compare notes on where we’ve been and where we’re headed next.

PRADNYA:

I hope so as well. Thank you.

The post LXT podcast – episode 7: Pradnya Desh – CEO at Advocat appeared first on High-Quality AI Data to Power Innovation | LXT.

Introducing Tech Entrepreneur Iain McCown, Director of AI at Dubber

PHIL:

Hello and welcome. My guest today is an industry veteran, a researcher, and a serial entrepreneur with more than 25 years experience in AI. He was principal research scientist both at CSIRO, Australia’s leading research and innovation organization, and at the Idiap Institute in Switzerland. He is the author of more than 50 research papers and holds multiple international patents. In the commercial domain, He founded Dev Audio, where he developed a speech-enabled microphone array, which was subsequently acquired by BiAmp Systems. And then he founded Notiv, which was subsequently acquired by Dubber, where he is currently director of AI. Please make welcome Iain McCown. Hi, Iain.

IAIN:

Hi Phil, thanks for the intro, looking forward to talking to you.

What are some of the emerging applications of generative AI that you are most excited about?

PHIL:

That’s great. Okay, so we’re more than a year into what I would call the post-ChatGPT era and AI, especially generative AI, remain hot topics. What are some of the emerging applications of generative AI that you are most excited about?

IAIN:

I mean, it’s a good question. There’s so much going on at the moment. It’s hard to keep up with all the changes and the different things coming every week, it seems. I think for me, I’ve been working in speech and language tech through my career, right? So that’s the area that interests me the most and is engaging me the most with what’s changed. And I think if you look at, you know, broadly, everyone’s aware, ChatGPT came out and that was kind of a step change in awareness and also a step change in capability of language models, to be fair. For the first time, human readable and output was available for different applications from AI models.

And so I think for me, what’s exciting at the moment is, you know, for the first time we’ve got these new tools that are able to actually understand conversation.  And I think what, you know, a lot of the use cases are focused on human bot communication, if you like. But I think what interests me is the potential to unlock human to human conversations and, and extract insights and get more value out of conversations like this from this AI. So for the first time in, in my career or in history, you know, AI is capable of understanding, synthesizing, summarizing, detecting key points in a human-to-human conversation. I think there’s a lot of exciting opportunities around that.

PHIL:

Very cool.  Yeah, I would certainly agree with all that. It is pretty exciting. I’m not from a technology background. I’m a linguist. But the idea that language technology is at the state of evolution that it’s reached now is very exciting to me.

IAIN:

Yeah, I know for like one, just one example, I guess the holy grail for us has been summarizing a conversation, right? And for years we struggled with extractive techniques where you try and get bits of transcript and piece them together and it just wasn’t readable. And whereas now, you know, you can get a condensed readable summary of a conversation. So that’s just one of the new capabilities there, which means that it’s unlocked new use cases for tech to actually give value to businesses, use cases for tech to actually give value to businesses.

Based on your entrepreneurial experience, what advice would you have for entrepreneurs who are considering this path?

PHIL:

Yeah, I can’t wait to skip meetings and read the summaries. I hope my colleagues aren’t listening to this. So, Ian, you’ve very successfully navigated from startup to acquisition twice.

Based on this experience, what advice would you have for entrepreneurs who are considering this path? And do you have specific advice on the challenges or opportunities inherent in addressing global markets from an Australian base?

IAIN:

Yeah, I mean obviously having gone through two startups, different journeys each time, you know, I could talk about that for hours or days, but I think what it comes down to is, you know, you’ve got to be creating something of value. Building value is what a startup is about. You’re starting from an idea or your abilities and trying to create value. in terms of delivering something to a customer, a new customer need that you think you can meet. Also building value for your investors and in your team, et cetera. So I guess if you bring that mindset to what you do is that every step of the way you’re trying to create value for either your customers or your team or investors, I think you’ll be successful at what you do. I mean, a startup can be an up and down journey.

So…  the other advice, I guess, is always do something you’re passionate about and you love, and that’ll get you through the ups and downs.

I think in terms of doing it from Australia particularly, or I guess from outside the tech center of the universe in the US, my approach is to abuse a phrase, I guess, think global, act local. So, you’ve got to be globally minded from day one, wherever you’re starting a business from. You can’t launch a tech company and be just thinking about people around you. You’ve got to be thinking globally from day one. But that said, there’s smart people everywhere. There’s resources everywhere. Connect to people and networks and programs, investors, government support around where you are. And from Australia. Australia has been a great place to launch businesses from and that continues to, you know, that ecosystem and support for entrepreneurs continues to improve.

Do you have a preference for working in research or commercial environments and what are the specific challenges and rewards of each?

PHIL:

Very cool. So you’ve worked extensively in both research and commercial environments. Do you have a preference and what are the specific challenges and rewards of each?

IAIN:

I think my career as you touched on at the start has been, you know, kind of see it’s spanning that gap I spent the first part of my career in the research environment. I love that, but was frustrated that, you know, the endpoint was writing a paper or, you know, publishing. And then that enabled you to maybe get a grant for the next one and the next one. So you’re always taking ideas from concept to the point where you’ve validated it, written about it, but not actually take it into the reality of getting in the hands of someone who can use it. And so I think, you know, there’s different types of researchers, but in the research world, I was frustrated wanting to make, you know, not let it go when it got to a paper, but take it all the way through to reality.

But the research environment’s not often set up to do that. On the flip side, in large enterprise, you’re really set up to take products to market, but maybe less scope to be creative and come up with, you know, far-fetched ideas and test them out and things. So I like being between both. I guess that’s why I’ve done two startups is really taking some new emerging capability and spanning, if you like, that chasm between research and reality and taking it across.

As a product leader in the world of AI, what are your views on the role of regulations and its impact, positive or negative, on innovation?

PHIL:

Recently at the World Economic Forum, there was discussion of 2024 as potentially being the year of AI implementation. And there was a focus, a strong focus on ethics and safety. As a product leader in the world of AI, what are your views on the role of regulations and its impact, positive or negative, on innovation?

IAIN:

I think my approach to this is like as researchers in this area, we know what we should do. We know what to do, we know what the correct thing is to do, what’s right in terms of how we deal with data, particularly training models and practice around that. So I think regulation shouldn’t change behavior, but it does create accountability there to make sure that, hey, in the worst cases, there’s accountability. So I’m supportive of the regulation. Obviously, in realm of responsible AI there’s some new regulations coming out in UK, US, Australia at the moment that we’re continuing to watch, but from my seat, it’s not changing what we’re doing, it’s just providing that, I guess, that accountability and that structure to how we report on that and our processes around that, which is a good thing.

I mean, I think to touch on your question around innovation is always pushing boundaries, right? And maybe, you know, that’s important. And so you don’t want regulation that squashes innovation. So it’s got to be well informed regulation, not just knee jerk overreaction to things. So I think where it’s well informed, well- intentioned regulation, it’s good. You know, that. I’d say innovation should push boundaries but it shouldn’t be crossing, you know, there are lines you don’t want to cross and I think that’s where regulation comes.

Is there a solution to generative AI and it’s tendency to have, and indeed, act upon biases and to hallucinate? And are these flaws likely to prevent AI based on LLMs from becoming truly scalable?

PHIL:

I recently read on a related note, I recently read an article which said that the Australian government was very concerned about ensuring that AI was trained on unbiased data and it really got me thinking about what that actually means. And my next question, which is a fairly lengthy set up and question, is a reaction to what I read about that, because I feel like this concept of unbiased is quite risky. I mean, in a sense, all the data that’s out there is unbiased.

It’s a direct reflection of the real world. And in that sense, it’s unbiased, but it will produce technology with biases.

And so I’m slightly concerned that the Australian government might be using what feels like the right terminology, but it’s not scientifically informed. Anyway, to my question. Among the major flaws in generative AI are its tendency to have and indeed to act upon biases and to hallucinate. Biases and hallucinations in GenAI output are typically a flow on from deep seated biases in the data used to train the models. When discussing this, I’ve often encountered the view that what’s needed is neutral or unbiased data. My own take is that there is no such thing as neutral, unbiased real-world data.

So it seems to be that the main options might be to use huge volumes of data that are out there in unfiltered form and accept the inherent biases. And in so doing to accept that those who are underrepresented in the data will in fact be underrepresented by the products or to proactively introduce bias to address imbalances in a form of affirmative action, if you like. But when you do that, this neutrality becomes subjective. And then there’s the question of whose interpretation of neutral will prevail. So is there a solution to this? And is this likely to prevent AI based on LLMs from becoming truly scalable?

IAIN:

Yeah, there’s a lot in there, a lot to pick through, I guess. So in terms of is there a solution to this, so if the problem is, is there a solution to bias or hallucination from models? Yeah, I think there’s multiple ways you can attack that problem. Data, you know, the training data  distribution being one of those ways that you can influence that. But you know, you’ve worked in data for a long time, so I’m not going to argue with your thoughts on the reality of actually collecting unbiased data, whatever that means. So I think, I guess the perspective I come at it from is a product perspective.

So we’ve got the problem of there may be bias in our system or it may hallucinate. How can we, knowing that our model may have those limitations or other limitations. How can we deliver a product that provides value to customers in a safe way? And so there’s different ways to do that. One of the ways we do it at Dubber, as you know, we’ve worked with LXT on this one, is, you know, not so much on the training data, but on evaluation data. Let’s collect evaluation data that represents particular population or demographics, such as languages or dialects or, you know, types of businesses, et cetera. And so let’s collect a smaller amount of data, which is probably more feasible and achievable and measure, quantify at least performance on different data. So at least then we understand, is there a bias and does my model work better for US Americans than it does for French Canadians, for example, if I’m doing transcription. And if so, what’s that difference and how can I address that? So I think step one is quantify and understand the bias in your product and that can be done with more reasonable amounts of data.  And then, you know, the other way is old fashioned engineering, I guess, probably aligned with your affirmative action type thought line of thought in that, you know, I’m not a smart car, self-driving car engineer, but I imagine that inside there, there’s some AI, right?

