EP: 481 The case for artificial useful intelligence (AUI) over AGI
35:29 Share

This is the Everyday AI Show, the everyday podcast where we simplify AI and bring its power to your fingertips. Listen daily for practical advice to boost your career, business, and everyday life. Sometimes when working with artificial intelligence, it can feel like you're dealing with alphabet soup, right? Yeah, we're leveraging AI and LLMs, large language models for this

Gen AI, right? And we're all chasing AGI, artificial general intelligence, but is all of that useful, right? Well, today we're going to be talking about a type of AI that is probably very useful, and that's the case for artificial useful intelligence and how that's probably more important than what we're all seemingly talking about and focusing on, which is artificial general intelligence. So, let's get started.

All right. I'm excited for today's conversation. I hope you are too. If this is your first time here, welcome. Thank you for listening. My name is Jordan Wilson and this is Everyday AI. So this is your daily live stream podcast and free daily newsletter, helping us all not just keep up

with generative AI, but how we can use all of this knowledge to grow our companies and to grow our careers. So if that sounds like what you're doing and what you're trying to do, if you're trying to be the smartest person in AI in your department, it starts here, but then it actually continues at our website. That's youreverydayai.com. So in our free daily newsletter, we're going to be giving you all the AI news and tips and tricks and everything like that to keep you up. But we're also going to be recapping today's conversation with an amazing guest.

So if you didn't catch everything, maybe you're in the car or walking your dog, don't worry. We're going to have all the takeaways and insights in our free daily newsletter. Also, this is pre-recorded. So if you're dropping in for the AI news, that's going to be in the newsletter as well.

All right. Enough chit chat. I'm excited for today's show and an amazing guest that, you know, a company in his work, I think we're all going to know and probably relate to. And I think, you know, even myself, I do get caught up in this alphabet soup, right? Of all these different types of AI and AGI and ASI. And well, let's just make it useful. All right. So please help me welcome to the show, uh,

We have joining us Rushir Puri, who is the chief scientist in IBM Research and IBM Fellow. Rushir, thank you so much for joining the Everyday AI Show. Hey, thank you, Jordan. And thanks to all your audience for listening in as well.

All right. Hey, I'm excited. Before we get into this topic of artificial useful intelligence, can you tell everyone, which might be hard to do it quickly, right? But can you quickly tell everyone a little bit about your background and what you do at IBM? Just to kind of set the stage here a little bit. Just I think just from my background point of view, I have focused for last almost four decades on

approaching four decades on technologies that relate to automation. At almost every level of abstraction of technology, you can think about from potentially two of the most important technologies in the world today, specifically AI and semiconductors.

Both of them have focused pretty much half of my career on semiconductors and its automation. And the second half last two decades on artificial intelligence. My background is in optimization and algorithms and, you know, have been doing computer science for very, very long time.

Yeah, and I'm excited to get in your background and talk about some of the advancements that you and your colleagues at IBM have made over the past couple of years and decades. But before we get into it, let's start at the top. What the heck is artificial useful intelligence? I think you started the intro very nicely, Jordan, and I think I'm just going to pick up from there. There's just a lot of talk about

AGI and the scaremongering that goes together with it regarding, you know, there'll be kind of robots walking around and we need to start becoming scared and life is going to be really pretty troublesome. First of all, there is not even a clear definition of general intelligence. And let me just say, every time we thought we have achieved it,

We kind of pushed that can down. Like, I don't think so. That's general intelligence. So let me start with something that we at IBM focused on for long. And that was the pinnacle of intelligence at that time, which is playing chess. And there is no better kind of, you know, I would say milestone than to defeat the reigning world champion of all times, Garry Kasparov.

