NBC News on CACAA Lomalinda sponsored by Teamsters Local nineteen thirty two, Protecting the Future of working Families Teamsters nineteen thirty two. Dot org. The information economy has a rid. The world is teeming with innovation as new business models reinvent every industry industry. Inside Analysis is your source of information and insight about how to make the most of this exciting new era. Learn more at inside analysis
dot Comsideanalysis dot com. And now here's your host, Eric Kavanaugh. Yes, oh, yes, indeed, folks, Welcome to the future. The future is here already. That's what William Gibson once wrote. It's just not evenly distributed. I have to say I heard that line years ago and it just blew my mind. I was like, that's pretty cool. The future is here already. It's just not evenly distributed yet. But that's happening right now. You see it with cell phones, for example. You see it
with certain kinds of construction, skyscrapers. You look at the new skyscrapers and places like Dubai in the UAE and they're just amazing. And that is the forefront of the industry. And it's primarily because they have a lot of money.
I want to build some cool buildings, and in the rest of the world you have older buildings that are still standing, obviously, But I bring in this metaphor just to point out that we are at a very transformative phase or moment in time as human beings, quite frankly, and with technology, of course, it's always out there improving, usually improving our lives, doing
cool things for us well. Artificial intelligence it's not new. It's been around for decades, fifty sixty years if you really push it all the way back even further depending upon your definition. But in terms of widespread adoption and use, boom, we are in the middle of that first major explosion. And you can thank the folks at open AI for sure with their chat GPT, which just blew the doors off. Now that even isn't all that new. GPT has been around for a number of years now, but this version that
they unleashed on the public really opened eyeballs and it forced the game. It was a forcing function. But I have someone on the show today who knows a whole lot about AI and has known a lot about AI for a long long time, and he taught me a lot about this stuff too. Asama Fayad. He's got a company, Open Insights, but also he was the chief data officer at a company called Yahoo. Some of you all remember those guys. They were the center of gravity in like the nineties and the early
aughts. That was where you wanted to be as an engineer, as an AI person. And then of course it kind of shifted to companies like cloud, Darra and Google of course, and now open Ai. But they're still out there. Yeah, he's still doing cool stuff, and so is usama A Sama Fayad. He's now the executive director for the Institute for Experiential AI at Northeastern University. Sama, welcome to the show here. Thanks so much
for your time and attention. And first of all, tell us about the Northeastern University Institute for Experiential AI. Where did this idea come from and what's your focus? Thanks? Thanks, Eric. Idea came from the president of the university, Joseph and the provost David Madigan, who's actually a well established statistician an AI expert in his own right, and some donors who gave some
large grants saying hey, AI is going to be big. This was about This was five years ago, So around late two thousand they started saying, Okay, we've got a lot of funding, let's recruit for someone to head it up and define it the way we I'm the inaugural executive director for this, as well as joining as afessor in the Computer Science College, Corey College for Computer Sciences. So experiential AI. The first thing we need to do is name it. And we use the term experiential AI to meet AI with
the human in the loop. Because the reality of my exposure to AI through my work, whether it was with Microsoft, NASA's GPL, my own startups, Yahoo, later with Barclays, and definitely with Open Insights, working with many companies around the world, including AI in the healthcare space, etc. Has shown me one theme that is constant, which is every working AI has
a huge element of the human in the loop. Whether you're talking about the core of the Google Search Engine or we're talking about open AI and chat GPT in any of these large language models, a lot of feedback all the time from humans supplying what AI cannot supply. And we, like you mentioned, we've spent over seven decades trying to figure out stuff like common sense reasoning. One of the big learning lessons is you only can get that from humans today.
You cannot get a machine or an algorithm to provide you that feedback, which is why there's a lot of human in the loop intervention in the operation of AI the other and that became a theme because what it means is you need to think about how AI help human intelligence and how human intelligence helps machine intelligence or AI become a reality. So it's about kind of the collaboration with
algorithms and robots and so forth. So that's how we define it. And then we had to go look for, Okay, what are areas that need focus where as an institute we can make a large contribution, We can become the leader in certain areas and quickly IDENTI responsible AI as one big area of focus from a research perspective, defining what it is, how do you do it, how do you do risk assessment, et cetera. And then we also identified kind of what I like to call new generation large language models or
new generation generative AI. So how do you reduce this obsession with larger and larger parameters in the model, Larger models that can so easily go out of control, start making errors, hallucinating as Google calls it or basically are extremely expensive to train because the more parameters you have, the more and of course capital equipment wise that the training infrastructure and the inference is so Rather than how can we make models bigger, the question is how do I get away with
the smallest possible large model and enhance it by stuff that we know prior knowledge? Do I have a knowledge graph about the domain? Do I understand who the people are, professional graph, a citation graph, et cetera. It is silly to try to let the AI rediscover from pure data, very very expensive data, what we already know. So can you supplement this? And
we call this better together? And then we picked certain areas and we are seeing better results indeed qualitatively and qualitatively when we apply these two things together. Finally, we had to pick kind of areas of focus for applications, which is critical for us. So applications, in my opinion, have been the driver in the developments of AI for the past decade or more. So. We looked at applied areas in AI plus health, AI plus life sciences,
drug discovery and molecule design and things like that, antibody design. And then the last area we added was AI for climate and sustainability. How do we leverage the new generations of data sets that are kind of hyper local, very detailed, huge big data, if you will, in the world of climate and then use it to model both the climate and the environment and predict things like here are the impacts of a potential disaster or a change or what have
you. So these areas define where we work and to drive the wheel forward. To start the flywheel going, we created the AI Solutions Hub. One of the big funders of the institute is Dave Rue, who funded also the Rue Institute up in Portland, Maine, in order to change the economy in
Maine, and AI in his vision, was a big area. So we created this AI Solutions Hub, which is a team, a delivery team of data scientists, data engineers, AI practitioners whose job is to engage with our partners companies, organizations, governments and work on real child just in the real world, in order to drive that activity, figure out what the real research agenda should be and at the same time figure out how do we train our
students and the corporate learners on AI in the wild, AI happening in reality? What are the real issues? How do you deal with it. You know, we all know that dependence deep dependence of data and generative AI, of generative AI on data. But the real issue here is what do you do? And the data is not an idealized circumstances. It's got quality problems. Half of it is missing, a third of what's there is wrong.