That sophisticated AI that’s detecting objects, classifying them, making sense of the world that the car is traveling in. There’s probabilistic models at the heart of that, right? But they are engineered into a system that makes that a safe, as a whole system, makes that a safe driving experience for the driver and other people on the road. So I think that that same thinking applies to, you know, for what we do at Dubber, for example, we use generative LLMs, but we package them in such a way that they’re anchored to the context of the conversation so that less likely to hallucinate and show bias, for example. We constrain rather than letting it just generate whatever text it feels like and sending that direct to a user, we structure, get structured output. So only certain, I guess, categories, we’re looking for structured output from the LLM so that we can control what the user experiences in a safe way.

So I guess there. Like it’s a deep topic. And to your point, I’m also be wary if people thought that the solution was just unbiased training data, because as you said, you know, most of these AI models just need more and more data and the data they can get reflects the world and the world has biases and you know, that’s the way it is.

I think in terms of your final question there. Can we achieve, can these LLMs achieve intelligence with these limitations or is there a solution? You probably consider yourself an intelligent being, I do as well and yet I’ve got biases. I tend to hallucinate after a couple of glasses of wine, I hallucinate and rabid on answer things that haven’t been asked to me. So I don’t think intelligence may still have biases and… hallucinate as part of it. Who knows?

In the face of the combined wealth and resources of these mega-corporations, Apple, Microsoft, Amazon, Tesla, Meta, do you think that there’s still an opportunity for smaller organizations to thrive in the AI space?

PHIL:

That’s a very cool answer, yeah. And I like, to paraphrase, engineers can save the world. My next question. There’s an enormous concentration of AI firepower in the top seven corporations globally, the so-called magnificent seven. So between them, Apple, Microsoft, Alphabet, Amazon, Nvidia, Tesla and Meta account for 27% of the S&P 500’s value. This week,

The Financial Times reported that Apple, Microsoft, Meta and Alphabet had between them made more than 50 AI-related acquisitions in the past five years. So there’s a trend. In the face of the combined wealth and resources of these mega corporations, do you think that there’s still an opportunity for smaller organizations to thrive in the AI space? Or is it possible that startups are simply seen by these larger organizations as incubators for ideas that can be acquired and integrated later?

IAIN:

Um, such an easy one for me as a startup guy, I have to say there’s a space for startups and there always will be. One thing I learned early on in my, in my entrepreneurial journey is that,  you know, every, every startup company is different, right? It depends on, you know, not every startup is out to become the next unicorn. So there’s a range of businesses and sizes and different objectives for why people start businesses. So I think there’s a place, a place for all sorts and all sizes, I guess specifically in tech startup land. I know there’s a number of smaller businesses that we work with because they provide the best solution for some of the things we need.

I know in the world of speech techs, for example, yes, the big tech companies like Microsoft, Google, Amazon have great platforms for that, but state-of-the-art performance is also available through smaller companies like Rev AI and Deepgram, for example. We’ve just spent two years building out our conversation intelligence platform at Dubber and at the heart of that we use an orchestration engine for our workflows called Temporal, which again is, I’d say probably a startup. Startup’s kind of a fuzzy term at what point you stop being a startup, I guess. So I think there’s always, there’s room for innovation outside the Big Tech, particularly as you get into specific niches.

So I guess come back to my early point around startups is what’s the value you’re creating and there’s, you know, there’s advantages to being a startup and a small company and that you can really focus, hone in on a niche or a particular piece of value that you want to deliver and really do that well and quickly and in a more agile manner than what a larger organization could be. But, you know, there’s no denying the environment is, scale, you know, those big companies are growing and there’s a momentum that, that causes and keeps building, right? But I mean, remember when I started my career, you’re probably the same. The big research, all the research papers came out of commercial labs, AT&T, Bell Labs, and, you know, IBM. You know, they were commercial labs. And then through the middle of my career, that’s why the pendulum swung back to academia, right? And all this deep learning and the current wave we’re in really came out of academic research labs initially and then has grown back into the commercial. So swings and roundabouts, everything’s in cycles, who knows where it goes.

What trends are coming up, or what shifts are happening in how data is going to fuel the next phases of AI?

PHIL:

Very cool. Okay, well, I have just one more question for you. If you were running this interview, what would be the one question that I haven’t asked you that you wish I had? And what’s the answer to it?

IAIN:

Um, so I think you’ve asked me a lot of questions and I’ve probably rattled on enough for this talk. So if I had the chance, I’d like to ask you a question if that’s all right. I guess, um, I mean, um, I’m on the, I guess the AI development product development side of things. I know you’ve spent 20 some years in the 30 years, who knows in the data side, right? And so AI, the current AI is, is largely built on data resources, right? That’s the fuel that’s driving this as well as processing power and other things.

So I guess from, I’m interested from your seat and where you’re at, we’re at an interesting point in the industry in that large reams of data are available. Models are now, you know, I remember back in the day we had the, the holy grail was unsupervised learning so that we didn’t have to hand label every piece of data. You know, we didn’t have to have the annotation expense. We could just give the model data and it would learn stuff. And so now we’ve got this, what do we call it, self-supervised learning is a big part of how these models are leveraging large data pools.

So I guess in this new world or current world, from your side of the fence, sitting on the data provider side of the fence, I guess what trends are you seeing coming up or are you seeing any shifts in how data is going to fuel next phases of AI or where the next wave is coming from.

PHIL:

Okay, good questions. And I will say you’ve mentioned two or three things in there and earlier in the interview, which I had to stop and think, have I actually published my previous interview already? Because they were red hot topics that we were discussing there with Roberto Pierroccini, who was formerly with AT&T, Bell Labs, IBM. And we discussed some of these very things and he referred to unsupervised learning. And we discussed some of these very things and he referred to unsupervised learning as the Holy Grail, which, yeah, I know you’re not blowing smoke when you say that. It’s definitely a widely held view. So yeah, where do I see the data field going? My view is that the need for data is not going away. My confidence level in my view is some fair distance short of 100%. So I say that with moderate confidence, but not high confidence that data is going to be required for quite a while to come.

In fact, one of my former customers in the automotive vehicle space, when I asked him how long he, how much more data he thought they were going to need, he said, he’s about the same age as me, and he said, well, Phil, I can’t quantify it exactly, but let me just say that when you and I are dead, we will still need more data. Yeah, and at that point we were generating data at a phenomenal rate. So, you know, his desired volume of data was unimaginably large. My experience with that is has throughout my 25 years working in this business, every time it looks like the data problem, well, every time it looks like the problems are solved and people start talking about, well, maybe nobody’s going to need data anymore.

What happens next is that we move on to try to solve bigger, more sophisticated problems and rather than the need for data going away, it 10xs. I’ve seen that 10x happen again and again and again. And so I still feel like if we are going to, if we’re going to build these super robust applications, the need for quality data and for human in the loop is still around for a little while. Am I answering the right question?

IAIN:

Yeah, I think so. Just your 10x comment made me think of back in 2005-ish, we collected the AMI Meeting Corpus when I was at IDIAP and we thought this was this huge, valuable, multimodal database. Then we recorded, it was 100 hours of transcribed speech in meetings.

And we thought that was, at the time, that was a significant resource. And now that’s a drop in the ocean. I guess my questions are, you know, we’ve got issues around, I guess, OpenAI and Wall Street Journal copyright of content that’s being used to train models. And then we’ve also entered a new world where a lot of the content on the web is, and probably newspapers and other things, is going to be partially generated by AI itself, right? So it’s kind of corrupt. It’s no longer pure human-sourced data. So I think my own take is that maybe we’ll swing back to quality over quantity. and just being more picky about the data. But I don’t know.

PHIL:

Interesting, yeah, actually that’s when you talk about pendulum swings. I’ve observed multiple times in my time the pendulum swing between just get me volume and don’t be too fussy. Get me volume cheaply. Don’t be too fussy about the quality, to okay we need the quality. We’ll pay for the quality. But that’s more important than volume. So we need to be very, very careful about the quality. And that one definitely goes backwards and forwards in quite tight cycles. Thank you for that. You’re the first person that’s actually turned the final question back on me and I’m very pleased you did. Thank you. Iain McCowan, it’s a pleasure talking to you as always, and I look forward to our next conversation sometime down the track.

IAIN:

Thanks again for taking part today. Thanks, Phil. I’ve enjoyed chatting to you.

The post LXT podcast – episode 6: Iain McCowan – Director of AI at Dubber appeared first on High-Quality AI Data to Power Innovation | LXT.

Introducing speech recognition expert and industry-veteran, Roberto Pieraccini

PHIL:

Hello and welcome. My guest today is an electrical engineer, a speech scientist, and an industry veteran, a true industry veteran. When I read his list of employers, it’s like laying out a roadmap for the contemporary history of speech recognition and AI.

Bell Labs, AT&T, SpeechWorks, ScanSoft, IBM, SpeechCycle, Jibo, Google. And in addition to his current job as chief scientist at Uniphore, he’s a talented and successful author, photographer and musician. It gives me enormous pleasure to welcome as our guest today, Roberto Pieraccini. Hello and welcome Roberto.

ROBERTO:

Thanks so much, Phil. I know we’ve been talking for a while about this podcast, and I’m glad we made it happen, and thanks for inviting me, it’s a pleasure.

Was Siri what scientists of the field hoped for to recover from all the frustration and unpopularity that speech recognition’s small and big earlier failures had raised in the popular culture?

PHIL:

It’s great to have you here. So, Roberto, you’ve published two wonderfully informative and accessible books on the evolution of AI, 2012’s Voice in the Machine and 2021’s AI Assistance.

I don’t imagine that at the time of their publication, you could have foreseen that they would roughly coincide with two of the biggest landmarks in the technology’s progress. That is, in 2011, Siri’s full integration into the iPhone, which really did change things for speech recognition, I believe. And then in 2022, the public release of ChatGPT.