Well, we built this machine called Deep Blue and we had a famous match and we defeated Garry Kasparov. The machine defeated Garry Kasparov. And, you know, we thought, hey, that should be it. Right. We should have surpassed.

you know general intelligence super intelligence whatever you call it like well i didn't felt very good but didn't feel like no that that was like humankind actually okay well then fast forward another decade um we actually you know very famously uh played on live tv which is kind of very uh hard uh this game called jeopardy and had a machine called watson uh play that game and

The game is well known now, and we defeated the reigning champions at that time. And we thought we had achieved general intelligence. Not really yet. And then the technology continues, and to me,

What is general intelligence? I don't even think so. In some words, like Sam Altman and others have formulated this regarding when AI can achieve $100 billion of revenue. That's kind of narrowing it too much. I really think so. That's a business lens too. Let's keep that aside. Intelligence, in my mind, is related to not even just pure what is known as IQ, actually, which is intelligence quotient.

It's also related to a large extent to what is known as emotional quotient. I'll say EQ. It's also related a lot extent to there's a third Q I coin, which is relationship quotient called RQ. So in my mind, intelligence is comprised of IQ, EQ and RQ.

And we tend to focus too much on this sort of very specific IQ part of it. And when we say somebody is intelligent, a human is very intelligent. It's combination of those three factors sort of generally speaking.

And leaving all of that aside, I think what we should really focus on is for your day-to-day listeners and for your really people who are decision makers to focus on, is this technology useful? That's all that counts. I really don't care whether we have achieved AGI, ASI or whatever it might be, AXI. I want this thing to be useful.

And usefulness can vary in your perspective. And we can talk a lot about sort of what does usefulness actually mean. But to me, it's about

a technology it's about automation and is this helpful in your day-to-day life in the whether you are an enterprise whether you are a business whether you are a small business owner is this being helpful in helping you accomplish things in day-to-day life and is the technology fulfilling that's all that matters actually

And that's why I'm a huge fan of AUI rather than AGI, ASI, AXI, whatever we may call it, actually, because that definition is not even clear what that means. And if you go back 10 years or even 20 years of history, the definition keeps on evolving all the time, actually.

Yeah. Yeah. Non-stop. Right. And I've talked about that on the show a couple of times, just looking at the definition of AGI from 15, 20 years ago, it's like, Oh, technically it's already achieved. Right. Like if you look at definitions from 20 years ago, but you know, I'm curious and I love how you brought this out and kind of this, this three pillars of intelligence, the IQ, EQ, and then the, you know, relational, right. RQ, you know,

One thing I'm curious about, even as it pertains to AGI, so that's, oh, when one AI system has the ability to understand or learn anything that humans can and perform tasks. But what about it seems, and maybe I'm wrong here, it seems like the tasks that us humans are performing are changing now more than ever.

Right. Like, you know, yeah, I've only been working for, I don't know, 25 years or something like that. But it seems like like what us humans are expected to do in the last two or three years since large language models has changed very quickly. Do you think that that may be just the ever evolving concept of human work and what we're doing with AI? Is that also kind of changing this fluid definition of AGI? Yeah.

I'm glad you brought it up, Jordan, because if you look at the evolution of sort of just we as humans, there was a time, so go back to industrial revolution. We went through automation technologies, which automated something that was very prized at that time, which was people really building or manufacturing with their hands to machines, building those things.

there was a profound actually revolution and gave rise to productivity and consumption which was literally unprecedented at that time what was priced in the new era post-industrial revolution what was priced knowledge work like if you could create knowledge if you could analyze knowledge if you could read write all of that that was prized more and more and in every revolution in human history

you should just go back literally almost thousands of years and just study that history if one would really conclude that knowledge workers have always been priced more and more and more it was something that was thought to be almost unachievable if i if i may say by machines um sort of this deep analysis of knowledge creation of new knowledge as well i'm gonna keep creation of new knowledge aside just for the time being because we should discuss it but

This is the first time I would argue in human evolution that we are very close to generating language seamlessly, actually, whether it is spoken language, whether it is analyzing language and a language of all kind, whether language happens to be a computer language code, whether that happens to be a spoken language, we are able to analyze, understand, reproduce,

generate things seamlessly. Now that has profound implications on how society does work. Very similar to the machines that automated, you know, really manufacturing had profound impact on how we did work. But we didn't stop work, by the way. Those became tools through which we did more work actually. So I'll give simplest of the examples. When somebody was banging nails with a stone, somebody invented a hammer.