But you've got to make the solution work, and you've got to make it work in the constraints of kind of real life and real businesses and real regulations and so forth. So all of this coming together gives us, if you like, the active lab that generates the research agenda and generates the chance to train our students and learners from the corporations on how to make AI solutions work.
Yeah. I really love the vision here because you're appreciating the importance of research but also the importance of training, the importance of understanding use cases. Hence the solution portfolio or the solution hub. And you know, I want to drive into one key issue here that you brought up, because I think it's the most important one to make this stuff enterprise ready and business ready and consumer friendly, which is the truth. Right, We have one of our
regular listeners Alex Husky, super smart guy. He works for rex On Mobile. He sent me an email we did an webinar a few weeks ago. He said, we cannot allow these large language models to pattern recognize their way to the truth. And that's kind of what they're doing, is there. It's basically a pattern recognition engine that predicts text based upon the prompt and all
the corpus of text that it has consumed. Well, you mentioned citations for example, and this whole concept of embeddings comes into play, and that's what you need, right, that's the ground truth our citations. This is where got the information so you can verify that it's correct. That's a cornerstone of making this stuff doable and workable for business, right, absolutely, absolutely,
And you know, citations are a beautiful example. You know, this autocomplete technology that we call large language models, it's trying to you know, build its output a token at a time, and the token is part of a word, part of a group of pixels, depending on what your output looks like. Well, trying to recollect a precise citation based on an autocomplete is
a very dangerous exercise and rife with room for making an error. You can get the year wrong, the author's wrong, you can make up, you can come up with a title, You could change the title all based on your recollection. Well, why do all that? If you already have these
citations in very precise format, how can you use them? And then tying the citations to the content, to the meaning is the other big challenge because most of these models, I mean we call them technically stochastic parrots because they really parrot out. They don't know anything, they don't understand what they're saying. So tying that to kind of the sources of truth, as you say, is another huge challenge which doesn't exist today. Well, and also we
have to realize there is a workflow to these models. And what I'm learning from folks is you can do one of two things. Basically, you can fine tune your model by embedding your corporate data into the model, so training the model itself on your corporate data. Or the other very common use case is to use a vector database, and that's where you store your embeddings. And so embeddings would be your financial reports for your business, or your training
modules or documents related to some process in your business. You want to populate all that and ideally just use the large language model for its text generative and syntactical knowledge essentially, so it's you're using the engine to spin up the words, but the facts, if you will, come from the embeddings. Isn't that the general idea? Yeah, and well you're touching on some of the more interesting aspects of large language models. So let me start by saying,
first of all, no one understands why they work. Of why a very large model that traditional statistics will tell you should never work because it's over parameterized, it's got too many degrees of freedom it can fit the data, and therefore no chance of learning why it works. That's one interesting mystery, right
that we don't write. The other mystery is this whole embedding space. Why is it that if I took a group of words, or a group of sentences, or a group of paragraphs and map them into this very large vector space, which is what's called embedding a bunch of numbers. Basically, why is it that those embeddings capture something about the domain that allows you to generalize,
that allows you to build on knowledge. So that representation is another mystery eric that we have people working on trying to demystify and understand what the heck is going on. Those two aspects are some of the deeper aspects of the technology that represent an academic and research challenge trying to understand what's going on and how do these models work? The fact is though they work, and therefore how do you best leverage them? And then how do you add these embeddings
with the other generative stuff which is straightforward and well understood. The autocomplete technology if you like, that kind of takes an input a prompt and produces the output. The other thing that I'll mention here, which is the third big mystery in the field, is the prompt themselves. How do you come up with the right prompt. We know there's a super high sensitivity between prompt variations and output variations, so small disturbances of the prompt may produce a completely different
answer. Some people claim that, hey, these are the new hot jobs. There was a New York Times and Financial Times articles on you this is the highest paid jobs, hottest jobs. My colleague and a core member of the faculty of the Institute for Experiential AI at Northeastern Byron Wallace, summarized it very nicely in our recent conference where he said, this is not prompt engineering. This is more like incantations and black magic, right, and you know,
you might stumble on the right incantation, you might not. And he has a field of study that has to do with Okay, how do we reduce that dependence, how do we reduce that sensitivity, because at the end of the day, to make it practical and useful, you can't have this huge variation and kind of weird stuff. And he says incantations because sometimes prefixes to prompts that you might you and I might think have nothing to do with it end up being very useful to guide the model to the right space.