So, looking at Siri and its integration into the iPhone, when it was already apparent that Siri would be impactful and relevant, you asked some great questions, but ultimately left them unanswered. So, I’d love to hear how you would respond to your own questions today. From your perspective, was Siri a beginning or an end? And was Siri what the scientists of the field hoped for to recover from all the frustration and unpopularity that speech recognition’s small and big earlier failures had raised in the popular culture?

ROBERTO:

Thanks Phil, that’s a great question. As a matter of fact, when Siri came out, I think it was October 4, 2011, exactly one day before, unfortunately, Steve Jobs passed away, I was in a hotel in Melbourne, in your country. So, yes, it was a time we were actually working, it was a speech cycle, and we were working with Telstra, the Australian large telco, telecommunication company. And then I learned, I knew Siri because Siri came out as an app, independent app on iPhone one year earlier, one or two years earlier than that.

And I was trying, I was mesmerized by Siri. And then I heard that it was coming out soon on iPhone 4S, if I’m correct. And then I… the proofs of my book were already at the publisher, MIT Press. I already signed them, the final manuscript. I said, I cannot publish a book without a chapter on Siri. So remember in that hotel room, very quickly I wrote the last chapter, which is called Siri, What’s the Meaning of Life? And the publisher liked it and we inserted that.

So, the question that Siri was beginning or end was both, right? Siri was the end of decades of frustration in speech recognition for many reasons. Technology was not quite there. I started working in speech recognition in 1980, but speech recognition started earlier than that, as you know. And we were moving ahead and there’s all this, people making fun, speech recognition isn’t working, and the problem was that… The world knew about speech recognition through the, what they call IVR, interactive voice response application that somehow simulate contact center agents.  And there were a lot of issues at the time. So first of all, people called agents when something was wrong.

So, they were not in the best of mood, right? You know, check bounces, the appliance you bought doesn’t work, your TV doesn’t work. And then instead of talking to a human, they talk to a machine. The machine did not always recognize them and so on. So Siri, I’d say was the end of that time, although the time is continuing today, right? It is still years away.  We’ll talk probably a little bit about that later.  But Siri opened up the technology of speech recognition to the whole world in a totally different manner. And it was for, you know, people were not captive on application, they didn’t want to talk, people could choose freely to use Siri for many things. And people started using it in more unusual way, asking where can I bury a body, I remember for a time was one of the most popular questions. Ask Siri, but also all that things like set an alarm and all that type of things. So that was a good thing, a good thing, right? Now, the problem was that it was very hard to know the limits of the voice assistant. Can I ask anything? Can I ask anything in the way I want, in any possible way, with any possible accent, in any possible noise situation? And the answer was no, and it was very hard to understand the limits, right? What I can ask and what I cannot ask.

And the second point, what are the capabilities? Can I ask to plan my next vacation? Can I ask to tell me what was that nice restaurant in Paris where I went three years ago, no, maybe it was four years ago with my wife and I had onion soup, right? Actually, we are laughing about that, except the onion soup. If you use Google Maps, Google Maps knows pretty much the restaurant you went to in Paris three or four years ago. So in principle, you could answer this question, but not at that time. So it was the beginning of a new era, the of the automated assistant or voice assistants, or as I called then in my second book, AI assistants. And then in fact, immediately after a few years, Alexa came out, Google Assistant came out, and Samsung Bixby came out. So that’s what I consider an epochal thing was a change of scenario for speech recognition.

Did Siri turn out to be the gentler version of HAL from 2001: A Space Odyssey that we’ve been waiting for, for more than 60 years?

PHIL

Yeah, absolutely. It is quite remarkable today with multiple choices. I’m sitting in my lounge room. I can actually call out any of those names and they’re there and ready to help. So the big question, in your books you often made reference to HAL, the personal assistant in 2001: A Space Odyssey. In fact, in the end, the villain in 2001: A Space Odyssey. So the question you asked, in your opinion, did Siri turn out to be the gentler version of HAL that we’ve been waiting for for more than 60 years?

ROBERTO:

Yeah, that’s a very good question. You know, when I started, when I was in high school, I went to see Kubrick’s 2001: A Space Odyssey. That was one of the things that really impressed me along with the moon landing. I want to be an engineer. This is great, this is so cool, right? You know, and then by chance ended up in speech recognition. I was on the streaming. Can we build a gentle version of HAL, not a villain version, but a gentle version. And I would say that comparing Siri with HAL is a little bit of a stretch, right?

You know, if you watch 2001: A Space Odyssey again, you’ll see that HAL has incredible capabilities, even today, we cannot do even with the latest ChatGPT and large image models… There is a scene where, Dave, the astronaut, is drawing something on a sketchbook. It was very boring life on the Discovery because everyone else was hibernated. There’s no one to talk to or very few people to talk to. And he was sketching the faces of his colleagues who were hibernated. And then he shows it to HAL, HAL asks, what are you doing there? Dave says, I’m sketching. HAL says, ah, nice rendering. He improved a lot in the past couple of months. Can you put it closer? And Dave puts it closer to the camera and say, ah, that’s Dr. Hunter, isn’t it? That’s amazing, right?

You cannot, and also the level of emotion, like when they play chess, of course, HAL cheats at a certain point. HAL is cheating, right? It’s calling a move, but it’s making another move, right? And so if you look at… carefully at the chessboard, but then say thank you for an enjoyable game, which is amazing, right? Of course you can record that, could be canned, right? But the level of expression, the level of things, it looks like, well, we are far from there, right?

Whether we’re gonna get to something like HAL, I don’t know, but today is a little bit of a stretch. I don’t think we are quite there. And whether we will or we want to be, something like that. You want to build something like that.

How concerned are you about the villain potential of generative AI?

PHIL:

Yeah, I mean, there’s been a lot written in the last, since launch of ChatGPT, which really awakened public consciousness of AI potential, generative AI potential in particular, that, how concerned are you about the villain potential, if you like?

ROBERTO:

Yeah, of course, being deep into that technology, I’m concerned, but not concerned for the, you know, the doom day when AI will imprison or will put an end to the human race, you know, or human species. I don’t think so. I’m not afraid of that. What I’m afraid about the use of AI for, by villains, by humans who are villains, right? For doing things that can cause a disaster.

Imagine interrupting the electrical power for days and imagine getting to the banks and using your money. These are real things that could happen.

So, like if you remember when we had virus, I don’t know, we didn’t hear about viruses anymore, since COVID right? But the computer viruses, but it was a continuous fight between the virus detectors and all the virus that runs on your PC and the virus. And every time there was a new virus, the virus detector came out with new solution. And I think that’s our future, right? We see, we need to create measures and guidelines and guardrails. It could be legal guidelines, and technology guardrails that prevent anything bad happen. That’s why everyone, including myself, is talking about responsible AI. We build AI, we need to be responsible about what we build. And that’s all the companies in the world are really on the same line. Of course, like in every case, the good people are on that, thinking that way. What we need to fear about is the people who are not as good as that. They want to do evil with that.

How would you characterize the contemporary value of AI systems, perhaps from a user perspective and then from a commercial perspective?

PHIL:

Yeah, no, no, absolutely. Thanks for that. I think people will find that answer quite reassuring, which is a good thing. So let me bounce back to the beginning of the 1990s. At that point, I think the way I saw it anyway is that the value proposition for speech technology was evident, but not particularly compelling, up until the point where Jay Wilpon and AT&T released the first meaningful voice automation system at scale. In your book, you referred to this and you said that there was a significant financial investment needed to support the emerging technology so it fulfills potential.

So, it was hardly surprising if at that time you could measure success in terms of the number of humans that can be replaced, for example. OK, so you showed that a simple five phrase application enabled six thousand layoffs. So that makes a financial, a compelling financial case, which perhaps didn’t exist before that event. And then later in the 90s, I love the phrase that there was a shift from carbon based transcription to silicon based.

ROBERTO:

Did it say that?

PHIL:

Yeah, I think so. So, yeah, so it doesn’t really surprise me that at that time, when people were trying to make a compelling case, trying to get the financial support that they needed, that they would make their case very strongly in financial terms. These are the savings we can make.

So, give us the money and we’ll make those savings for you. How would you characterize the contemporary value of AI systems, perhaps from a user perspective and then from a commercial perspective?

ROBERTO:

Thanks. This is a very, very good question. First of all, you mentioned Jay, Jay Wilpon, who is a great friend of mine. We are born the same year. We all love seeing each other. And if he listens to this podcast, I say hello to Jay. As a matter of fact, the AT&T guaranteed me, or Jay guaranteed me, that the 6,000 layoff never happened, but they found a way. So I was, I don’t know if that’s true or not, right? But I was reassured that Jay was working on this technology probably one or two years before I joined AT&T Bell Labs.

But it’s always the case, if you look at the history, the most evident case is the invention of the assembly line by Ford in 1913, I think. That displaced a lot of people, right? And any new technology does that. And we need to be, the whole society and the whole system needs to be aware of that, and needs to provide help for people at this place. People need to be trained on different jobs and different careers.

And so that’s a responsibility that we have as a society. But we cannot stop the… the progress of technology, right? Otherwise today we still use horses, right? Not Teslas to go to work, right? So, talking about this particular technology, the reason why it came out in the mid-90s or was because if you remember before there was the what’s called DTMF, press 1 for this, press 2, and then voice recognition in the mid-90s thanks to speech works and nuance, at least in the US, became very reasonably accurate. And we started using that. And the reason is not just to lay off people to save money, but because if you look at the consequence of the spending necessary for an enterprise to maintain a trained task force for customer care is huge. And what happened, the agents were poorly trained. So, they didn’t provide much value to the end customer. And since you don’t have an infinite number of agents responding to the demand of an increasing the amount of customers, customers will end up waiting in a queue with music for tens of minutes or half hours or more. And it’s not the best use of that time.

So, the idea to automate part of that, and to always provide a way to escalate to a human agent when the problem was hard to solve for the machine. In a sense, it was not just for saving money, but also for providing a better customer care. So today, I am back to that after Google, after Jibo, and I work at a company called Uniphore where that is one of the many value propositions that we provide. So, what you call self-serve, right? So building applications that somehow automate certain functions, not all of that, certain function of agent. Actually, there are more interesting applications than that where AI is helping. One is the agent assist. I told you that it’s hard to train agents and there’s a big turnover and attrition among the agents.

