8. History's Guide To The Future of AI
>> Hello, Scott Hebner here. Thank you for tuning into this episode of the Next Frontiers of AI Podcast, where we partner with industry pioneers to explore the latest developments in AI and how businesses are applying these advancements to create competitive advantage. With innovation cycles moving at warp speed, staying up to date is more important than ever. Today I'm thrilled to be joined by Irving Wladawsky-Berger. Irving is an MIT research affiliate, adjunct professor, senior fellow, advisory board member, and a legendary game-changing executive from IBM who I like to refer to as the father of E-business. During my days at IBM, I learned a ton from Irving's inherent skill to seamlessly fuse technology innovation with business strategy.
Today we'll explore how the lessons from the last 50 years of technology transformations may provide a guide into the future of AI. Irving will share his unique insights gained from experiencing multiple eras of technology- driven transformation, including compute, mainframes and PCs, interconnection. First with client-server, and then with the open internet. And instrumentation with the internet of things. And now intelligence, advanced analytics, and of course AI. For those of us that have been through these progressive waves of transformation, it begs a simple question.
With AI, are we watching the same movie again, just with different characters? What can we learn from the lessons of history? So with that, let me bring up Irving. >> Hello, Scott. It's a pleasure being with you here. >> Hey, thank you very much.
I really appreciate you taking the time. And the weather's starting to get a little bit warmer. We'll have to take our walk again down in Westport.
>> I look forward to that. >> Yeah. What have you been doing over the last many months since I last saw you? >> You know, same things. Going for walks, although it's a little harder when it's so cold, and it's been very windy. Reading a lot about AI and writing about it in my blog, trying to wrap my head around agentic AI, which is sort of the hottest new thing.
Very involved now in helping to organize the MIT CIO Symposium, the annual MIT CIO Symposium that will take place in May. So pretty much things I've been doing year in, year out for the last several years. >> Well, I'm so glad that you found some time to be here , because we have a lot to talk about. Things have come a very long way in a relatively short amount of time. What used to be the big wow is now just duh. Right?
It's what people expect. My kids, I don't think appreciate what it was like, let's say 25 years ago when the internet was just coming on the scene here. >> Well, and we had no smartphones. >> We had no smartphones, right.
>> Remember when you traveled and you got to the airport, and you had to call the car to come get you. How did we do that if we didn't have a smartphone? It's hard to imagine. >> Yeah, and I think back to when I was in college, you're right.
The phone was wired to the wall. Actually, my dad had got me one of the first PCs, IBM PCs, and I remember it was like carrying luggage. I used it for word processing, I think when I was a junior or something like that. But even if you just go back 2025 in the internet age, let me show you a video that I just happened to see over the weekend that I think helps to bring this to light here.
Let me bring that up right now. >> The X10s are online. Gentlemen, I am now about to send a signal from this laptop through our local ISP, racing down fiber optic cable at the speed of light to San Francisco, bouncing off a satellite in geosynchronous orbit to Lisbon, Portugal, where the data packets will be handed off to submerge transatlantic cables terminating in Halifax, Nova Scotia, and transferred across the continent via microwave relays back to our ISP and the extended receiver attached to this lamp. Look at me. Look at me. I've got goosebumps. Are we ready on the stereo? Go for stereo. >> That's very cute.
And actually, there was a conference in 1995, and I was working with some people at Argonne National Lab and so on. And in the conference, we were trying to do experiments about being in touch with people in San Francisco where the conference was, and in Chicago over the internet. Now, that seemed to have been a very big deal then because we were trying to do that over cables as opposed to over phone communication. So that we take for granted now. Other things we shouldn't take for granted yet, but that we do >> Right.
Now when you're on vacation somewhere and you want to turn your lights on at home in the early evenings, it looks like you're home. You can do it. And it is quite amazing. Actually, I looked up, that episode ran on March 19th, 2005, and I think the internet stuff started 10 years earlier than that. >> Yeah, for sure. >> And we're going to get into that for sure.