So it's people are adding weird stuff in the front and in the end of a prompt just to make it work. Right. That is not a science. That is not engineering. We got to figure this out. That's funny. You know, I'm gonna I'm gonna close this opening segment here with a fun story and basically say that this is not as crazy and different and unique
as some people might think. So, I had a business partner for a number of years back in the year two thousand and I think earlier you were talking twenty men twenty twenty, I think when they started this whole thing. So it's five years ago. But back in two thousand, I was working for a company Keem Jim, a couple Russians, and the good Russian was named Nikola niko Lem like this one thing. He said to me. The funny thing about the web is nobody know how it works. All we know
is it works right, And it's the same thing. So now what's interesting is if you look at Internet and cloud and hybrid architecture and all this that people are talking about today, containerization, Kubernetes, all that stuff, Well what happened is Google came along with Kubernetes as this orchestration layer to sit on top of Docker as a way to containerize applications make them really really small, which is like the vision of service storryed architecture, but a different bus,
if you will. And what's happening is now we have all this data, we have tons and tons. It's called observability. So what's happening now is we're finally starting to kind of figure out how the web actually works. So I'm not saying we didn't know anything, but there are a lot of stuff that we didn't really fully understand about how the web actually works, like which ports are important in what particular process. I mean, there's all this stuff
going on. So my point view is that this AI stuff, You're right, we don't know how it works. It's not the first time we've found some engine that works and we're like, well, we don't know exactly how it works, but it sure does work. But folks, don't touch that dollar. We'll be right back talking to Osama Faiad of the AI Institute from the Experiential AI Institute, will be right back. Welcome back to Inside Analysis.
Here's your host Eric Tavanaugh. All right, folks, back here on Inside Analysis, your host Eric Kavanaugh with doctor Usama Fayad from Northeastern Universities the Institute for Experiential AI, which basically just means humans in the loop. Humans are really important. Even with self driving cars, you have to give correction to the vehicle. I've actually driven a couple of these now, and it's kind of freaky when it starts taking the wheel from you and like, WHOA,
what are you doing. I didn't even know it was self driving. It was a rental car. I'm like, oh, what's going on here. Okay, it's taking over for me and seeing you shouldn't be doing something. We're getting there. It's I mean, thanks to innovations, thanks to Elon Musk and Tesla and all these innovations. Those folks are doing pretty cool stuff. But doctor Fayad, we wanted to dive into the critical six sess
factors, and one absolutely positively is the data. You know, I keep cautioning people, and we're going to probably dive into this more in our own business of consulting with organizations. Be very careful how you train your models. You want to train your model on your data. You want to give it curated, trusted data. You don't want to just point it at your whatever information system you have, because there's a bunch of junk in there. And
anyone who's ever done data quality initiative knows. If you get an executive doesn't believe the data is crap, just show them. Just open up any database and point to any corner of it, and you'll find some problem somewhere. I promise, it's just everywhere, whether it's data in the wrong fields, data in the wrong format, data that was truncated during some ETL script or
something, and it's just everywhere. There's bad data everywhere. So getting the right data is really important, doctor Fiab, How do you do that? What do you recommend? When people come to you and ask you what are the best practices? What do you tell them? Yeah? Great, great question. So so data is a whole ocean to open up. There's the collecting the data at quality. That's collecting data at the right level of granularity.
So we have spent decades, as you well know, Eric, in the whole area of data warehousing and kind of information systems management and so forth, and data management thinking about data for use by humans. Now what the AI haves. The AI first companies or the companies who really are using AI today, they have a deeper understanding that, Hey, machines needed data at
a different level. They need high granularity. Not necessarily stuff understandable by humans, but every little detail of what happened, every log of every event, every kind of intervention correction by a human, et cetera. Captured. This is not how most organizations or companies work today. So that is one big shift that needs to happen. The second thing is, if you notice a lot of the literature and lower in the world of data has to do with
structured data. Well, according to Gartner and many others, the majority of data in any organization, in any company is actually unstructured data. If you actually talk to the IT teams and the technology teams and so forth, they only typically only know how to deal with structured data. And they hit unstructured
data, they're stuck. It's like in the headlights. We have to change this dynamic to understand that, hey, we got to embrace the unstructured data, the text, the images, the video, the audio, interactions in the customer center, all the conversations and leverage that information properly, and that needs. There exists many good tools out there. Most of them also, fortunately are in the open source, which is something I'm very big fan of.
But you can put these things together in solutions that work for businesses, and this is something we're very passionate about. How do we partner to figure out, how we help our partners figure out how to approach an area, how to the data, et cetera. Now that leads me to another very very important data that's information or aspect that's very related to data, which is
kind of many companies out there. Think that all we have to do is wait for Microsoft or open AI or whoever to refine their models so that they can meet our needs. What happens then, is you become users of the technology. You don't have your own private language model. You don't have the
ability to capture all these corrections. You mentioned corrections when you're driving right, Well, it's great to intervene and correct, but if you don't capture that correction, the context for that correction, why you made that correction, your model is never going to benefit from it unless you're willing to share it back with the company who's providing this public utility, this large language model. Often
this information is sensitive, it's proprietary. I argue it's one of the most valuable sources of IP than any company operating in any space has, which is kind of the secret sauce. How do our processes work? Where do we intervene, why do we fix what is our way of doing things? Today? Companies let all of that go into the exhaust by never capturing it. In reality, you need to be able to not just capture it, but feed it back. And feeding it back leads us to another whole interesting area,
which means you really need to have your own private language model. That model can't be too huge, like the likes of open ai or chag GPT or any of these. It has to be as small as possible to maintain its stability, to maintain its maintainability and its reliability and so forth. But
it needs to get feedback. So you need to be able to capture all that feedback, all those corrections, and feed them right back and retrain that language model in an affordable way to maintain your own IP, your own knowledge, and so forth. Now, many of the companies who offer the large language models will offer you the chance, of course, for money, to kind of fine tune the model or create your own tuned version of the model,
and there we can talk about technically what happens. All they do is they cut off the output layer, they throw it out, and retrain an output layer they call these adapters that is kind of fine tuned just to your problem and on your feedback. But we know that that's less effective than retraining the model and reintegrating that data into the whole stack. So coming up with a large language model, your own private language model, is not impossible.