So, talking to an agent who is not trained doesn’t help. So, AI can provide support to agents, tell them what is the next thing to do, give them answers from the knowledge base that the user is asking. And also making summaries. Today we can do summaries quite well using the latest technology. Summaries are something every agent has to wrap up a call at the end of the call, it takes time. That means money for the company. So if summaries can be generated automatically for the agent to review and correct when it’s the case. Other applications analyze the tens of thousands, hundreds of thousands of calls that the company gets every day. Companies that are big enterprise don’t even know what the agents talk about, what the customer asking about and then new trends, new problems arising. So, if you can do that automatically, it’s a big, big improvement of the customer care situation for a company. And also, there are other applications that we are working on like support for sales, emotional detection in a sales goal to understand what is the sentiment of the clients and provides hints to the salesperson or when the clients are not engaged and so on.

So, there are a lot of applications, which is not just the horrible voice VR-like people. If you remember at the time there was a website, I want human dot com that tells, just to tell the tricks like for example, you call the AT&T, you get an automated machine, push zero three times, turn back. five times, say the magic words, and then operator will come, right? Yeah, so, actually it was a serious thing. It was a serious thing.

In terms of advances in machine learning and AI, does it still boil down to solving communication problems with elegant mathematical solutions?

PHIL:

Yeah, that’s a, yeah. I can imagine, yeah. Yeah, so yeah, I think it’s clear that that’s a really good explanation of the value proposition commercially, but also that from a user perspective,

these days, it’s not something that people aren’t going to go and look at a site like that anymore because they’re quite satisfied with what they get from the interactions and, and perhaps would choose it over human in some situations because they don’t have to engage, they don’t have to do small talk, they can just move right along.

Okay, in 2012, you framed AI as language, thought, understanding, solving communication problems with elegant mathematical solutions. And I’m interested in that – the elegant mathematical solutions part of this. Is that still how you view it today? In terms of advances in machine learning and AI, does it actually still boil down to solving communication problems with elegant mathematical solutions? Or is it now something different?

ROBERTO:

Yeah, that’s a very, very, very good question that gets into epistemology and philosophy, and I could talk for, I would love to talk for hours with you in front of a good Australian red wine.

PHIL:

Yeah, we can arrange that.

ROBERTO:

We can do that next time. But that’s, that’s very interesting. I’ve been very fortunate to live through the evolution of AI or machine learning. I like to use it until today. When I say that, I was referring in particular to speech recognition or ASR as people call it, because ASR was not invented because people started to try to find solutions to ASR. To this, as far as we know, in the 1950s, using analog computers, not everything analog machines with resistors and capacitors and tubes.

But in the 1970s, the work done at IBM Research by Fred Jelinek and Jim Baker, they came out with an elegant mathematical formulation of the problem of speech recognition. And it’s the only equation that I have in my book, which is the equation that everyone who works in speech recognition should at least know that puts out the problem, solving the problem of speech recognition and solving the problem of the optimal receiver in presence of noise. We have thoughts, we express our thoughts with words, the words get into a noisy, and the noise is also the different variations that we use to talk, different dialects and so on. And then it gets to the ear of a listener, and then you can express it mathematically. And the noise, you want to get the best possible sequence of words given the noise. And the equation highlights two important things that have been with us until I would say 10 years ago, and this, the acoustic model and the language model. So we knew language models way before language, large language models came to. New language models since then, the end of the 1940s with Claude Shannon talking about that in his famous seminal paper on information theory.

So, so what happened? What happened today? So if you look about what happened today, I would like to cite one of my philosopher heroes, is Daniel Dennett, who wrote in a recent book that makes the parallel between the evolution theory, Darwinism, and Alan Turing, the Turing machine. So, the evolution theory taught us that you can build complex, sophisticated organisms without comprehension. We don’t even, the evolution theory does

not understand how a virus, a frog or a human being work, but that happens to an algorithm that is the survival of the fittest. So many variations of that, right? So it could talk about competence without comprehension. Alan Turing showed a very simple machine can solve all the problems. Of course, it’s a virtual, theoretical machine.

And today we have large models. Language, large language models, don’t have the modus inside. They take the speech, talking about natural language understanding, take the string of words, find the adjectives, find the nouns, find the verbs, makes an hypothesis about, there’s nothing of that. It’s just a mass of artificial neurons that have been trained into so much text that they have been training only to predict the next word. And it’s amazing how just predicting the next word or the next token, they have a behavior that seems like intelligent. I say seems because I don’t believe it’s totally intelligent in many people, but it demonstrates some rationality and some intelligence and makes a lot of mistakes sometimes. Right, and again, it’s competence without comprehension. The individual artificial neurons inside the machine don’t understand anything, don’t have comprehension. But the massive thing, the neurons in our brain eventually get to a competence level.

Is there an absence of design in generative AI?

PHIL:

Does that suggest that there’s an absence of design at this point as well?

ROBERTO:

So, I like to talk about ChatGPT and large language model, large language model is not just ChatGPT, but when I was at Google, I was working on Lambda and MENA and the evolution actually wanted to be going to Bard and Gemini. I like to say about the history of AI, like the end of the intelligent designer. So, there’s no intelligent design. It’s not totally true. In order to design a ChatGPT there is a lot of engineering behind there, right? And the same for all that I was talking about. But we don’t design modules. We don’t design algorithms. We don’t design what we used to do in MIR 12, 13 years ago, right?

PHIL:

Right, so yeah, so there is a very large phase, perhaps the largest phase in which there is an absence of design.

ROBERTO:

Yes, yes, yeah. And see, we need to, we are the beginning of that. What I call actually the democratization of AI, right? Many people call it that because everyone can use ChatGPT to create interesting application. Even if you don’t have started natural language understanding, speech recognition, and so on.

Is unsupervised learning the holy grail for speech recognition?

PHIL:

That’s great. Which is a nice lead into the next question I have here. So, you noted that unsupervised learning is the holy grail for speech recognition. I think it’s safe to say that this is, today, perhaps the holy grail across the entirety of AI and machine learning. To what extent do you think it’s possible and practical to achieve this holy grail? Is there a limit that you think it might be impossible to cross?

ROBERTO:

I think we have to cross the limit. If you want to build machines that are more and more competent, we need to be able to train them with more and more data and more and more multi models. They’re not just text. The limit of today’s large language models is that they have seen only text. Imagine a brain in the bat that’s only read the web, never touched a cold surface, it never tasted the taste of a lemon and so on, right? So there is a limitation to that, right? So, we need, because we cannot use what we used to do before annotations, right?

Annotate curated data. We need to curate the data in a way to avoid redundancy and multiplication, clean up data from a lot of things. But that can be done programmatically. When I started speech recognition in the 1980s, we had to record words and make sure that they were the words that were tagged and so on. We cannot do that anymore. Also, there is a more strict guidelines of privacy. I cannot use recordings freely, right? I cannot, right? And in fact, we need to think about how… And you know, large language models are trained mostly in a supervised manner. They learn how to predict the next token. So it’s easy to take text and mask, remove the next, once at a time, remove the next token and ask the large language model to predict it.

So, that’s one thing. The other thing is that we see more and more generation of synthetic data that is helping, that happened a few years ago with speech recognition. So we can generate data, we know how to generate speech. We can generate pattern between noise and variations and so on. So this is the new world and this is what we need to, these are the problems. So now we are seeing a lot of use of the human in the loop in what is called reinforcement learning with human feedback. That’s very useful, but it happens on a limited amount of this. We want to make sure. And that’s useful for many things like working on hallucinations, bias, and safety. So, I still see that, but it’s not the big majority of the big amount of data is going to be unsupervised.

Scalability issues, increase in volumes of data, human-like performance of generative AI, and human in the loop: Are there near term solutions for scalability concerns around these topics?

PHIL:

My next question does dig a little bit deeper into that. So over recent decades, the volumes of data used have increased massively. And you’ve expressed concerns about how this impacts scalability. So, I’ll just read a fairly lengthy quote here and then there’s a few questions.

“What plays a big role in the more difficult attainment of human-like performance in language, understanding and generation, is that even today we still need to rely on representations of meanings such as intents and arguments which are not naturally available to us and these need to be crafted on a case-by-case basis and crafting an abstract representation requires a lot of work that hardly scales to cover all possible meanings and their variations.”  Now, I’m sure at the time that you wrote that that made a lot of sense. What’s happened in this time since then has probably obviated that even further.

So, do you see a near term solution to this scaling problem? Do you think that those case-by-case crafting that you talked about, that that can be automated? But if it is automated, does the issue of getting human-like performance extend to some of the bigger issues, such as hallucination and elimination or management of bias? And can this be achieved without human in the loop?

ROBERTO:

That’s a great set of questions. Let me start saying that in my opinion, and of course, all of this is my opinion, anytime we tried to impose a human invented representation on a machine, on a language machine, speech machine, I’m not sure I’m going to understand. we didn’t get great results. And a clear example is picture recognition. Until, I would say, 12, 15 years ago, we imposed the phonetic transcription. And we, we couldn’t get a more certain, why? This means to a certain extent, and I know I will attract the irate responses of linguists. It’s a human invention. If you’re going to take someone in the streets, and say, can you tell me what are the phonyms? People don’t know. We know the words, and we know how to pronounce them, but nothing in between,.

So in fact, today, speech recognition does not need to have a phonetic transcription. It creates its own phonetic theory, which is very amazing, right? One of the layers, or the many layers, so that we’ll have something which not exactly correspond. By the way, linguists fight, right? About what is the correct presentation of phonyms and so on, right?