But let's kind of start with you for a little bit of time here. I mean, you retired from IBM, I believe, in 2011. >> 2011. I retired from my full time job at IBM in June of 2007, after 37 years. And then I stayed another three and a half years as a consultant, and I finally fully retired in December of 2011.
>> Yeah. And ever since then, I know you've been keeping super busy. You went through some of it a little bit early. But tell us a little bit more about the MIT affiliate research role that you're doing. >> Yeah. Okay. So I've been affiliated
with MIT since 2005. It's one of the things I have most enjoyed. I've taught some master level courses, not a lot because I don't live in the Boston area.
I live in Connecticut, so it's a bit difficult to go every week to give a lesson. But I am a research affiliate at the Sloan School of Management, and I'm a particularly a member of two major groups in the Sloan School of Management. The Initiative on the Digital Economy, which as the name says, it's about looking at research on lots and lots of things to be done as we move into an increasingly digital economy.
I became a member of another research group on cybersecurity, also associated with the Sloan School, and I asked to join that a couple of years ago when, I mean, it was clear that cybersecurity was becoming a much, much more serious issue and cyber attacks. So I've learned a lot from attending our weekly meetings. A lot of it scares the hell out of me I have to tell you the kinds of things businesses have to face. And of course, all the phishing things that individuals are facing.
And I am very, very involved as an organizer of the annual MIT CIO Symposium, which takes place every year. In May this year will take place, May 19th and 20th. And I've been doing that now for at least, I don't know, 10, 12 years.
And the last three years I've led, I've been the moderator of a panel. I like to moderate a panel with two young up and coming MIT graduate students or postdocs. And this year, for example, I'm doing a panel on the impact of AI on jobs and skills.
I'm looking forward to that. And a number of activities of that nature. I'm a mentor in one of the classes in the Sloan School called the Analytics Lab, where students are given a problem that a company actually has. Companies contribute a problem they want the students to analyze and they contribute data to help them analyze the problem. And then throughout the semester, that's the key project that groups of students are doing, and a mentor is assigned to each group, and I work with the students.
So it's a combination. Working with students is the nicest thing. >> Yeah. I remember when we were taking a walk with my son Jackson, you were talking about- >> That's why I was so excited that he was about to go to Oxford.
>> Yeah. And it wasn't lost on me the conversation about you having to mentor them on business impact and business context. And it's not just about the technology, and it sounds like a really rewarding thing to do, to work. >> It is rewarding. And let's say in these groups, most of the time they have to be admitted to the class. And for the most part, their analytics skills are very strong.
They've all learned Python. They know the latest analytics kinds of things. Stuff, by the way, that they know that far better than me because I got my PhD in the 1960s in physics doing eigenvalue problems and partial differential equations. So my math is a little bit behind the new math.
But the thing where I could be most helpful is it's not enough to do the technology. You have to then explain the implications of your analysis to a group of people who may not know your technology. And actually, they don't care. What they really care about is, what does that mean to the business? Why should I do differently based on the technologies you're showing me? And that's where I can be most helpful to the students. >> Yeah. I remember when we were taking the walk, it was like
you and Jackson were talking Chinese or something like that with all the math stuff. I kind of followed it a little bit. But anyhow, having spent so much time in this industry and in particularly all those years at IBM, you've been through a lot of technology transformations. >> Oh God, yes. - Yeah. Have reshaped the business landscape. Of all those, what do you think had the biggest impact on change in business, the business landscape? >> What's so interesting about my career is that the different technologies I've been involved have been quite interrelated in the sense that one led to the next, to the next.