It is very doable. Many tools are available out there, whether you're talking Falcon which came from Abudawi or Lama or Lama too. I'm a big fan of at Facebook or Meta because they actually not only did they make it open source, they gave a commercial open source license. So companies, sorry, yeah, and that opens up a whole other approach and a whole other mechanism. But to do that and to take advantage of it, you need to have the talent and then know how of how do I use that language model,
when do I find tune it? How large does it have to be? How small can I make it? All of these questions are the kinds of questions we partner with companies on kind of answering and figure out what is the right approach here in this very very confusing, and I might add fast evolving space. I mean I try to keep up and I am challenged, right every week, every month there's new innovations, new ways of doing things. Somebody's crapping something old and saying it didn't work. Actually it was part
of the system, but now we've got to do something completely new. All good, all healthy, but it's a challenge to keep up with the latest, greatest and the right best practice of how to use it. Yeah, start with I just want to remind everybody that data that you're letting go into the exhaust by not capturing it systematically. That is gold that is actually more
valuable than gold. It's not just oil. It's more valuable than gold, and you're letting it go out in the exhaust and losing all that ip that your teams, your employees are working on day in, day out and keeping operations going. Yeah. Well, you know, I just as you were talking, I thought of an interesting metaphor here, and I'll throw it at
you. I think you'll you'll find it interesting. So think about language and feeding a dictionary of English to something like one of these large language models. You feed it. The dictionary knows all the different words, but people don't talk the way they write, and people don't talk the way dictionaries are written. Necessarily, there's a lot of slang, there are lots of expressions, idioms, all these different things that don't fit neatly into the definitions of words.
But it's how you actually speak when you're talking to someone. And that's the unstructured data, really, er, that's the unstructured syntax. Almost of how people actually interact with words on the phone in person, and certain cultures and in certain regions, and so if if you don't absorb that that contextual information, your engine will never speak like a normal person does. Right, So this is the nuance you have to appreciate, is that what's on paper
often is very different in the spoken word. What do you think? Absolutely, and it's critical in many applications, actually, I would argue in most of them, if you're dealing with conversational AI, if you're dealing with dealing with humans, whether you're doing it for marketing purposes, whether you're doing it for customer service, whether you're doing it for predictive maintenance. And it's not only kind of the idiom or kind of what the public, how the public
or the customer speak, it's often the culture within the company. Your engineering team has their own terminology where they use certain terms. If you come up in the rights, they mean something of it, but they're using the company to mean a certain aspect of a process, a certain aspect of a system. Learning all of that and making it work in the right context is a real interesting challenge, and it's very possible. I mean, that's why the
data is. That's why I describe it as gold or more valuable than gold, because it contains all these instances and it gives you the keys to allowing these systems to have the bigger and correct impact on how you do acceleration, how you do automation, how do you leverage generative AI to kind of give you that unique advantage competitive advantage out there. I've experienced it personally in healthcare.
Right if you're in an operating room or in an X ray room or in an imaging center, you know, the language being used is completely different than what's being used in you know, a pathology lab or a practice for physical therapy or what have you. So all of these things, all of these aspects are important. What contents to use, the words in, what they mean, new meanings that are not available in the dictionary. All of
that is so important and so critical technically, socially and marketing Whites. Yeah, well, and there's also tone of voice. And I remember, going all the way back to like third grade or something, our teacher asked us a question and it stumped all of us, and I thought about it later, I was like, Oh, tone of voice, that's what he was talking about You can say something and mean it honestly, or you can say it sarcastically. And usually it's the tone of voice and the context that will
tip off a human being. But it's going to be harder for a machine to understand that, but not if it has the intonation that if you capture the audio and it can kind of pick up on that sort of thing. But these are the sort of deeply understood aspects of communication that we just know
as grown ups, that we've learned over time. I mean, I remember, this is funny, but probably most people don't remember this, but I remember the first time I really experienced sarcasm because I was like two or three or something and I'd spilled milk and my mother and I knew it was like, oh no, this is bad. And my mother looked at me, she goes, well, that's just great, and I was like, hold on, I know that great is good, but that tone is not good.
What's happening here? And I was really defuddled as a child. But that was my first experience with sarcasm. Right, I'll give you here. You're touching on one of the biggest challenges that, for example, search engines today face, which are memes. Right, you hit a meme on the Internet, the search engine thinks it means one thing. Reality is, it's something completely different. And the only way you catch that, The only way
you catch that today is human intervention. Search editorial team saying, wait a second, this shouldn't be in your search result. This is a meme. It means something completely different. It has nothing to do with biology or whatever it's it's being used to refer to a group of people or a group of practices or what have you. So those are those are real things that matter
a lot, and you have to have mechanisms for catching them. And you know, in a way, you can argue that a lot of the internal culture of a company's operations team, a company's engineering team, a company's business team, a company's sales team, a company's marketing team is filled with these memes. We don't call them memes, but they are there. That's funny.