Did we stop banging nails?

We really banged more nails. In fact, we invented machines that can just put more nails in, actually automated machines that put more nails in. And we discovered new sort of usage of those nails. Actually, if I may say, we build auto new houses, like manufacturing grew and so on. I fully expect this revolution to be very similar in perspective. These will become amazing tools through which people will

unleash new productivity. I'll give you a very simple example. Actually, I happen to be on the campus of MIT this weekend, be with some students actually. And I was talking to them and says, we used to wait for reading hours, what is known as sort of really reading hours with professors and TAs and so on. And now I can literally take, if I don't understand a question,

I screenshot, I put that in your large language model of choice, whatever chat GPT, cloud. And I say, explain this to me. And it does it. I can really talk to it and it does an amazing job overall. And I can get to that question and understanding much faster than I would otherwise.

Does that mean students have stopped understanding? Not at all. I think they've discovered new ways through which they can understand a whole lot better and debate with the system in a very seamless way. They were waiting for their turn in the line to talk to someone.

Now they can talk to someone as well as they can really debate that internally and be much better prepared for that conversation. I think that's a good example of how these technologies are impacting day-to-day life and are being useful to these students and other, you know, really people in terms of helping them

made their life more fulfilling. So I know one recent kind of advancement, at least when it comes to large language models, is their ability to reason, right? So I know IBM's new Granite 3.2 has a reasoning mode, but essentially now all of these large language models, when it comes to intelligence benchmarks, right,

they're off the charts of where we might have expected them to be four or five years ago when we were looking at early GPT models. So that leads me to think, ever since you broke this intelligence down into three categories, now I'm just scratching, is it even useful for us humans to have

All of intelligence, right? The IQ side. Is it useful for us to have that if we can't actually put it into use on the emotional side or the relational side, right? I'm like, sometimes I have all these models that can do anything and everything. And I'm just like, huh.

Right. Like I like I just get stumped sometimes at the amount of intelligence that I have at my fingertips. So what is actually useful in this in this age of, you know, these large language models on the IQ side are so high. Are you still running in circles trying to figure out how to actually grow your business with AI? Maybe your company has been tinkering with large language models for a year or more, but can't really get traction to find ROI on Gen AI. Hey, this is Jordan Wilson, host of this very podcast.

Companies like Adobe, Microsoft, and NVIDIA have partnered with us because they trust our expertise in educating the masses around generative AI to get ahead. And some of the most innovative companies in the country hire us to help with their AI strategy and to train hundreds of their employees on how to use Gen AI. So whether you're looking for chat GPT training for thousands,

or just need help building your front-end AI strategy, you can partner with us too, just like some of the biggest companies in the world do. Go to youreverydayai.com slash partner to get in contact with our team, or you can just click on the partner section of our website. We'll help you stop running in those AI circles and help get your team ahead and build a straight path to ROI on Gen AI. I still definitely think so. I think you should really look at

I think what you're bringing up is really a profound point on we are in the middle of defining the future of work, by the way. I really think so. You are at a profound point in history where, like industrial revolution, once again at a juncture where you're defining the future of work. And I think it's been said before also, but I would capture it again here as well, that

with reasoning models, these amazing reasoning models and even more importantly, these reasoning models with tools at their disposal. By tools I mean

Tools that we enterprises and businesses use every day, a database, you know, really the tools for business processes and so on, which are digital tools that you can call at any time, at any instance and say, accomplish this subtask. Give me back the results. I'm going to integrate that back in and continue in my sort of business process.