Now, I was lucky enough that the summer before I started college at the University of Chicago, I was able to get a job in the brand new computation center that was being organized by a professor of physics and chemistry, one of the top computational physicists in the world. And I was 17, I was about to start college, and he offered me the job. And I said, "What do I do now? " And he gave me some manuals and said, "Well, go read them and see how you're learning. " So I got started programming in assembly language back in 1962 and debugging programs by taking assembly language dumps and going through them. Did you ever do that, Scott? You're a little younger than me, so you probably didn't quite have to get to that level. >> Not that level. I programmed in Lisp.
>> Yeah, that came later. Later on, I switched to programming in Fortran because you know, you could do a lot more. When I joined IBM research, I started doing programming in PL/I and so on. And I would say among some of the projects that I was very involved.
And actually later on I switched from being in the research divisions to more being in the product divisions. And for example, I organized IBM's efforts in parallel supercomputing. And that was very, very enjoyable. It was just starting, parallel supercomputing in the early '90s.
And that parallel supercomputing continues to be incredibly important first as internet servers. But in AI, all of the data analysis, and all of the training of algorithms, and on and on requires gigantically powerful parallel supercomputers. So I was involved in that.
Then you and I worked together in the internet division when IBM made the decision to start a new internet division and embrace the internet. And then the best decision we made is to focus our internet strategy in what we call e-business, which meant leverage the internet for business value. And the reason I think it was such a good strategy, again, there was a lot of new technology around the internet. Of course TCP/IP protocols, HTML, HTTP, etc., the new email protocol, new transaction kinds of algorithms to make sure that when you were paying with a credit card, it was all done very safely.
E-commerce wouldn't have taken off if people didn't feel safe using their credit cards. But what we did best is that for the most part, the customer... Remember, IBM customers were business customers. They assume we knew the technology was going to work. We had already embraced the standards of the internet.
They wanted to know, what is this good for? What should I do with this? And what we did was instead of picking the most complicated examples we could pick, I still remember one of the first examples we picked was to work with UPS so that if you wanted to know where your package was, instead of having to call a UPS operator with an 800 number, and then he or she would enter the ID of your package into a terminal connected to the mainframe where they had all the data, we worked with them to develop a browser application that then connected to a web server. And the web server then tricked the mainframe and say, "Here is a request from a terminal. " Obviously it wasn't talking to it. And the mainframe could care less. It got a request. It gave it back to the web server, which then translated it and gave it back to the browser. That sounds trivial, doesn't it? But it was so easy to understand.
Really, people loved it because now they could find out where their package was anytime they wanted to, and they didn't have to see if the operator was busy or not. A lesson there is especially in the early days of a new technology, don't pick the most complicated applications you can imagine. Pick the simplest that you can explain to the customer in a short elevator ride and explain how it works. And because it's simple, then you can get it done.
Now later on, we did more complicated things. We worked with L.L.Bean. L.L. Bean, I don't know if you remember in the '90s. They had a catalog business. They would mail your catalogs and you look at the catalogs. And once more, you had to call somebody on an 800 number and order from the catalog, and then they would have it shipped to you.
Well, we worked with them. Number one, create an online catalog. Created the online catalog.
So now you no longer have to work with the operator. And because it was online, you can see this stuff. You can see the pictures on the browser of what you were ordering, and you finished.
And then we integrated the credit card application so that they could pay for it. We made sure it was all done in the safest way possible at the time. And I don't need to tell you, e-commerce took off.
I mean, since taking off. So it was very nice to be able to be at the beginning. But let me repeat again. The key experimentation we did was in the marketplace working with customers, and jointly helping to develop the applications, and seeing which ones were very useful and which one...
Did we do applications that didn't bring so much value? We did. But the only way to find out what works and what doesn't is by working with customers in the marketplace. And it's as true of AI now as it was with the internet and in business then. >> Yeah, it was all novel back then, and I know- >> It was magical. It was more than novel. I think when you get a new technology, it feels like magic.
>> Yeah. Well, I know it played a... I'm going to show you another clip here, which I know you have not seen, but you may remember it. And this has to do with IBM's near-death experience and what the internet did and how IBM, under your leadership with Lou Gerstner turned it into the e-business strategy.