Now, the same thing with intents. Intents are a human invention. The human invention of  intents, especially for a visual assistant, it’s very important, necessary. Why? Because eventually, if you ask a human, a virtual assistant to do something, it has to call the API, what we call the API, the function that defines a functionality. If I say, what’s the weather tomorrow in Sydney? Eventually, once I understand this, I had to create a HTTP request to the weather.com site and get the response and interpret the response, right? We all created intermediate because it could be weather.com or could be another site. Intermediate representation that is weather, open parenthesis Sydney, comma, tomorrow, close parenthesis.

But now that creates a scaling problem. The API exists already because people will be at this website, they will create an API. We need an API for everyone to be able to query that thing. And then we want to create… and that requires a lot of people, a lot of engineers, defining these intents and arguments or entities, how we used to call them. Now, if you look at things that, you know, Bard, Gemini, and ChatGPT, they don’t have intents. Why? Because they’re able to provide the answer to a question exact without going to intermediate phase.

Now the problem, if I ask what’s the weather in Sydney, they don’t know that, because they were trained three months ago, six months ago, and they may knew what the weather in Sydney was, but it’s not of use to us, right? So, there are somehow they have to know the API, what we call the API or the function call and how to invoke it. And then I see we are moving in the direction where with the GPTs and agents, and so on, we can do that by providing to the chatbot or to the large-language model the knowledge of the API without designing and engineering an intermediate representation. So, I think there is a hope and I think probably the world is moving so fast right now that many of the things we do today don’t require the design of an intent and argument – intent and argument schema.

Can the human in the loop be taken out?

PHIL:

Yeah, that’s, that’s, so. And therefore the human in the loop can be taken out?

ROBERTO:

No, the human loop, as I said in one of my previous answers, is still to validate. You mentioned hallucinations. You mentioned safety. Safety is an important problem, what we call the alignment problem, aligning the values of AI to the values of humans, like safety, bias, fairness and so on, right? We don’t want a chat bot, like ChatGPT to be abusive, to have abusive language. And we don’t want them to give advice on how to do violent things, to build bombs or to kill people, right? Or simply advise on health, right? You ask, I have a headache, what should I do? They should not. They should tell you, go to a doctor. And not allowed to give you health, I think, right? But they do that.

So, all these things which go under the umbrella of responsible AI and safety and trust, they still require some form of human in the loop. And I said that we could have like quality assurance loops where some humans interact and say, oh, this… not to be safe. It’s an abusive answer. I mark it as abusive. I believe, and I probably see some article where we start doing that automatically. Imagine we have a much more expert language model, expert exact on safety issues that could actually correct, teach the other model. This is not very good to do, but this is still a big research issue and not a problem.

Can we predict the future of AI?

PHIL:

Okay, well, I have just one last question. And that question is, if you were running the interview and not me, what is the question that you would ask yourself? Is there something really important that I’ve forgotten to ask?

ROBERTO

Ah, that’s interesting. You asked exactly the questions that I would ask myself because probably you are, you read books, thank you for reading my books. And you got to know me and to know my interests and my points, the points that I was trying to make there. I don’t have a specific question to ask myself. I know probably, the question is what will happen in the next five years? And the answer is somewhat that some of the some great scientists gave. Making prediction is very hard, especially ahead of time. So, so, I mean, who could predict the ChatGPT 20 years ago, 20, no one did, right?

PHIL:

My wife and I, my wife and I often sit down and say, you know, talk about could we have predicted where we are today if we looked just two years ago? And usually the answer is across a lot of big things there’s something that… that we would never have imagined, places

our lives have gone.

ROBERTO:

It’s interesting, we had a New York City party with some friends here and I ran a game that I’ve done in another party, say, let’s predict what will happen at the end of this 2024 and let’s meet again at the end of 2024. I wrote down the predictions actually, actually. And the prediction was about political status, about the AI and all this stuff, the economy and stuff like that. We’ll see how good. But who could predict COVID and how Mel Brooks would say, who could expect the Spanish Inquisition? No one did expect that.

PHIL:

Well, Roberto, can’t thank you enough for your time and your patient and insightful responses to the questions. It was really great to spend this time together and to dig a little bit deeper

into things that I think are really the important topics in our industry today. So huge thank you.

ROBERTO:

Thank you so much, Phil. I know that we have been dating for some time before we actually took, I probably started a couple of years ago and I changed jobs and all these other things,

but it’s great. I really enjoyed. Thanks a lot for the great questions. Next podcast, we do it in front of a good glass of Pinot Noir from from some of the great wineries in Australia or in Italy.

PHIL:

They’re both great suggestions and perhaps we can talk about art and photography in the next session.

ROBERTO:

That would be great. Thank you, Phil.

PHIL:

Thanks again Roberto, take care.

The post LXT podcast – episode 5: Roberto Pieraccini – Chief Scientist at Uniphore appeared first on High-Quality AI Data to Power Innovation | LXT.

The answers weren’t all that surprising given how many businesses use these tools, but it got me wondering how, and to what extent, generative AI has impacted these solutions, now that it has burst onto the scene.

Generative AI is most often associated with using AI to create content, so it might be easy to think that it couldn’t directly improve upon tools that are designed to defend against cyber attacks or to facilitate communication. But what I found proved to the contrary: the same fundamental advances that make tools like ChatGPT and Bard to be the game changers that they are can also be applied to more specialized applications of AI.  

Improving the conversational performance of chatbots and virtual assistants

Early versions of chatbots were rules-based and locked into the conversational flow for which they were trained. Recent improvements in conversational AI and natural language processing have made huge strides in chatbots’ ability to sound more human, but asking a chatbot any question or statement outside its predefined rules would still lead to a dead end. Anyone who’s spent much time working in this space has heard stories about how customer experiences went sideways thanks to a broken chatbot conversation. 

Now generative AI is enabling chatbots to be more flexible and respond to customer queries in a more nuanced way. Instead of being at a loss for words when a query strays outside of training, chatbots that are powered by generative AI have the potential to curate technical content from an organizational database, or from the archives of reconciled service calls.

An early example of this is Duolingo, an organization that prides itself on creating fun and effective learning experiences. Duolingo augmented its latest release, Duolingo Max, with GPT-4, which powers two new chatbot features: Explain My Answer and Roleplay.

Explain My Answer offers students a lesson on proper word choice, and students can engage a chatbot within the content of the lesson to ask whatever follow-up questions come to mind. This can be very helpful whether students repeatedly make the same mistake, or simply need further elaboration to clarify the nuances between different words. 

Roleplay provides language learners with real-world scenarios in which they carry-on conversations with chatbots and gain experience points. Thanks to features like Roleplay, Duolingo Max provides an interactive and responsive experience based upon lessons and scenarios that are created by experts in curriculum design. 

Envisioning more secure mobile platforms

There’s no denying the allure of generative AI. Getting an informed, near-instantaneous answer that’s based on all of the information available online about a particular topic can be incredibly empowering. But any scenario that involves submitting your personal or business information out into the cloud is inadvisable.

For their part, security professionals have wisely cautioned against submitting ANYTHING about themselves or their business to generative AI tools, and to adopt a zero-trust security stance instead. 

Qualcomm is taking a more visionary approach by inviting readers to imagine a world where an on-device generative AI security application could act as an intermediary between your device and the online services you use throughout the day. Generative AI security apps like these would protect your personal information, while enabling online services to securely access said information when consulting a financial adviser or making dinner reservations online, for example. 

It’s anyone’s guess how much work needs to be done before such a vision could become reality, but Qualcomm’s idea for the future is certainly more compelling than the status quo. Meanwhile, cybersecurity research firms such as Tenable Research, are experimenting with how they can use large language models to streamline their work when reverse engineering, debugging code, improving web app security, and increasing visibility into cloud-based tools.

Perhaps the two companies can work together and meet somewhere in the middle?

Creating better user experiences through created/cloned speech

Smart assistants first came on the scene in the mid 90’s, but, as is often the case, the early technology was a bit rudimentary. Since then, large language models, and advances in natural language processing and deep learning have led to significant breakthroughs in the capacity of computers to understand and recreate text. We’ve all seen the astonishing ability of generative AI to summarize existing content, as well as to write original works. 

The same technologies have made it much easier to create and clone human speech, a process that used to require a great deal of training, with results that weren’t terribly convincing.  

In just the last few months, Meta has released generative AI systems with text-to-speech capabilities that can edit an audio clip using text-to-speech capabilities, or generate a voice in a wide variety of styles. Meta AI’s Voicebox models can also take a portion of text in one language and translate it into speech in another language.  Google’s Audio LM offers many of the same capabilities. 

Generative adversarial networks enable you to clone a voice and iterate upon it until you have a version that’s virtually indistinguishable from the individual’s actual voice. And products such as  ElevenLabs or Podcastle offer tools such as text-to-speech software that can create life-like voices, or that will clone your own voice by reading as little as 70 sentences. 

Cloning a voice brings with it many benefits and threats, as spelled out recently by global consulting firm, Ankura. For example, with the ability to recreate their voice, people who can no longer talk due to medical reasons could benefit dramatically from an improved quality of life. Companies can reduce operating costs and streamline call center operations by deploying virtual assistants. And Hollywood can bring back characters that were voiced by actors who have since passed away. 

With advances like these, we may have finally reached the point where voice interaction seems possible AND preferable. Not only is voice interaction faster than typing, it’s also safer, can make online searches more effective, and could give way to new use cases for technology.

On the flip side, there are many threats, virtually all of which fall under the heading of security attacks targeting individuals and businesses. In anticipation of this, Google is working on another language model that will be able to detect any audio snippets generated by its own AudioLM speech model.  

This is a lot to take in, and breakthroughs in generative AI seem to be coming at a faster pace than anything we’ve seen before, at least in my living memory. It might be awhile before advances like these reach your bottom line. Either way, stay buckled to the edge of your seat.

The post Generative AI in the real world: pervasive today; ubiquitous tomorrow? appeared first on High-Quality AI Data to Power Innovation | LXT.

At the same time, many organizations are demonstrating increased levels of AI maturity through the success of their AI projects in production. And as they become more AI mature, demonstrable ROI will follow.