There's actually a great Seinfeld episode where George Costanza, he's working for the Yankees and I think he was talking with it was like the Texas Rangers or something, and as a baseball team, and the culture of the executives for that team loved using curse words. So they were insulting each other all the time, but as a compliment. And then George's boss walks by as he's on the phone using all these expletives, and the boss is like, oh no, what are you doing. This is terrible. In reality, he's
bonding with these guys because the context was there. Classic dramatic irony, right, the one guy didn't know the context. But this is the importance of nuance and understanding context. And to your point, just getting it back to training AI your own private language model, which is what most businesses are going to want to have. I'm pretty sure you have to bake in that kind
of stuff. You have to have that human in the loop that goes, okay, no, this is an example of sarcasm or these guys are just being goofy or whatever in their context. Calling you a jerk or a so and so is actually a compliment. That's the human invention part, right, absolutely, absolutely, And I remember that episode. But the thing I wanted to mention is, you know, language, the language use is very different
than kind of the formal language defined. That is why I believe, like you, most companies will find out that achieve true competitive advantage to have differentiation, You're going to have your own private language models, which opens up a whole new area. We are a company that's in the you know, quick service restaurant business, or we are a company that's doing manufacturing, what do we know about large language model AI, et cetera, and how do we
approach it? And part of this demonstification and kind of working with you on the AI journey is one thing our institute is very passionate about because that actually puts us right up front on the front lines of getting the applications to work and helping our partners getting it to work. And through that we will learn what the real research problems are, where the real challenges are, and we will learn what should we be teaching our students and our corporate learners in order
to be effective in that world. So that whole area of the infrastructure, the talent that know how you will need to build your own private language models to capture your own IP and your own knowledge. It's going to be a whole, huge new trend that no one out there is paying attention to in the business world. They're focused on being of these public right, No, that's the very very good point. Well, folks, we're talking to doctor
Usama Fai will be right back. Don't touch that doubt. You're listening to Inside Analysis. Welcome back to Inside Analysis. Here's your host, Eric Tavanaugh. All right, folks, back here on Inside Analysis talking all about experiential AI. That's humans in the Loop with doctor Usama Fayad of the Northeastern University Institute for Experiential AI. Got to get the all clean and straight in my head. The Institute for Experiential AI, it just means humans in the loop
basicly. But we're talking about the power of these models, which is amazing. We're talking about how a lot of people don't really know how they work, including people who design them. We just see that they work, and so we're coming up with strategies to help companies really harness the technology. And I think that's the key, right is this is if unbridled, this thing will go wacko and cause you all kinds of trouble. But if you harness
it effectively, and that's your governance structure, it's your embeddings. It's using ontologies and I mentioned my friends from Praxy Data who are using ontologies to really narrow the focus of their work and thus be able to deliver some tangible business value. And you know that's always been the case with machine learning and AI is you have to have a very narrow focus and you have to know what you're trying to accomplish. I mean, you can use deep learning to discover
things, but even then you're trying to discover things. So you have to have a purpose and a mission and an objective upon which you train these models to be able to understand if you're getting anywhere right. Tell us a bit about that and why ontologies matter, Yosama, Well, I mean, ontologies give you the basics of understanding domains. There are many things you have to
worry about. And by the way, not coincidentally, a lot of the work in generative AI now is using what we call the RAG framework, which stands for retrieval augmented generation, which is another form of kind of enforcing the mode to stay within certain guardrails and to say here's what we mean. And it's aided by kind of retrieving the right references and all that to help it
from straying, both in training and in answering. But anyway, you know, working in different domains you need to figure out ways of bringing in that domain knowledge, that domain into into the problem in order to constrain the space in order to focus it. You know, I was talking about the secrets of making AI work, and we talked about the humans in the loop or must interventions are necessary? Data at the right level is absolutely necessary. Most
data is being lost. But I will I will add the last bit is how do we bring in non domain knowledge and how do we narrow the problem down to a very narrow, very specific problem. Because trying to solve AI in general has proven to be an impossible tasks, a very difficult task for the past seven decades, and it will continue to be that, and our best prescription for fixing it is by narrowing the problem and resorting to what we call narrow AI versus general AI. General AI is still far out of reach.
