Episode Transcript
[00:00:00] Today I'm going to talk about a really big philosophical topic that's been on my mind for a couple of years, and that is what is intelligence? Because the word is getting thrown around willy nilly and becoming quite meaningless unless somebody defines it. And what really threw me off was when Elon Musk was quoted in the Economist stating that within five years AI will have more intelligence than all of humanity. Which is the most absurd, ridiculous comment I could have ever heard. But of course, who knows what he thinks thinks the word intelligence means? And if you read the philosophers and how they tried to define it, they really talk a lot about reasoning and judgment, decision making, problem solving, not facts. So let me posit a very simple model that I think works for me, that might work for you. I think there's four parts to this phrase. First is horsepower reasoning horsepower. There are some of us that were born with quick minds. We analyze information quickly, we respond quickly, we can deal with large amounts of data and information. We can deal with numbers or words. And we have speed and capacity to handle high volumes of information or judgment or decision making. And I think the IQ test kind of measures that. And you see that in people you meet where some people pick things up very, very fast and some people take longer to learn. And I don't know if that's genetic or learned. It's probably both. The computer of course has almost unlimited horsepower. Not really, in fact a less horsepower than we thought because the way we've now programmed it with AI, it's quite slow to make decisions. And I think there's frankly some fundamental problems with the LLM model from the standpoint of intelligence as I see it. But anyway, that's number one. We'll talk more about that in a minute. The second is knowledge, facts, information, data. When you go to school or when you grow up and learn how to speak or walk or study or go into work, at some point you're learning facts, you're learning science, you're learning words, you're learning languages, you're learning, learning information about the world. And the AI systems of today are focused on that second level. Can we amass all of the facts in the world now? Of course the obvious issue is that not all facts are facts. There's a lot of things in the world that we don't know the right answers to. And so there's different versions of the same facts. I don't like the idea of alternative facts, but anyway, you know what I'm talking about. But nevertheless, if you look at the data labeling industry and what the frontier vendors doing, they are literally trying to assemble every fact, every image, every scientific or physical relationship or object in the world. Now, the problem with doing that, of course, is that new things are always being discovered. New laws of matter, new types of science. And so even if you did have an AI that somehow knew everything that was knowable, whatever that means, it would be out of date every second, because every second something new is being known. So this idea that we're going to train our models to know everything is kind of absurd, but that's level two. And I think perhaps when Elon refers to this, he may be talking about the density of knowledge that will be in these big systems. The third level to me of intelligence is a little bit different. I would call it judgment. Judgment is a word that works for me that describes somebody or a machine's experience in different uncertain situations so that they recognize patterns or they deduce patterns to try to make good decisions in the future. And when I say decision, I'm basically saying every decision is uncertain. There are no right and wrong decisions in most situations where we need AI, who to hire, how much to pay somebody, whether to fire somebody, et cetera. And some of the most intelligent people in intelligent organizations make terrible decisions. I mean, just talking about Elon himself, he's obviously a brilliant man in many ways as a scientist, but what he did to Doge didn't exactly work out. Apple, who you might argue is the most intelligent company in the world, wasted some number of billions of dollars on a car that either failed or was killed. And there were a lot of judgment decisions made, I'm sure, on the design of that car, the engineering of that car, and then the decision to kill the project. And we may at some point go back and undo the information that went into all those decisions and realize that maybe they didn't have very good judgment over there, or maybe they did. And you could go after almost any business and you will find judgment that is later proven to be flawed for a variety of reasons, sometimes because the world changed. And that third level of intelligence is very important. And I think in the human context, certainly in hr, that level three is extremely subtle and sensitive. And my belief, knowing what I do about work and business and psychology and genetics, is that judgment comes from intuition, from feelings, from emotions, and from a lot of non scientific things. If you listen to music, if you listen, if you read poetry, if you read literature, if you know a lot about the world and the way people and the world works, you're going to have better judgment about people and the world than somebody who didn't. Even though you may both have exactly the same number and facts and the same capacity to analyze those facts. And I think in the third level of intelligence, the AI has little contribution. Now that may change over time. But my experience with all of the models that I use and pushing them very, very hard, I asked these models do some very extensive things is they're very solicitous and they really don't make decisions very well. What they do is they try to make you happy so that you feel like you made the right decision. And I think there's judgment that's created through experience.
[00:06:04] It turns out, if most of you have been around as long as I have, that more senior people or more tenured people who've worked in the same domain or in the same company, or in the same function or in multiple industries, tend to be able to make better decisions because they've seen more uncertain situations and learned what did and didn't work and why. Now, let me make a very strange analogy here. There's a very, very interesting book called the Hidden Life of Trees.