It becomes even more important for us to be able to learn to manage these tools, if I may say so. People have said it, as I said earlier as well, that if you look at the information technology department of the future, currently it's comprised of lots of people. I would say.

Information technology department of the future is people who are operating these tools, managing these tools, governing these tools, but will become a lot more automated in the future. So it's like people managing these agents, if I may say, these agents happen to be digital agents.

Having said that, the value has to be very clear in terms of the business value being delivered. That value is still managed by the people and the relationships among those people as well. We humans are never going to prize less in any way, either the emotional part

or the relationship part. I think those two parts grow even more in importance if I may say because now so far it is about people managing people. Then it will be about people, still people by the way, managing these agents that happen to be digital agents because somebody needs to make sure they are doing the right thing because somebody needs to have accountability by the way. It all comes down to accountability. Who has the accountability?

Finally, where does the no in sort of general English, the buck stops where?

And if the bug stops at human, then I better know what is going on. Actually, somebody should tell me what is going on. If the bug doesn't stop at me, do whatever. It's fine. But if the bug stops at me, then you should better like I should have a control over it. I should govern it. I should know how to operate it. And I should be able to manage it as well. Actually, I think that is the most important part. Where does the bug stop?

I believe the buck will continue to stop at humans for a very long time. Yeah, that's a great point, right? Yeah, like even myself, as these models become more and more powerful, like you feel almost like,

Not pressured, but all of a sudden you're just taking a back seat and just kind of marveling at what AI can accomplish. But it's like, no, the human role and the human of the loop becomes more and more important as we talk about agentic AI, reasoning AI, all those things. But I have to ask you this. So as we make the case for artificial useful intelligence over AGI, what's

What are you currently finding useful, right? How are you measuring AUI, whether it's in your own work, in your team's work, you know, at IBM, how are you actually measuring it? And how can you define what's actually useful when using artificial intelligence? So I'll take sort of, I think there are a couple of forces that have literally revolutionized our lives in last, I would say a couple of decades.

I think it was Mark Anderson of Anderson Horowitz fame who said, maybe in 2011, if I remember right, software is eating the world. And I think it'll be fair to say software has eaten the world. I think we are in the middle of an era which is defined by software to a large extent. And the development of software,

The testing of software has become a major endeavor and we've got literally millions and millions, tens of millions of software developers. And one thing that we are very focused on from an IBM perspective is how do we make the lives of enterprise developers, business developers much easier? I think I'm going to sort of lay out a use case in which how we measure this. So.

Given that software is so important to the world and given that software developers are like it has been a prized commodity. The daily life of a software developer is like this, actually. It's almost like a doctor, although I don't want to compare saving lives to fixing software bugs. But it's almost like you come in and you look at your your you open your tab and say, well, Richard, you've got 40 issues to fix today. OK, well, that's kind of going to be a busy day.

and you start going through the list and you are fixing issues one after another. You are testing it, you are patching it, you are releasing it. And you get to the end of the day, you are about to sign off and, you know, five more show up. This is urgent, needs to be fixed. Okay, you know, you want to go home. And you wish at that time there was a technology available for you to help automatically.

This is the technology that we are launching, that we are working on today, or something called software engineering agent, which is able to look at your complex software development landscape, hundreds of files, thousands of lines of code, hundreds and thousands of lines of code, just English description of an issue that you have. Pinpoint the issue for me, where it is,

Tell me what is the why this issue is there. Second one, suggest a fix and tell me the reasoning for that fix and go fix it. So if you had 40 issues lined up on your plate or in your list, I fixed, assume for the time being 15 for you automatically.

That's real productivity in your day-to-day life that you can measure. This is not about how many calls you made to a large language model. I don't care. What I care about is the end value that you deliver to me. The end value is the time consumed in my day-to-day life, which is going into things that may not be productive. By productive, I don't mean fulfilling. I wanted to go home at that time. I wish there was a technology available.