Narrow AI has worked and has been working for many years and decades actually in different domains, and the key there is domain knowledge. The key there is narrowing it enough so you know that not every possibility is here, and you don't have to worry about every language, and you don't have to worry about every variation and every meaning no, restrict yourself to this domain. Here's what things mean. Try to get complete knowledge around that domain. And that's
a secret of making it work. That only comes by figuring out what are the use cases, what are the domains of applications, and how do I come up with a methodology that allows me to build solutions for each specific use case. Which is why we kind of focused on things like AI plus health. How do we get healthcare applications into AI? How do we leverage digital health? You know, we live in a world today, think about at
the highest level. You go see your doctor, hopefully once a year, maybe twice a year, right only when something is wrong other than the routine visits. And what happens between these routine visits is a total mystery to the provider, to your physician, or because who knows what happened to Eric during this one year? Now Eric needs to fill me in, and who knows whether Eric can fill me in completely or skip a whole bunch of important things. Well, guess what. We live in a world now where there's a
lot of instruments. We've got these watches, we've got the mobile phones, we've got the you know, all sorts of devices, cameras, other things in our homes. They can actually capture a lot of the information that typically is only captured in a clinic. So how do we leverage that to close the loop at the highest level that we call that digital health. Another form
of digital health is things like oncology, or actually more generally pathology. How do we bring these digital techniques down to a pathology lab where somebody is taking a sample, looking at it in a microscope and kind of storing it in
this glass slide. Well, guess what digitization can help a lot. You can look at it with instruments and with algorithms instead of just human eyes, instead of just relying on staff to do you can strike capturing what's happening in the heads of these very talented pathologists who are looking at an image and concluding a result. This happened in radiology in a big way. Radiology has been digitized. It benefits from a lot of these things. In pathology, it
hasn't happened. Pathology is still practiced like it was, you know, three or four decades ago. It can benefit from a lot of digitization, which brings in a lot of AI and a lot of use of these things. In life sciences, we have many use cases, whether it's the acceleration of drug discovery, how do you design how do you accelerate the design of the next antibody? Right antibodies are very effective ways of coming up with cures,
vaccines, ways to mitigate disease spread. But it typically takes about three years to come up with the right antibody for something. So it wasn't a viable solution for COVID back in twenty twenty or twenty twenty one because it took three years. But if I can get that three years down to three months by accelerating it through compute and through modeling, amazing things can happen. Right, take things like that happen in a hospital or a physician when you're being taken
care of. In one of the projects we have is called heart, which expands to something involving AI. But what it is one of the applications and working with a large hospital system, is how do you determine when somebody should
not be downgraded from critical care? They're in the ICU and things look good, we can downgrade them, but the algorithms can tell you no, no, no, no, there's a huge danger here if you downgrade that issues would recur, which is horrible for the patient, horrible for the hospital, horrible for the physician. And you want to make that decision when you're absolutely
sure that no complications will happen. Similarly, you don't want to keep them in ICU too long because that's not great for the patient and it's a use of a very limited resource, the ICU. So a lot of these problems come up in a lot of places. And there with my colleague Ray Winslow, who's also a core member of the faculty in the Experiential AI Institute, he leverages what's called computational medicine. How do we model these things? How
do we model processes, how do we model patients? How do we model diseases and use those to come up with predictions that are hard to get otherwise. So those are applications in healthcare, in life sciences. And our newest area is the climate and sustainability. How do we leverage kind of new data
sources and models, et cetera. Yeah, And to drill real quick into this, the ICU example, which is a fantastic one I'd like to point out for our audience, And let's get some feedback on this, se me an email info at dmradio dot biz or info inside Analysis dot com both come to me. You know, at any point in time in a workflow, you can insert an AI suggestion, for example, or you can have it
sort of waiting to make a suggestion. So anytime the clinician is there and they're typing information, they're doing stuff as you go to the screen where you're going to think about maybe downgrading this person from ICU, that's an excellent point
at which to offer an opportunity for the AI to give a suggestion. Right, So it'll come in and say, I see you've just launched this screen, maybe you should reconsider because of X y Z. And I think that's really where a lot of this AI value is going to manifest is in suggestions
at certain points in a workflow. Right, absolutely, Eric, And you've touched on a very important area where my colleague Professor Ray Winslow, who I mentioned working also with Professor Gene Tunnik, who runs our AI plus health area for the institute. They like to call it, how do I make AI a valued member of the medical team. That word you use, the recommendation is key, right, AI doesn't know everything. It may sometimes it's right,
sometimes it's wrong. The feedback from the physicians or the care team and nurse could be anybody, would be very beneficial to both improve the AI and to allow that AI to contribute to the proper care of that patient. So I love that analogy, which is, you know, AI needs to be just another member of the healthcare team, making its own recommendations and learning from corrections, interventions or getting it right and wrong. That's right in learning.
And here's the beautiful thing. You know, doctor I had mentioned you have to capture all this information. When you correct the engine or when you accept the recommendation, all of that should be captured as well, because then over time you can see, well, doctor Jones never accepts the recommendation of the AI engine, and doctor Phillips always accepts the recommendation of the AI engine. Somewhere in between is probably where we want to be. But the point is,
as human beings, we are actually very predictable. And you will know that if you start looking at your device that giving you feedback on how often you are on this app or that app or all these different things, you will see a reflection of your behavior and you'll realize why I am pretty darn predictable every day at this time I do X. But hey, hang out for one second. We'll be right back after this break. You are listening to Inside Analysis. Okay, folks, time for the podcast bonus segment here
on Inside Analysis is always so much fun talking to doctor Fayad. And you made a point there after in the break of out the importance of trust and building trust in the system, for the system, for the doctors, for the team. Go ahead with that. Yeah, So, I mean one of the critical pieces of making the AI a valued member of the medical team or the care team is that ability to kind of make suggestions and get feedback
those suggestions of recommendations. They get value, you know, they get evaluated by a physician or a nurse or whatever. They may get corrected every once in a while, which gives them a chance to contribute to the system and it's evolution and they feel that their baby. But at the same time they get kind of this incremental exposure to hey, this thing is right. It
caught something I missed. So it's useful and by the way, applies equally well whether you're dealing with a medical team or a factory, a team of engineers in a factory, a team of operators in a monitoring center. Every application where the AI can contribute, you need to gain the trust of the users and domain experts, because that's the only chance you're going to have of getting acceptance, getting important data, getting feedback, and keeping the system healthy.
Yeah, that's right. And you know what I've noticed about the AI that I've seen in practice, which I see in my email or my text messages in other places, it's like, oh, you haven't followed up on this text, you know, do you think there was three days ago?