And then it was all history because, so let me show you this video. >> Company was going to make it, but now it was time to try to restore IBM's fame. We didn't want to be known, let's say you go to a party and somebody sees you and they say, "Where do you work? " I say, IBM. It wouldn't be nice if they say, "Oh my God, I thought you were dead. Thank God you're alive.
" You want to do a little bit better than that. And I think that God took pity on us because right around the time we were ready, they sent us this gift, the internet. The mid-'90s is when the internet was taking off. And I think the Gods were saying, "Here it is, guys.
Don't screw it up this time. " Yeah, IBM almost died in the early '90s during the transition from mainframes to client servers, because client server systems were based on pretty much PC and Unix technologies, which were much cheaper than mainframes. And in the early '90s, it was touch and go whether IBM would survive. Quite a number of companies didn't make it.
I don't know if you remember digital DEC, the mini computer company. And Wang, who was the leader in office systems and so on. And we were very lucky. Yeah, I used to use this slide when I talked about that experience.
Because in 1984, IBM was the most profitable company in the stock market. In 1984, all because of mainframes. And then only eight years later, we almost went out of business, which shows that if you don't pay attention to what's going on in the marketplace, you can go from the top company to one that's barely alive. And luckily for us, we got one of the top executives that I've ever worked with, I think in the world, Lou Gerstner. And with Lou Gerstner, we transformed the company, we embraced the internet.
And when he asked me, "Let's organize a new internet division," which we did. And that restored IBM. That stopped the death experience we were going through.
And it restored IBM to once more be a major technology leader then. >> Yeah. I'll never forget those days because I joined IBM in 1989. My dad worked- >> We were already sliding down then. >> Yeah, no, when I joined, we were starting to slide down. But it was okay. But by the time 1993, 1994,
that's when my dad retired. You start to worry about the company. And I remember losing billions. And like you said, and I think you've made the point many times ever since, is that there's nothing like a near-death experience to just shake things up. Right? And I'll never forget that journey, because we had the software group coming online.
We had services, and we had the overriding E- business strategy. And in my 34 years at IBM, I have never seen more of a one IBM. We win or lose together, do or die. And no one was confused about what IBM was about.
>> Yeah, I give Lou Gerstner huge credit, because it was his leadership that he made sure no one was confused. This is what we're going to do. Not everybody liked it.
For example, IBM had its own proprietary network architecture system, network architecture. Do you remember? >> Yeah. SAA, systems architecture, <inaudible> 6.2. >> Yeah, all that. And the people who were leading the networking division didn't want to change.
And they said, "Well, the internet is for supercomputing and for technical computing world," and we had to force it down their throat. I could never have done it if it hadn't been for Lou saying, "This is what we're going to do. " And we did it. And... >> I remember the shock waves when he went out, he was at a financial analyst session or something, and he's like, "The Internet's not about browsing.
It's about business. " And I just remember the negative press about it, and he turned out to be right. >> Absolutely. Absolutely. It's all about business, which brings us to AI now. >> Yeah, let's do that. - And the analogy is even more than the internet, when it first emerged in the commercial world.
Remember, it was being developed in the technical world, and research labs, and universities, and government labs since 1969. And you couldn't use the internet for commercial purposes until 20 years later, the late 1980s. And that's when it entered the business world. And right now, once more, we have AI, which in many ways is even more impressive than the internet. And you can go and have a conversation with ChatGPT about all kinds of things, and you will either text or write in English.
And with large language models, it will understand you. It can do real-life translation. You can ask it to translate a Shakespeare poem into Spanish, and it will do that in a second.
Or you can ask it to write you a poem to send to your son in Oxford saying how much you love him, and you hope he's happy in Oxford and so on. >> <inaudible>. - Hope he's having fun, and drinking good wines, and so on, which they do. As I understand, the different houses in Oxford have their own wine cellars, so they take that very seriously there. But what really counts is, what's the business value? What are you going to do with AI that brings you real business value? And I think for the most part, we are in the very early stages of that.