[00:06:34] It's actually about trees, but it's a fascinating book. And because what it's about is all of the decisions and intelligence and judgment and information that trees collect and use, grow. Trees respond to light, to drought, to vermin and other parasites, to other trees. And they're incredibly long lasting living things. Even though they don't have a brain per se, but encoded into their genetic makeup, they have a lot of judgment. So judgment doesn't necessarily require the intellectual or horsepower part of your brain because there is no brain in a tree, but there's a lot of cells, or some equivalent of cells, whatever they're called, that have a lot of history embed in them in their response to conditions. And so I think judgment is very largely driven by these non computational parts of our brain or our bodies, and also by our experiences and to some degree by our genes and the genetic experiences that we have encoded into our very bodies. The fourth level of intelligence I would call wisdom, and the word wisdom of course, is slightly different than judgment. And it implies, and when you read the philosophers, you hear this a lot. It implies that not only are you quick to figure out things, you have a lot of information and you're well read or well informed and you have a lot of experience to make decisions well, but you have an ethical or moral or societal framework from which to make decisions that lasts over time. Now, you know, I don't think you would say that Adolf Hitler was wise. I don't think too many of you would. It could be somebody who says that. But when you see people that use their intelligence for evil or dysfunctional or selfish things, you would probably, probably say to yourself, they're missing judgment, they're missing principles. They're missing something in their intelligence. And I don't think it's fair for Elon or anybody else to claim that an AI is more intelligent than humans when it has no wisdom. It has no judgment or ethics or morals unless we tell it what those are. And I think most of you know that ethical issues are vague and gray and uncertain and surprising. And we're all susceptible to making the wrong decision at different times, depending on who we are, how we were raised, and the experiences we've had. And there are some people, in my experience, that have a lot of wisdom. And many times they're older, not always, and they just have a certain framework for thinking about the world and about life and about business and people that give them this sense that they know things that we need to pay attention to. And good leaders, great leaders who we follow tend to have wisdom of some kind. In fact, it's interesting. One of the most interesting facts of leadership was actually Colin Powell, I think, who said this was one of the only ways to measure a great leader is do they have followers, and how many or how avidly or how well do people follow them? And I know in the business world, you oftentimes don't have any choice but to follow the leader, but in the real world, you do. And there are many, many people who have wisdom, and we follow them for that wisdom. And to some degree, the social media industry, the podcasts, and all of the noisemakers we have in our society have different amounts of wisdom, and we find the one whose wisdom we respect the most as individuals. And that sort of leads to this other dimension besides these four levels or these four categories. And that is that intelligence is spiky. Spiky in the sense that no two people have exactly the same levels of intelligence in the same areas. You'll find somebody who, at the age of 10, becomes a beautiful musician or pianist or some other instrument and just was born with this incredibly natural affinity for music and notes and the variety of feelings and mathematical issues that are involved in either generating music or playing music. Math, science, art, acting. We're all gifted in certain areas, and in some ways, every one of us, I believe, have magnificent areas of high intelligence in some parts of life. I can't prove that to you. But I certainly in my work, in my life, try to respect everybody I meet and learn from them, their unique areas of intelligence and depth. And I'll tell you every, everybody I know has certain strengths that are unique to them. So if you add these four things up, horsepower, knowledge, judgment, wisdom, you could ask yourself, does an LLM have those four things? Well, could it possibly have those four things? Of course not. An LLM is fundamentally a statistical model that uses a very, very complex large amount of statistics to correlate the possible or likely arrangement of tokens. Based on an arrangement, tokens that come in creates an arrangement of tokens that go out. A self driving car doesn't really see anything, but we give it bits of information from either a camera or radar. And it uses those bits to manipulate and statistically correlate how those bits compare to bits of other things that it has seen in the past to interpret that there may be a person walking across the street, or it might be raining and slippery, et cetera. I don't call that intelligence. I think that's extremely powerful pattern matching. And over time, if it had access to deep, deep, deep levels of judgment or experience in driving, it would be better and better and better at that pattern matching. But when a whole new situation comes along, a child or a dog or some other weird situation that happens on the road, we don't really know how well it will interpret a very vague and uncertain situation. The tree has proven itself over tens of thousands of years that it knows how to adapt because it's still here. I have redwood trees in my yard and they've been, there's a lot older than I am. And so yes, I have confidence that it has a pretty good amount of wisdom about how to survive in the climate of Northern California, where I live. But an AI that's five years old, trained by a bunch of data labelers who are just out to make 50 bucks an hour, labeling data. Somehow I don't think it has that.