Basically, that's a perfect example of a reminder that is helpful because, let's face it, even the best doctor in the world will forget sometimes to do one little process, one little piece, you know, and depending upon the use case, that could be a very serious thing or not so serious thing. But you know, you forgot to wash your hands, for example, you forgot to put the mask on, you forgot to put the scalpel back
where it belonged, these little teeny tiny things. Because guess what, People get distracted, even very focused people can get distracted, and so the AI is there as your little helper watching you, making little suggestions along the way, not wrapping you on the knuckles, getting you in trouble with your boss necessarily, but just giving you a little bit of a hint and a nudge. And it reminds me one of my favorite expressions. That's an old cliche,
don't here much anymore. And of course it's it's posed on man, but it means men and women. It said, man doesn't need to be taught so much as reminded. Right, we already learned. We just have to be reminded. Final thoughts from doctor Fayette go ahead, Yeah, and absolutely, and you actually need both. You know. Learning in the system is also critical because you know no system is going to get it right. So every once in a while it's going to guess the wrong thing and it
becomes annoying. It's reminding me of something I already know. Right, well, I need to give it that feedback, hey, in this context, don't remind me. And then if it adapts to that. Now you have intelligence. Now you have something you value and it's personalized. So now you want to keep it because now it's beginning to get to know you and your
preferences versus me and my preferences which could be completely different. So bringing machine learning into the loop and making sure these systems learn and adapt to the feedback they're getting is very Yeah, this is wonderful stuff. Well look this, gentleman up online and look up Northeastern University, the Institute for Experiential AI.
That just means humans in the loop. Doctor Fidgen here and you folks have put together a great program, and you understand the interplay between training and research and development and solution designed. All these things feed each other, and we'll feed each other over time. So you know how to train people, know how to get people excited about stuff and really nudge them and get them involved. Because the humans in the loop, they're the ones who are going to
make all this stuff successful and worthwhile. And it's going to save lives in the healthcare space. I promise you it's going to save lives. It's going to save time and effort. It's going to do the most important thing I think for any business that has improved morale. When MOREL goes up, good things happen. When MOREL is down, good things do not happen. So just that's the Dallas Cowboys. Thank you, Thank you, Eric, and thank you for getting our mission and vision for the institute. You got it
perfectly. I love it. Well, we'll talk to these folks again. I'm sure folks don't touch that boy. Actually, well, we'll touch you next time. Infodinsid Analysis dot Com. Bye bye Marino, Valley, Corona and Riverside. Listen to KCAA, KCAA the station that leaves no listener behind and now the voices of KCAA was an exciting announcement. Want to hear NBC and News or KCAA anywhere you go? Well, now there's an app for that. KCAA. You is celebrating twenty five years in our silver anniversary with
a brand new app. The new KCAA App is now available on your smart device, cell phone, in your car, or any place. Just search KCAA on Google Play or in the Apple Store. One touch and you can listen on your car radio, Bluetooth device, Android Auto or Applecar Play. Catch the KCAA buzz in your earbuds or on the streets. Celebrating twenty five years of talk news and excellence with our new kc AAP. Just do it
and download it. K c A celebrating twenty five years t Hebot Club's original pure Power Drgo super TA helps build red corpusols in the blood, which carry oxygen to organs and cells. Our organs and cells need oxygen to regenerate themselves. The immune system needs oxygen to develop, and cancer dies in oxygen. So the T is great for healthy people because it helps build the immune system, and it can truly be miraculous for someone fting a potentially life threatening disease
due to an infection, diabetes, or cancer. The T is also organic and naturally caffeine free. A one pound package of TA is forty nine ninety five, which includes shipping. To order, please visit to Hebot club dot com. T hebow is spelled T like Tom, a h ee b like Boyo. They continue with the word T and then the word club. The complete website is to Hebot Club dot com or call us at eight one eight six one zero eight zero eight eight Monday through Saturday nine am to five pm
California time. That's eight one eight six one zero eight zero eight eight to Hebot club dot com. Hey, the Big Games this Sunday. Who you're rooting for? Joe the Browns? Well keep waiting, No really, Joe, what you doing for the big Game? Well? Who has an eighty five inch TV and you type it to watch it? Oh? I know that place. It's the place that you type on and you type of boulevard across from the golf course. The Fat Greek. They have an eighty five
inch TV to watch a big game on. Hey Joe, why does my dog not like the Big Game? Probably likes the Puppy Bowl. Oh, the puppy Bowl because he's a boxer. Soo's your underwear? Ah? The Fat Greek has the big Game. Yeah, Fat Greek has the big Game. The Fat Greek has the Big Game, big stream and big special. So Mark, what is Sir mix a Lot's favorite part of the soup? I mean the big Game? Oh? No, what what? He likes big punks that he can out lot? He likes big punks and he cannot
like. Oh okay, the Big Game starts at three thirty pm. But the Fat Greek has their free game going, so get there early. Hey Joe, what are the Kansas City Chiefs and a Chick fil a manager having common? What neither one shows up for work on Sunday? Gotcha that Jogins is bat team but serious the people. The food is super at the Fat Greek thirty three to sixty five you Paga Boulevard or you can go to go Fat Greek dot com for more information. Nick Chris and their whole family say
they'll be waiting at the Big Game party at the Fat Greek. It's the big game. Bring your big appetite and have big fun. It'll be a big easy to plan for your game celebration at the Fat Greek in Ukaipa, Go Fat Greek. Oh bah, it's that time of year again. No, not the holidays. Medicare open enrollment and if you have questions about Medicare, you should talk to the local experts. Paul Barrett and associates. All of his agents are certified with plans that are accepted by most of the medical
groups in our area. Call nine oh nine seven nine three oh three eight five. Their services free and after forty two years of the business, their agents are trained to help you pick the plan that's right for you. Last July, several GOP senators combine their fi WoT intellects to charge that inflation was rising because of quote insane tax and spending spree of President Biden and the Democrats, never mind that the insane spending is for such sensible, productive, and
enormously popular national needs as childcare and childless benefits. Mitch McConnell's rapidly partisan flock saw the chance to politicize the public's legitimate worries about rising prices. You, poor consumers are made to pay more for basics they squawked because of socialist Joe's investment, and grassroots people follow the ricocheting pinball of the GOP's logic. One they say that helping hard hit families induces them to refuse to go to work.