But just six months ago, we saw the release of ChatGPT, a transformational generative AI tool that has taken the larger business and technology industry by storm, and spurred a frenzy of other applications and the integration of the technology into the major platforms, from Bard to Bing and beyond. This will have huge implications for the way organizations approach their AI strategy and investment.

Let’s take a step back and look at how the industry was evolving before the advent of generative AI, where companies were seeing the greatest ROI, and how that may change in the future.

AI maturity in the enterprise

In our latest survey of executives at US-based organizations with annual revenue over $100 million and more than 500 employees, we found that 48% rate themselves at the higher levels of AI maturity (up from 40% last year), which includes companies that have AI in production, as well as those at the systemic or transformational stage, where AI is already a part of their business DNA.

We also found that despite the economic slowdown, investment in AI is strong with almost half of all organizations spending $76 million or more on AI. An interesting note is that when you look under the hood at how budgets are being distributed, it’s evenly split across a range of categories, from training data to talent, and hardware to software, among others. This suggests a roulette wheel of sorts, where companies are placing their bets evenly to see where the biggest payoffs will be.

The Path to AI Maturity survey was conducted in late 2022, just before the launch of ChatGPT and the resulting interest in generative AI. At that time, it was still difficult to identify a truly dominant trend driving AI investment and ROI. There were many promising applications of the technology across business functions and industries, but no singular driver.

No longer. At least for the time being, that driver is generative AI, and the curiosity and urgency that organizations are beginning to feel is increasingly clear. Executives across industries are asking themselves how they can use tools like ChatGPT to gain competitive advantage, and stay ahead of challengers who are also wrestling with how best to leverage this new technology.

Natural language processing and conversational AI set the stage

Our research shows that even before the shockwave-like launch of ChatGPT, natural language processing (NLP) and conversational AI (CAI) were two of the top three most widely deployed AI applications, alongside a wide range of applications – everything from predictive analytics to security and robotics, which highlights the wide applicability of AI across the enterprise.

Conversational AI and natural language processing were also at the top of the list of AI applications to deliver ROI in the enterprise, which may be due in part to being more widely deployed. In particular, these solutions are seeing strong penetration in the automotive, professional services, retail and tech industries.

The rise of generative AI has simply turbo-charged these applications and their potential ROI.

What was behind that interest? Enterprises are using CAI in order to build chatbots, digital assistants and other automated interfaces that allow them to connect with customers, build relationships and reduce cost. And they are leveraging NLP and speech/voice recognition to unlock the potential of their language data for things like contract review, knowledge management and sentiment analysis.

The critical importance of a data strategy, and human involvement in AI

How will this change with the rise of generative AI? The short answer is that it will simply accelerate what has already been taking place. In particular, ChatGPT and similar tools provide an easy to use interface that helps to accelerate how enterprises unlock their language data.

But human involvement is still critical for oversight.

As I have been saying for a long time, if you don’t have an AI data strategy, you don’t have an AI strategy, because data is what fuels AI. That is becoming more true today.

Even before generative AI, as organizations moved to higher levels of AI maturity, they were increasingly leaning into supervised machine learning with human annotated data, and moving away from unsupervised learning. I expect this trend to continue.

As we saw in our research, organizations that have reached AI maturity are more likely to have implemented a data strategy focused on data enhancement or annotation, and less likely to rely on unannotated data. In other words, human involvement is increasingly necessary.

Many have also recently voiced the need for a more ethical approach to AI development, including more than 1,000 tech executives and researchers in an open letter calling for a six-month pause on the development of advanced AI systems so that safety protocols can be established. Naturally, these will need to be enforced through some form of human oversight.

For those organizations looking to explore and leverage the potential of generative AI, it is imperative to look back at where ROI has already been generated, strategically choose an appropriate application for the technology, and implement a human-infused data strategy that will guide your solution to success.

The post AI maturity, ROI and the immediate impact of ChatGPT appeared first on High-Quality AI Data to Power Innovation | LXT.

Phil:

Hello and welcome. My guest today is Prem Natarajan, Chief Scientist and Head of Enterprise Data and AI at Capital One. Prior to joining Capital One, Prem spent 12 years at Raytheon BBN Technology and was VP of Alexa AI at Amazon. In parallel with this, he spent ten years at the University of Southern California. Now, what most people might know is that Prem is an inspiration and that he is an industry legend. But what those of you who haven’t met Prem yet might not know is that he also has a deeply incisive sense of humor. With these things in mind, I’m excited – nervously excited – to welcome Prem to today’s edition of Speaking of AI. Prem, hello and welcome!

Prem:

Thanks Phil! Excited to be here. You know, our association goes back a long way, and I couldn’t be more delighted to be talking with someone about AI.

Phil:

Fantastic. I’m going to lead off with a question about where things might be headed. What do you think might become the battleground capabilities for generative AI in the next year or two?

Prem:

My answer might feel a little general, but actually I can’t imagine an area of human endeavor that is not going to become an area where somebody or the other is trying to apply and employ and harness the potential of generative AI. So in fact, in my mind from coding, to code development, to writing documents, to communicating with others, to finding information, to organizing our lives, etc. I think generative AI is going to have – applied properly, thoughtfully, responsibly – it’s going to have a beneficial impact on pretty much every aspect of our social and professional life. I don’t quite see it as much as a battleground as an opportunity to deliver so much magic for so many people around the world that there is enough magic to be created for everybody to work on it.

Phil:

A very diplomatic response, I think. Well, a couple of observations I’ve made. One is that we did a survey of senior executives working in AI-related areas and we asked about areas of application. And when we got the results, it was one area where we couldn’t see any trend at all. And my first reaction when I looked at the data was, “Okay, there is no trend in this area”. My reaction when I thought about it a little bit more was, “Oh, no, no, no”. What this is telling me is that it’s everywhere. It’s not that there’s the absence of a trend. The trend is that it’s being applied everywhere. But my second comment on that, though, is with regard to the battleground status of this, and I suspect that Microsoft and Google might see that differently. It feels a bit like an arms race between those two organizations.

Prem:

I don’t know about specific organizations and what their perspectives might be on it. I like to think about it in terms of individuals and humans and where are we likely to be feeling most opportunity driven or like which leads to competition in some sense. There is an ecosystem developing around AI and around generative AI in particular, right? And so when we look at that ecosystem and we look at some of the fundamental building blocks of that ecosystem, obviously the large language models or foundation models are one place where there is going to be a lot of competition and there are some open hypotheses that are still being explored. There are some people who are betting on one model increasingly becoming better at everything, right? There are others that have the belief that custom tuned models for specific domains are the way to go. There are others who are somewhere in between on that spectrum of one thing that is everything and versus lots of siloed things, like some balance of customization, the fine tuning, supervised fine tuning balance with large language models trained on a broad variety of data, etc. So I think if we think about it as one of the battlegrounds for human endeavor, I think one of the hypotheses that’s going to be tested is what is the right balance of massive models that are capable of doing more and more and everything well, versus specialization. So that’s certainly one area that’s an intellectual battleground, if you will. 

Then the next level of battleground is what are some of the capabilities that are needed to unlock the application to two different use cases? If you’re in health care, for example, you may want the models to be more explainable or be more able to trace their reasoning when they come to a conclusion, right? And so that’s going to be another area of competition. Who has the models that give you the maximum performance consistent with what we see today, but are also able to explain their working, if you will. Yet another area of work would be in terms of reasoning. All of these are having interactions, whether responding to requests or, you know, the human is playing a big role in directing the conversation in some way. But when you get into more complex tasks, how are they at reasoning, etc.? So that’s yet another battleground. 

Then I’ll elevate it and say there’s also a bunch of supporting technologies that need to be built around these, because these things, as we know, can make up their mind that certain things exist that don’t exist. And can do their best to try to convince you that they exist.

So things like vector databases, other knowledge bases in combination with these things. There’s a whole battleground for that. Then there is a lot of competition around which model-building platform should be used. Is it one that one of the big providers of cloud services do, or is it some of the smaller people are saying, these are also specialized endeavors. So there’s just a number of these areas. I will just say, like you said in your survey, you know, where you could either see battlegrounds or fertile fields of opportunity. And right now when I stand and if some generative AI thing and I look all around and I’m seeing, you know, 360 degrees of fertile fields all around.

Phil:

And when you mention this continuum between the idea that one model will dominate everything and at the other end that specialized models will be required for each vertical. Where do you stand on that continuum? What advice are you giving to folks?

Prem:

My belief right now, because I’m building stuff or I’m in a place where I have to build stuff

that can be used, and because we put a premium on the reliability and the robustness and in a lot of other things, given the kind of enterprise we are, things have to really work properly. We have to appropriately govern things, etc. My view is kind of more conditioned by the state of affairs right now. And I think we’re somewhere in the middle in the sense that we need the pre-trained models, but we can’t actually employ them for use cases without customizing them on our data. And that customization, Phil as you know, takes different forms. One is supervised fine tuning where you might collect data that is representative of your use cases and do that. And that’s one of the battlegrounds is what LXT is, who is going to be the, you know, the most proficient developer of instruction tuning data set and who can actually partner closely with people who know AI to be a data partner for that. I mean, this is your history. You’ve led the way there on the data side in the previous generation of AI development. You know, our joined work on trans tech, some of the fun that we’ve had. But coming back to this, I think that balance of finding, but the finding, it takes multiple things. One is supervised fine tuning. The other is how do you construct the prompt? And once you start about thinking, constructing prompts, now you’re saying, how do I involve vector databases in anchoring on curated facts, on data that I know is on hard ground that I have complete confidence in, etc. To me it feels like in the near-term future, maybe near to mid-term, we’re very definitely living somewhere in the middle of that spectrum.

Phil:

I’m going to shift gears slightly here. So you’re at Capital One. Capital One is very evidently a tech forward organization, but it is before being a tech forward organization, it’s a banking organization, it’s a financial institution. So it’s a financial institution first, very tech forward second. I’m not sure if I’m asking you a question that’s sort of out of limits here, but if you were advising your competitors on what they what anybody would need to do, maybe let’s put aside your competitors, but organizations that are less tech forward than Capital One, what do they need to be thinking about in order to not be left behind?