>> I agree. - I don't think we are anywhere near the equivalent of the e- business stage with the internet. There is one exception in my mind.
The exception is the internet at heart is all about analyzing gigantic amounts of information, gigantic amounts. A lot of the progress that we've seen with ChatGPT and the different AI models, it's all based on data. Lots and lots of data.
You use special GPUs from companies like NVIDIA to help you analyze the data, but it's all about analyzing gigantic amounts of data. And then either answering questions, translating from one language to another, etc. Now, where the internet already is very, very useful because of the data analysis is in the world of research. For example, a young graduate student at MIT, a second year graduate student in the economics department wrote a superb paper about the use of internet algorithms to develop new media products, that is new materials products.
And the way the people who did that were able to do that is they were inputting into the algorithms characteristics of certain materials and asking it to give you examples of similar materials, which then they can analyze and test. Similarly, people are applying this ability to analyze huge amounts of data to find cures for cancer, to be able to link characteristics, let's say, of a breast cancer tumor to the genomic characteristic of the woman who had the tumor. And then analyze a lot of databases to see what has worked in the past, what hasn't worked, find things that work that are very similar to this combination of tumors and genetic stuff. The reason that works so well, that's really not a business.
That's scientific research. That's what that is. And we've been doing this kind of supercomputing- based scientific research for decades, maybe longer. But especially using very powerful parallel supercomputers, the people who do this are researchers.
Physicists, biologists, etc. Remember when the human genome was decoded? I believe that happened in the 1990s or thereabouts. And so you have a highly skilled group of people who don't know how to do that. And then that research then has to be the recommendations that you get from that research.
Then they have to make a judgment. Which ones we should be patenting, which ones will lead to new products? And they go through all those processes. But the AI comes in in the research. That's very different from using it in a seemingly much simpler business application.
But that now requires, for example, to help a customer service rep understand if you call the company and you are confused about something. And if you talk to a person, people are very good at having a dialogue with you and clear up the confusion. If you talk to an AI agent, then the AI gets confused very quickly because it doesn't have the same human understanding of the problem that a human has.
That's the state of the art today. Will it get better? Absolutely. When will it get better? It's too early to tell. That's <inaudible>. >> Yeah. My view of all this, and again, trying to learn from the past is one thing I think I've learned in my career is what you think is important and critical in the beginning of the transformation turns out the end not to be.
So example, I remember in those days, the browser wars and Netscape. It was the most critical decision you could ever make was what browser you picked. And what ended up happening is it faded into the infrastructure. At best right now, it's a preference. It basically created a gateway into the internet. I think what you got today with AI is LLMs and generative AI, most important decision you can make right now, but it's going to fade into the infrastructure.
It's going to become just a preference. And it essentially is providing a gateway into the world of AI. And just like with the internet, with e-business, and I think what we're going to see today with AI is how you build around that foundational, LLM, generative AI capability within your business, your ecosystem of specialized models that get you into decision intelligence. >> By the way, what settled the browser wars, the reason we had browser wars is different companies had developed browsers, but the browsers couldn't access all the websites from each other.
And finally, the W3C got organized, which was a standard organization, and got everybody around the table, and they say, "We must have one goddamn browser based on standards that everybody's going to agree on. " And they all did to their credit. And that's what allowed the worldwide web to explode. By the way, in the same way, the reason the TCP/ IP replaced proprietary networks is because everybody agreed to now use the internet standard protocols for their networks.
So the old proprietary networks that couldn't talk to each other, that went away. So that's what takes time to try to get agreements. And then as you were just saying, because we don't know, because we're in the early stages, what will really work well in the marketplace and what will not.
What is very cool, but it won't work. It doesn't bring you business value. We need a lot of experimentation. I like to say that you have to play the game in the field, and the baseball season is about to start.