Two, this creates blockages in the global supply chain. Three this causes shortages of everything. Four this forces corporate bosses to raise all prices, which five slams the middle class and poor. So six lazy workers cause inflation. Whooh, Rube Goldberg couldn't have dreamed up a more fantastical diagram to deflect attention from what's really happening, namely that instead of an inflation problem, we have
a corporate greed problem. Of course, the greed meisters insist that their pursuit of excess corporate profits has not driven any price surges in our economy of free market competition, they snap, Prices are established by the law of supply and demand. It's the magic of the market place, they explain. But magicians don't do magic, They perform illusions, and the illusion of free market competition implodes when it hits the reality that our economy doesn't remotely resemble a competitive market
place. This is Jim Heiitar saying, for some forty years, corporate directed government policies have been transforming America into monopoly nation, letting the few gouge the manuye. That's where inflation comes from. Lauren sp century twenty one. Masters reminds listeners that a family pet is a year round commitment. Proper care includes regular checkups and vaccinations, dental care, space, and neuters, along with
flea, tick and heartworm preventions. This reminder is brought to you by laurenesb. Century twenty one Masters, reminding everybody pets are people too. Lauren from century twenty one Masters goes to extra mile to keep all her clients, having to put her fifteen years of experience to work for you and call her today at nine by one, six two three, six y nine five five. That's nine by one six, two, three, sixty nine fifty five.
That's Lauren esb century twenty one Masters on the Air Because they Care. Redlands Ranch Market is a unique, full service international grocery store that specializes in authentic food items from Mexico, India, and from many Mediterranean and Asian countries, including popular items from the US. They offer fresh baked items from their in house bakery, housemade tortillas from their Tortillirea, a delicious array of prepared Mexican
foods, a terrific fresh food and juice bar, and a large selection of meats, seafoods and deli sandwiches, salads, and hellal meats. Their produce department is stock full with fresh, local and hard to find international fruits and vegetables that you cannot find anywhere else. Don't forget to step into the massive beer cave and experience the largest selection of domestic, artisan and imported beers in the ie. They can also cater your next event with one of the delicious
takeout catering trades of food. Visit them at Redlands ranch Market dot com. That's Redlands ranch Market dot com. Redlands Ranch Market a unique and fun shopping destination. KCAA Radio has openings for one hour talk shows. If you want to host a radio show, now is the time. Make KCA your flangship station. Our rates are affordable and our services are second to none. We broadcast to a population of five million people plus. We stream and podcast on
all major online audio and video systems. If you've been thinking about broadcasting a weekly radio program on real radio plus the Internet, contact our CEO at two eight one five nine nine ninety eight hundred. Two eight one five nine nine ninety eight hundred. You can skype your show from your home to our Redlands, California studio, where our live producers and engineers are ready to work with you personally. A radio program on KCIA is the perfect work from home advocation
in these stressful times. Just type KCA radio dot com into your browser to learn more about hosting a show on the best station in the nation, or call our CEO for details too. Eight one five nine nine ninety eight hundred NBC News Radio, I'm Chris Graggio. Two people are injured and the shooter is dead following a shooting at Jolo Stein's Lakewood Church in Houston. The woman entered the church this afternoon, wearing a trench coat and backpack, armed with
what authorities called a long rifle, and then began shooting. Two off duty officers engaged the suspects, striking her. She reportedly entered with a five year old child who was hit during the incident and is in critical condition. Another man was hitting the leg and is receiving treatment at a local hospital. It's currently unknown what the motive was or how the child was related to the shooter. Memphis police say a suspect is in custody after multiple shootings across the city
that left one person dead and multiple people injured. Authorities say the shootings took place across three different locations in the city, with a carjacking being connected to all three crime scenes. The suspect was caught, with authorities saying he was out on one hundred thousand dollars bond for attempt at first degree murder and attempted aggravated robbery when he committed the crimes. At least four people were reportedly shot.
A former GOP presidential candidate says if the twenty twenty four election comes down to President Biden versus former President Donald trum Up, then the American people are left with two really, really bad choices. Both of these candidates are people who are being questioned for their competence and being questioned for their character, and
that's a problem for the American people. Speaking on NBC's Meet the Press, Chris Christie said he knew the GOP race was over in the first debate when six of the eight candidates on the stage said they would support Trump even if
he's a convicted felon. The former New Jersey governor and US attorney talked about the Justice Department report that said there will not be any charges against Biden related to classified documents, saying comments about the president's memory were appropriate because they explained why he wasn't being charged. When asked, Christy said he has not looked into running as a third party candidate and hasn't had any talks with the No
Label's movement. I'm Chris Garagio, NBC News Radio, NBC News on CACAA Lomalinde sponsored by Teamsters Local nineteen thirty two. Protecting the Future Share of Working Families Teamsters nineteen thirty two, dot org