There is a technology that's available, actually. Again, the second part of this could be, you know, fixing vulnerabilities automatically. I mean, the world is full of, at this point in time, cybercrime. Identifying when the cybercrime is going to happen, where it's going to happen, if the software has leaked, on a dynamic basis all the time. You know, what is known as AI for security, actually.

is again in a world that is full of risks, if I may say. For every business and every entity in the world, that becomes an extremely useful scenario that there are not enough human hands in the world with the right expertise to be able to identify. You can only minimize the risk. If there's a technology that's available to help you reduce the risk even further, God bless it, actually.

So it's not taking away from any human activity. It's just making your risk lower and your life much better. So those are some very tangible way in which we are measuring things that are impacting day to day activity of sort of normal human endeavors.

Yeah, I think the day-to-day activity of what, you know, knowledge workers normally do, it's right, like that's where, you know, AI, I think, is truly useful, right? When you can get time back, when you can get focus back, when you can get creativity back, right? All of those things. But, you know,

I want to hit rewind here real quick, you know, cause we briefly mentioned, you know, some of your, your background and, you know, IBM's achievements in the field of artificial intelligence, right. It's been around for many, many decades and it's been useful as well for many, many decades. Right. So, uh, maybe the timing here, it's, it's, it's kind of, uh, interesting, right. Because you had a deep blue, I believe that was, you know, around 97. Uh, then you had, you know, becoming the chess champion and then you had, uh,

you know, Watson on Jeopardy about 14 years later, you know, both of those two very useful, right. In terms of, of where the artificial intelligence, you know, is, is at and where it's going. And here we are now 14 years later. So 14 years between each one. So, you know, what's that next big kind of landmark, right? So first it was, you know,

beating the smartest chess player, then it's winning Jeopardy. What's that next big milestone with everything that we have in AI right now? You know, what's that next big thing that you're like, oh, okay, this just opens up a whole nother echelon of AI being useful. I think that word has been overused, abused, but I'll really clarify why I'm so excited about that technology because that's a profound shift in technology, right?

which is what I'll say agents and I'll describe what I mean because it's been really lot is written on it kind of abused in many ways so far we've been working with systems that in engineering terms is called feed forward systems feed forward systems are you give that system a input it gives you output

If you don't like the output, you as human don't like the output, what do you do? You as human change the input. That's called prompt engineering. You don't like the output of a chat GPT, you change the prompt. You change the prompt. Agents take it at a whole different level actually, like exponentially smarter. They say, you give me input, I'm going to give you output. I'm going to analyze that output. I, machine is going to analyze that output.

Compare it to your intent of the input and continue to iterate internally until I get it right. That actually entails, we were talking about reasoning earlier, deep reasoning of these AI systems. In particular, another step they take, this is not just the only step they take, another step they take is, you know,

So it's okay now actually in ChatGPT, but go back around a year back and there were many Twitter posts on it as well that if you give ChatGPT two numbers to add, just give it two kind of random numbers to add, a little bit larger.

it's likely to get the addition wrong because adding two numbers is not a, what in no LLM terms or large language model terms is known as next token prediction problem. I think, can somebody tell Chad GPD to please use a tool called calculator that we have used for, you know, thousands of years, not just decades, thousands of years. Please use it. Please don't use it as a language problem. This is not a language problem. This is a math problem.

can somebody use a calculator okay i think then realizing when to call that tool you gave me an english problem i said oh that's an addition problem i should call a calculator and then you call a calculator look at the result plug the result back in in that english words and say your addition is x and you continue actually from there so this ability to be able to take a task break it down into sub tasks

call the right set of tools for the right sub tasks, integrate all of that together, reflect on the results and continue until I accomplish that task is what in the next level of technology is called agents. And it takes the word from what is known as feed forward systems to feed back systems.