Prem:

So I don’t know necessarily that I mean, anybody needs to be left behind. But my view is each of us has to chart a strategy that reflects an approach to serving our customers. I like to think about it as all of us exist to serve our end customers in some way. And we’ve each adopted a particular approach. And at Capital One, we certainly adopted an approach that is tech based. To leverage technology to deliver the best value that we can to our end users and do so responsibly and thoughtfully. And so the path for us is around the way I told you, I think we’re somewhere in the middle. All of this in the end is simply a means to providing the value and the service to the customer that you want to provide. So in the abstract, it’s kind of hard to say like, you know, because somebody might say they want to provide customer service in a particular way and that that’s how they are differentiated, etc. But I think simply the fact that there is such a variety of services and service providers available and ways to leverage this technology, that every enterprise, whether it’s in finance or in law or in healthcare or in education, right? Everybody will find a way to leverage this technology to become better. 

In some sense, you know, Phil, this was my view when we were in the darker world as well. The thing that we’re competing with is the current version of ourselves. How can we leverage technology to become better every year, every week, every month? And so I look at what’s available today and say literally everybody will be able to be on this continual improvement path, which means society at large and their end users and their customers will benefit. So I kind of think it’s in that more exciting phase now than like just a relative advantage, though, I certainly think like we, you know, we’re kind of leading the way in thinking about how to leverage all of this technology, harness it responsibly, think about the implications, but still use it properly. You know, apply the governance that we know how to and use it to deliver the best value we can to our customers.

Phil:

Right. I read one of your recent interviews and in the course of the interview, you said there is a deep imperative to operate all of this in a responsible, thoughtful way, which I think, you know, I think most people would agree with you. But when I think about a point like that, I wonder what is responsible, what is thoughtful. And in thinking about this question, are there some baseline truths or are those truths rationalized by competing perspectives and interests? For example, if you’re thinking about operating responsibly, is there a prioritization of being responsible to customers, to employees, to shareholders, to society as a whole? And is there a conflict in thinking about that sense of responsibility and thoughtfulness?

Prem:

I think one of the great, one of the great aspects of modern technology and its advances

has been that we’re able to make advances on all of these, because these are fundamentally things that make us all more capable at doing something, more capable of finding the stuff that we want, more capable of delivering value to people that we want to deliver value to. So in that sense, I’ve long seen technology as one of the democratizing influences. If, for example, back when we were working on speech-to-speech translation, the thing that excited me about that was the fact that even if you and I don’t share a common language, you know, as humans, we want to share a conversation, right? And so the fact that speech-to-speech, you know, real-time spoken language translation could allow us to have that conversation, even when you don’t share the language, was super exciting to me. In that same way, I think all of this technology, all of this AI, so when I say responsibly and thoughtfully, I’m kind of connecting to a few different motivations that at least are within me. One is, I want to make sure that this technology is making things better for everyone. It’s not in some way a trade off in the sense things are better for someone vs. worse for someone. In fact, that’s one of the things I find most inspiring about advances in technology, not just AI, is that it allows us to lift things for everyone. So that’s the first part. 

Second is, as these things become more and more effective, more capable, I think we want to bring in more perspectives right from the design stage. I strongly believe in the notion of  inclusion at the design stage. How do we bring in many different perspectives. Ultimately, this thing is meant to serve humans. So the more diverse perspectives I can bring in terms of its use and how something might be used, etc., etc., help us, help inform us to build a better and more robust product for everyone, right? 

A third thing is there are lots of things we care about when we’re in such a business as we are, you know, customer trust and safety and all of that. So thinking about those things as we’re designing it, not just as an optimization after the fact, but right from the design stage, how do we bring all of those perspectives and to me, that’s the essence of a responsible and thoughtful approach. Be inclusive, make sure diverse perspectives are seen, make sure you’re constantly testing all of your hypotheses, that you’re generating hypotheses to test in your testing, and then that you’re also testing them in the real world before you actually launch them into a product.

Phil:

It’s very reassuring to have someone like you in this role. I’m really pleased to see that. So I mentioned DARPA connection a couple of times during the course of the conversation here.

Prem:

And, well, that’s how we’ve met you know…

Phil:

It was an exciting time. It was. Yeah, I would come to those DARPA group meetings where for those who in the background here who don’t know, we worked on DARPA projects where multiple technology leaders would be competing with one another to make advances. And the competitive environment actually produced very rapid advancement in the technology. And it was exciting. It felt like we were doing something that was going to change the world. Or, well, let me back that off slightly. It felt like we were doing something that might change the world

if we could get it right. But it also, to me, felt far from certain that we were ever going to get there. 

Now, I’ve got a quote here at the end of five years of hard work on the trans tech program, the director of the program, Mary Mader, described the accuracy that we achieved

as “enough to be interesting but not enough to be useful”. And at that time, I have to say I had doubts as to whether this would ever be possible. It was a struggle to make the technology work in even the cleanest and most controlled of environments. And the reality is that we were designing technology to work in the dirtiest and most difficult of circumstances with all the issues that come with that. How strong was your confidence that we would ultimately achieve

the breakthroughs that we’ve seen, particularly since, let’s say, 2012, 2013? There’s been massive breakthrough after massive breakthrough. And now the recognition, translation and synthesis pieces that went into trans tech are all at level of advancement that I found difficult to imagine back then.

Prem:

Indeed. So I’ll say this, I was super hopeful. I mean, partly when you’re you know, you’re in the business that I’m in, which is building technology that is going to be useful a few years from now, right? Deep inside you, deep inside you lives a technology optimist, right?

Phil:

Yeah, absolutely.

Prem:

So in some sense, it’s also a mindset in my mind that you think this problem is so worth solving. I think the way I’ve approached a lot of these things Phil, is think about the feasibility later. But first, ask yourself, is this problem worth solving? Is this something that is a hard problem, but one that will make so much value to the world or deliver so much value to the world when it is solved, right? Once you pass that filter, then the next thing you have to ask is, are the elements of this problem interesting to me and to me, anything that involved human language and machines learning them, etc., etc., that I’m and still does remain very interesting. Just fundamentally, they appeal to me, you know. It’s one of those things you don’t know why, but they appeal to you, right? And so I’d say my mindset at the time was, boy, this is such an important capability or set of technologies to develop. And boy, this was so hard. And to me all that meant was, oh my God, I’m going to have fun for so many years solving this problem, which is that, you know, so it was it was less and less the question of, oh, this is overwhelmingly hard, and more like, oh my God, this is going to be so awesome to work on because… and deep inside that, the basis is if you thought you would never succeed, then it would be hard to keep that optimism channeled, right? So I think deep inside I always believed this is coming. Could we have predicted the specific arc of things? Probably not, right? We didn’t predict the specific arc of things. But when you look back at where we even during the speech-to-speech, translation times, you look back, you can see that there was several inflection points in the technology even during the course of the program. Even during that five years, which she said it went from being interesting. I don’t remember the exact quote, but not entirely useful or something like that. I would say even the users of it were finding it useful in very specific instances, right? It was just not useful in every situation. That’s a big change. Like at the start you had no technology and people were simply stringing together different components to deliver a capability. By the end, we had an integrated thing that was talking to each other, etc., etc. and being able to deliver a thing which some people found useful in some cases. So my way of thinking about it was like, oh my God, look at this. At the start of this thing, we thought we could never get that in five years. Halfway through it, we said, oh, maybe we can get that in five years. And then in five years we look back and we say, oh my God, I can’t imagine what a distance we have traveled in five years. And if that’s how I feel about the first five years, do I want to continue on that journey for the next five years? Heck yeah.

Phil:

Yeah. Awesome, right? Absolutely. 

Prem:

Yeah.

Phil:

Very cool. I have to say, that was some of the most enjoyable time in my life. I felt like I was on such a voyage of discovery and I loved attending the events associated with it. It was something else. I just have a couple of wrap up questions here. One is, what do you see as the biggest risk to successful AI deployment?

Prem:

I think there are a few performance curves that we have to bend, right? I mean, we have to make this more affordable across the spectrum. We have to bend the cost of being able to deliver good, useful AI across a broad spectrum of use cases. I think that’s the first in my mind. The second, I think, is I’ll go back to my responsible point, I really think across the board we need to make sure that we’re focused on understanding the behavior of the things we’re building and the potential outcomes and bring in a diversity of perspectives. If that means that we move a little bit slower than we could otherwise, in my mind, that’s okay, because at this point I see the potential for so much impact and for so much beneficial impact if it is done right, that I think one of the fears could be that, you know, so that there is a race to kind of be the first on something and that’s not ideal or something. But beyond that I think there are a bunch of other factors to consider: who owns the data, and how do we build these models? There are some business models to be worked out in different things, but in the kind of society we live in where we’ve managed to find our solution, I’m pretty sure those I’m optimistic about, like we’ll find a fair arrangement that works for everyone. But there are some of those things that need to be worked out. But overall, you know, Phil, I’ll go back to being my optimist self.

Phil:

Yeah.

Prem:

I think we’re going to figure it out, you know?

Phil:

Yeah, I was actually on a call about 12 hours ago late last night discussing a fairly difficult problem that we’re facing and our head of technology said to me, he said, look I’m pretty confident about this. I’m 90% confident that I’m right. And I said, my optimism is at the same level as your confidence. My confidence is not, but my optimism is at the same level as your confidence.

Prem:

But this is why we all love language. So, I mean, look. Absolutely. Wonderful semantic nuance that you’re teased out and it’s just such a joy to hear it, right?

Phil:

So, Prem, my final question for you. If you were running this interview, what is the question that you would really be hoping I would ask and what’s the answer?

Prem:

If I was running this and I know this is in the back of your mind, how is my family doing? How’s my wife? How’re my three girls? And you know, they’re doing wonderfully. You know, they’re the joy of my life. And, you know, part of what also drives me is to, you know, keep using AI to make this a better, more joyful place for all of them to be in. So I’ll say that was the question that I know is at the back of your mind and I asked it for you and I answered.