You have to play the game in the field. It's not a theoretical game. You have to get out there. And some of the applications will be very, very useful and valuable. And other applications that on paper look very good.
>> Well, I got something for you. Watch this. >> But you know, you're going to the Red Sox game tomorrow, right? >> I'm going to the Red Sox game. >> The game is played in the field, not in the dugout. So the labs are... No, I'm very serious.
>> That was a key message to me, right? And I think, and I remember Steve Mills used to always preach this too, which is to be successful when you're in the middle of a transformation, you have to spend at least half your time experimenting with what's coming next, because you have to make a decision. You're either going to lead, lag, or simply fail. And I think that's another lesson where don't get overly consumed with LLMs and generative AI today, because the real impact is emerging now. And yes, you have to standardize that and you have to build it into your infrastructure, but you can't not be experimenting with what's coming next, right. Otherwise- >> Just as an example, lots of people are very excited about agentic AI, and I'm trying to wrap my...
You've heard of it, obviously. You've written about it, right? >> Absolutely. >> I mean when you read it on paper, it sounds magical. It sounds like oh my God, you can develop this AI module that will automatically do what you want. And what's not to like about something that does what you want and it's a collaborator for you? The way I'm looking at it is that this is the evolution of application programming. Now, when I say the evolution of application programming, remember we started our chat when I told you that when I was entering college, I was programming in assembly language.
That's in 1962, '63. And later on, I was programming in Fortran. That's a long time ago. Here we are, I don't know, 70 years later.
And programming has gone higher, and higher, and higher, and higher, and higher, and it makes it easier. And you can get more tools, more abstractions, and things like that. And now, you can do incredible things with programming, some very sophisticated applications with "agentic" AIs.
But if you ask me now, " But you just talk to it with a large language model, which is English- like," well Scott, depends what you mean by you can just talk to it. Because the people who've shown me what they expect this will work, like Jerry Cuomo, who just retired from IBM, who's an IBM fellow. Yes. But you have to know how to do the hints to the model as to what you like. And it helps if you know Python and if you know other techniques. And you're not working a program, you're using them to tell the agentic AI system, "This is what I want.
" But you have to tell it what you want in specific languages. And if you don't have that kind of experience, if you just start talking like you and I are talking, expecting the agent AI to translate what we're saying into a program that will do incredible stuff, boy, you don't know what you're talking about. >> Well, what's going on out there today, you're right.
And everyone's saying that these agents are going to be able to make decisions, and they're going to be goal- oriented versus just prompt-oriented. >> But who's telling them the goals? The humans have to explain the goals in ways the poor AI system understands. >> Well, I think the challenge they have today is that they're trying to build this on generative AI and LLMs, and an LLM is insufficient. It's correlation at best. It is not AI reasoning. It's not decision intelligence. It's simply not capable of doing that.
And that goes back to the browser analogy that you got to build on top of it with semantic reasoning and causal reasoning. And those technologies are out there, and our research shows about 10 to 15% of companies are using it today. Netflix, Uber, BMW, Fitch Ratings. There's a ton of them out there that are using it.
That's what's going to differentiate an agent's ability to actually make some degree of decisions. All right, Irving, thank you very, very much for taking the time. This has been a fabulous conversation. >> Thank you for inviting me.
It's a pleasure doing this with you. >> You bet. We'll do it again sometime, and we'll certainly as soon as now that things are getting warm, we'll take another walk down there on Westport on the beach. So look forward to that. And thanks again.
And for all of you out there, thank you for joining us today. I really appreciate you being here on the Next Frontiers of AI Podcast. So appreciate it. Stay tuned for more research
and analysis from our team as we dive deeper into what is shaping the future of AI in business. Visit us at thecuberesearch.com and siliconangle. com to stay connected to and get access to all the AI research and podcasts from across the team. We are the leader in tech news and analysis. Bye for now.
2025-03-29 22:17