All intelligent systems in the world are feedback systems. I'll give you simplest of example where it will make very, very clear. Assume for the time being you are trying to send a rocket to the moon. If you were 0.0000001 degrees off launching from the earth, you ain't going to the moon. You're going somewhere else. Yeah, you're messing. I don't know where you're going, but you're not going to the moon actually. So the whole point is for rocket launching systems is not just to have the strongest possible rocket.

Think of that as a very strong model. But to have the ability to be able to correct every time you realize, oh, I'm not going there. Let me correct. I'm not going there. Let me correct. This ability to be able to correct is so important in intelligent systems that it can mean the difference between

Landing, no, not going anywhere to really going where you want to go. I think that's a good analogy to what agents are in this new era to what technology was before, actually. And that's why I'm so amazingly excited about sort of agents and what they can do to the world.

So speaking of that, right, and tying it back to this, you know, artificial useful intelligence, when we talk about agentic AI, you know, AI that can reason, you know, agentic AI that can reason that has tool access and knows when to use the right tools, right? It seems like we're on the precipice of all these things coming together. So speaking of useful,

what's useful right now for business leaders to be focusing their own time on aside from, you know, using the right system, right. And actually, you know, AI strategy and AI implementation, but what about on their own EQ in our Q side, right? What are those useful human skills that we need to be learning and practicing to properly take advantage of

of AI's nonstop development that's happening seemingly on its own, right? What do we need to be preparing for to actually make good use out of all of this AI? I think I encourage every business owner, every business decision maker, every business strategist, number one, to get involved.

And it's almost like by default should be true, but get educated beyond the hype. Even start like I encourage everyone to be hands on. You don't need to build the software. You don't need to be a software developer, but please play with the technology yourself. The technology is available at your fingertips. Please play with the technology yourself and that will give you a notion of the power of the technology.

Second, I would suggest will be for everyone

have a strategy and a plan of how AI is going to disrupt and reconstruct your business. It's like mandatory actually. It's like, you know, everybody has a plan for if I lose electricity, I'm going to have a backup generator or something. It's almost like that. It's like electricity actually. Have a plan of how AI is going to disrupt and reconstruct your business. I don't mean just, that's why I didn't stop at the disruption part.

And the third one will be make sure your employees, your teams have the right set of skills. Because we are in the middle of a transition, as I said. It can be very scary for people who don't have the right skills. Because people can be very scared of technology if they don't know how to transition into the new world, actually.

So those three factors together have opinions that are grounded in sort of hands-on activities.

Second one, really have a plan at a business level, at a decision maker level. And the third one, bring your teams together, which is the part on the EQ and the RQ. If you leave it, then it becomes cultural. Actually, they're going to resist it. You can shove it down their throat. It's just, you know how it goes, actually, in general. It's not the best fulfilling activity a decision maker wants to have. Right. Yeah.

That was such a good way to end today's show with that, you know, on the fly, by the way, my gosh, right? We didn't, we didn't talk that one out beforehand, but I love that. Just the education, the strategy, you know, to not just, you know, deal with the disruption, but the reconstruction is huge. And then making sure employees have the right set of skills.

Amazing. Today's conversation was a fantastic one. So thank you so much, Rushir, for joining the Everyday AI Show. We very much appreciate your time and your insights. Thank you, Jordan. It was a pleasure.

All right, y'all. That was a lot. I'm excited. I'm going to go listen to this show and I'm going to type up this newsletter. This is one that I think you need to re-listen to at least twice because Rashear dropped a lot of great information on our heads live, unscripted. Love to see it. So thank you for tuning in. If this was helpful, please go to youreverydayai.com. Sign up for that free daily newsletter. Thank you for tuning in. We'll see you back tomorrow and every day for more Everyday AI. Thanks, y'all.

And that's a wrap for today's edition of Everyday AI. Thanks for joining us. If you enjoyed this episode, please subscribe and leave us a rating. It helps keep us going. For a little more AI magic, visit youreverydayai.com and sign up to our daily newsletter so you don't get left behind. Go break some barriers and we'll see you next time.