Phil:

Prem, it is always a delight to spend time with you. I’m pleased we could do this today. I’m sad that we’re not in the same room or perhaps over a dining table.

Prem:

Same. But I know we will be again soon.

Phil:

I can’t thank you enough for your time today. I think people are going to find this tremendously insightful and interesting, and I thank you from the bottom of my heart, Prem.

Prem:

Thank you, Phil. You know, the feeling is entirely mutual. And I’d also say, you know, I’ve seen you go through so many incarnations, you know, musician, you know, data maven, you know, growth officer and now podcast host! It’s just amazing. So, you know, I have to figure out

how to follow that kind of trajectory.

Phil:

But thanks again, Prem. I really look forward to the next time we speak.

Prem:

Yeah, take care.

The post LXT podcast – episode 4: Prem Natarajan – Chief Scientist at Capital One appeared first on High-Quality AI Data to Power Innovation | LXT.

Let’s dive in.

What is generative AI?

According to the World Economic Forum, generative AI “refers to a category of artificial intelligence algorithms that generate new outputs based on the data they have been trained on.”

How does ChatGPT describe it? Generative AI “is a subset of artificial intelligence that involves the creation of new and original content, such as images, music or text, through machine learning algorithms.” It goes on to mention that these models are trained on a large dataset of examples and then use this knowledge to generate new content that is similar to the original dataset but has never been seen before.

A pretty solid answer.

A brief history of generative AI

The history of generative AI dates to the 1950s and 1960s, when researchers first began exploring the possibilities of artificial intelligence (AI). At that time, AI researchers were focused on developing rule-based systems that could simulate human thinking and decision-making. Throughout the years, researchers began to experiment with generative models used in speech recognition, image processing and natural language processing (NLP).

As we approached the new millennium, new generative models such as Bayesian networks and Markov models were used in robotics and computer vision. Once deep learning was introduced and further developed, there was major growth. By 2014, the introduction of generative adversarial networks (GANs), a type of machine learning algorithm, generative AI applications were able to create authentic images, videos and audio of real people.

Today, generative AI is used in a wide range of applications, from creating art and music to designing new products and improving healthcare. With advances in technology and increased access to data, the field of generative AI is continuing to evolve and expand, providing new opportunities for innovation and discovery.

Generative AI industry impacts

The growth of generative AI models and applications has expanded past technology-based industries, impacting fields from creative-based industries to customer service. Let’s chat about some of those industries.

Data Analysis

Generative AI has transformed the way people work in the field of data analysis. With the ability to process vast amounts of data in real time, generative AI can identify patterns and make predictions that would otherwise be difficult – if not impossible – for humans. This has led to a significant increase in efficiency and accuracy in fields such as finance, marketing, and healthcare.

In finance, generative AI is used to analyze market trends and identify potential investment opportunities. If you remember from the first edition of this blog series, I touched on Robo-Advisors and the creation of automated investing.

Language Translation

The field of language translation has also benefitted significantly from generative AI technology. With the help of NLP algorithms, generative AI can translate text and speech between languages with great accuracy. This enables businesses to communicate more effectively with customers and partners across different countries and languages.

Customer Service

Chatbots and virtual assistants powered by generative AI have become increasingly popular, allowing businesses to provide 24/7 customer service and reducing the need for human intervention. This has led to significant cost savings for businesses and improved customer satisfaction.

Education

Within education, generative AI is coming up with new ways of creating and delivering content to students. For example, it is used to create interactive simulations that help students learn complex concepts in science and engineering. It can also be used to create personalized learning experiences by generating quizzes and assessments that adapt to each student’s level of understanding.

Further, generative AI helps teachers create new educational materials. For instance, teachers can use generative AI to create new lesson plans, generate customized worksheets, and even create learning materials such as textbooks. These tools are especially helpful in subjects like literature, where generative AI can create new stories or analyze existing texts to help students understand complex themes.

Marketing

Generative AI is transforming marketing by providing new ways to create and distribute content. The technology is used to generate personalized advertisements for each customer based on their preferences, browsing history, and buying behavior. This improves the effectiveness of advertising by creating targeted and relevant content.

Generative AI also helps companies create new content quickly and efficiently. Using an application like ChatGPT, social media posts, product descriptions, and even entire marketing campaigns can be developed quickly and efficiently.

Creativity

Generative AI has significantly impacted the creative industry, allowing artists and designers to create new and unique content much easier and quicker than before. For instance, generative design tools have become increasingly popular in architecture and engineering, allowing architects to create complex structures and designs with the help of AI algorithms.

In the entertainment industry, generative AI is used to create music and videos. The popular music app Amper, for example, uses generative AI to create original music for content creators, allowing them to produce professional-quality music in a matter of seconds.

Having spent most of my adult life working in the creative arts (my first career was in music), this application of the technology is of particular interest to me, but full disclosure – I am hardly a neutral observer. Keeping in mind my inevitable bias when it comes to the arts, I would want to make a distinction between the conception of “generative” as it applies to AI, and that of “creative” – this is a distinction analogous in my view to that between “craft” and “art,” or between “artisan” and “artist.” In each of these oppositions, there is genuine value on both sides of the equation. This is to say that the work of the artisan or the craftsperson can be highly valuable/valued, but it is arguable whether this work is necessarily art.

In my view at least, the value proposition for the respective halves of each opposition: creative vs generative; artistic vs artisanal; etc., is quite different. To my mind, great art (or literature, music, poetry) is characterized by motivated innovation. So whereas AI-generated art (in all its forms) is, I think, entirely a product of what it has learned or absorbed, real human-generated art (as opposed to craft) is a product of all that has come before while at the same time, a reaction- and often rejection-of what has come before. This is to say that the relationship between AI and training data is not, for the moment at least, the same as the relationship between human intelligence and the world of experience that forms our training data. AI absorbs data, whereas human intelligence has a more polemic relationship with data that involves challenging and arguing the validity of what it encounters. And this is a gap that I don’t see AI being able to bridge any time soon. But, I could be wrong (and have often been before!).

So, to conclude, as with any new developments in tech and AI, human interaction is still needed today and into the foreseeable future. AI isn’t quite sophisticated enough to pull in the same emotional appeal that a real person can add. But the future is bright and generative AI can support us as we create new bodies of work and take some of the hardships out of the process. For example, most of this blog post was written by a generative AI tool, specifically ChatGPT. A post about generative AI, written by generative AI (spot the boundaries?). I don’t know about you, but I think that is pretty cool.

Until next time!

The post Generative AI: A brief overview of its history and impact appeared first on High-Quality AI Data to Power Innovation | LXT.

Automated financial investing 

Whether you’re new to investing or well-traveled on your investing journey, automated financial investing may be an interesting way to level up your financial portfolio. Also known as “robo advisors,” automated investing uses computer algorithms to generate tailored financial planning or retirement advice. These computer algorithms are used to select and trade stocks, exchange-traded funds (ETFs), or other assets without the need for oversight of a human financial advisor. This cuts out the middleman and some of the barriers to entry for many investors who are just starting to build their net worth through investing. 

Robo-advisors are more affordable and convenient. Less research and management are needed to oversee your stocks since algorithms are fed all the information they need through AI and machine learning. If you want to learn more, check out SoFI’s post about Automated Investing 101. 

Vehicle design with digital twins and AI 

When it comes to AI in the automotive industry, most of us probably default to self-driving cars and vehicle’s virtual assistants. However, manufacturers are also enabling AI algorithms to build cars. Enter: digital twins. A digital twin is a virtual replica of physical objects. Their purpose? To run cost-effective simulations that mimic real-world assets in a digital space. 

In the automotive industry, digital twins are used to create replicas of an entire car, including software, mechanics, electronics, and the vehicle’s physical behavior. The digital twin holds all data from real-time performance to warranty data. This technology uses IoT sensors, log files, and other relevant information to collect this real world data and then is combined with AI-powered analytics tools in a virtual setting. 

The use of digital twins enhances factories and manufacturer efficiency. It cuts cost for product testing and employee training as this can all be done in a virtual environment. It also benefits the health of production lines and factories by determining the maintenance needs in real time. So, what happens when factories and more efficient and can cut costs on the making of cars? Vehicles become more affordable for consumers. 

Restoring ancient texts 

AI and machine learning are not only changing how businesses operate; they are also impacting history as we know it. DeepMind, pioneers in the field of AI, have developed a tool called Ithaca, an interactive interface that restores and attributes Ancient Greek inscriptions. 

Why is this so cool? Well first off, to the best of my knowledge, it is “the first Deep Neural Network for the textual restoration, geographical, and chronological attribution of ancient Greek inscriptions.” Basically, it’s a human brain on hyper drive and then putting it to work on the restoration of ancient Greek texts.  

Having a tool that can continue to learn and expand its intelligence while operating similarly to a human brain is what makes Ithaca so unique. Epigraphy – the study of inscribed texts or inscriptions-  is the main discipline for gathering evidence of the thought, language, society, and history of past civilizations. But between the constraints of current epigraphic methods and the fact that many ancient texts are damaged to the point of illegibility, historians can only do so much.  

Ithaca is designed to assist and expand the historian’s workflow: its architecture focuses on collaboration, decision support, and interpretability. What the creators of Ithaca have done is unlock the cooperative potential between AI and historians, which can transform the way we study and write about ancient civilizations – for Ithaca, specifically Ancient Greece. 

Can you imagine? Ancient Greek texts that have never been able to be restored due to years of ruin suddenly being restored and attributed to the correct time period. As a history buff myself, I can’t wait to see what they discover! 

As you can see from these examples, the impact of AI is far-reaching and is enabling breakthroughs that will fundamentally change the world we live in. Join me next month when I’ll share more of these exciting applications with you! 

The post Real world AI applications: robo advisors, vehicle design and restoring ancient texts appeared first on High-Quality AI Data to Power Innovation | LXT.