[00:12:58] Now, you know, those of you who are AI geeks are going to try to prove me wrong and you're going to show me statistically why the AI is smarter than the humans. But I think we're going to have to just deal with this very unstable certain issue. And I'm a scientist, I understand math, I understand statistics, and I understand what AI does. And I'm just as surprised and thrilled by what it does as anybody else. But we live in a world of uncertain things and situations that we haven't seen before. I mean, you know the most every week, every day, I see this IBM, which is one of the oldest companies in the tech industry, more than 100 years old. So there's an enormous amount of wisdom in IBM. Of course, a lot of the people that with that wisdom have left, but other people have come in, just saw its largest stock drop in 60 years this last month, because they completely misforecasted their revenue in certain areas of hardware and software. You know, I worked there for 10 years. I have opinions about the place. I thought they were crazy to sell off their hardware business and their PC business. But what do I know? Now that the world is all about hardware and not about software anymore, they're very poorly positioned. Somebody made a whole bunch of decisions in that company that in retrospect don't look very intelligent.
[00:14:13] And you know, so how do we possibly think that an AI could be more intelligent than we are now? There is one element to AI that I think could change our thinking in this area. Long range planning. There are a new breed of AI companies trying to build planning systems to try to plan for uncertain events like global warming or the future of the economy, whatever you may have, or, you know, the future of your company or your industry, whatever. And if you talk to those guys and you listen to them, and I've been listening to some podcasts about them, what they do is they basically say that the real benefit that they have is the depth and quantity of information. The actual processing that they use is very similar to the processing used in any AI. But what they're trying to do is amass huge amounts of historic information so that their AI works harder and uses more history to predict the future. Time will tell if that turns into wisdom. I really doubt it will, but that anything could happen. But I think the idea of intelligence being, you know, less than human is really important for us to consider because so far, in my experience, in our experience in our business, talking to hundreds of companies about hundreds of AI projects, there are lots of things that AI is really good at. Candidate experience, answering questions about your benefits, looking things up, analyzing data, analyzing a spreadsheet and so forth. But there's lots of things it's not clear if it's good at. And as we learn how to apply AI more and more and more, I think it's going to be important for us to think about what the word intelligence really means. Then I would also add the other conversation that we just had this week with a big organization, the human intelligence needed to manage the machine intelligence. By the way, I've mentioned this before, but I'll just tell you one example of how silly this is. I've been doing business with American express since the 1970s. So American Express has a database of almost everything I've purchased in my business career for decades. And you know, I'm a white collar guy and so I do the typical things that most white collar people do. Somebody got my Amex number and purchased a big heavy machine tool for construction and charged it to my account. I don't know, $10,000 tool of some kind. And I suppose what they were going to do is sell it. The AMEX AI did not identify that that was out of range and didn't block the purchase. I looked at that and I said to myself, I think American Express is one of the most sophisticated credit card companies companies in the world, maybe number one. And they have lots and lots of history, but they missed this one. And I guarantee you if I had been looking at these transactions as a human, I would have noticed that that one was weird. So these, you know, very, very statistically smart, highly validated systems may not be as intelligent in certain situations as we hope. And we as the human masters or trainers or users of these things are going to have to apply our levels of judgment and wisdom to train.
[00:17:17] If we can train them. I'm not sure if they can accept the kind of training that we have. Maybe they can. We'll find out and interpret them and manage them and control them and give them rules so that they don't make too many decisions. Now, you know, Elon Mus knows a lot more about the engineering of AI than I do, as does Nvidia, as do a lot of the engineers working on this. And maybe, you know, what I'm saying is absurd and silly as far as they're concerned. But I think for most of you who deal with human capital issues and leadership and management, I think it's healthy to be skeptical about the power of these systems systems and don't expect that they're going to give you brilliant answers out of the box. And I think in our case having our own AI, which we've been training just extensively for the last few years, it's very, very, very good at many, many things, but it's very narrow and it doesn't try to solve every world problem, only the problems in HR and management and leadership. And so I don't claim that it has the judgment that I have. I may have more judgment or less depending on the situation. But we test it and we testing it and I think that's the way we need to think about this stuff. Anyway, if any of you are experts at this, or philosophy majors or AI engineers and you want to discuss this, I'd love to have a panel and talk about it and compare notes on what areas of intelligence we truly have been able to automate in what areas we have not. Thank you.
