A Brief History of AI: From Machine Learning to Gen AI to Agentic AI

Artificial intelligence may feel like some brand-new tech trend, but the truth is AI has been evolving for over 70 years. From simple math puzzles to today's powerful neural networks, each generation built on the previous one. Let's take a look at where we've been and where we might be going with this two-part AI series, beginning with A Brief History of AI.

Let's start our tour of AI with a guy named Alan Turing, who back in 1950, proposed what became known as the Turing Test. Now, Turing is known as the father of computer science. So, the guy did a lot. And one of his contributions was this as a way to

measure if the a computer was really intelligent or not. So, this is how the Turing Test works. You have

a human subject and they're separated it by a wall. They can't see who it is. They're typing on a keyboard, and they're gonna communicate with either an a computer or another person on the other side of this. And if they're typing messages and getting responses back with

these two things, if this person cannot tell if they're talking to another person or a computer, then we will judge that this thing is considered to be intelligent. So that was what he proposed with this. And that was the gold standard that was taught to me back when I was in undergrad, riding

with this. And that was the gold standard that was taught to me back when I was in undergrad, riding my dinosaur to class. This is how we measured things, and this is where all of that stuff started off. The term AI actually was coined a little bit later in 1956, and

started off. The term AI actually was coined a little bit later in 1956, and then we started really progressing along this timeline.Ah. Back in the late, ah, 50s, there was a programing language that came out called Lisp. And Lisp was for short for list processing. And in

my early days of AI programming, this is what we used. So, that was back in the the early 80s. That

was really still considered to be the predominant way you you did things with AI. Now remember, I said programming. Our modern stuff isn't so much programmed as it is learning and will come to that

programming. Our modern stuff isn't so much programmed as it is learning and will come to that in a few minutes. But Lisp, ah, interestingly enough, was first implemented on an IBM 704 system. So, IBM was back there, ah, in those very early days, and it relied very heavily on this notion of

system. So, IBM was back there, ah, in those very early days, and it relied very heavily on this notion of recursion, which is something that doubles back on itself. Ah. It was very complicated to program in. But

think about it this way if you don't know what recursion is, I I saw a definition that said the definition of recursion is c recursion. So again, the thing doubles back on itself. It gets very complicated really quickly, but it can also be very powerful and very elegant if you do it right.

But if you wanted to change and make your system smarter, you had to go back in and write more code.

This was programming. Now, in ah, in later in the 60s, we came out with something called ELIZA.

And ELIZA was really the first, ah, chatbot if you wanna think of it that way, ah, well before the chatbots of today, and not nearly as sophisticated. It was designed to kind of be conversational and it talked to you very much like a psychologist would. So, it would ask you, you know, "How are you doing today?" You would respond and whatever you responded with, it would do the

standard kind of "And how do you feel about that?" and, and go with those kind of of responses. But it

gave us the first sense of a system that felt like it was understanding us. Now, it it also did, ah, some crude version of natural language processing. So you could put your your words not just in specific commands, but you could actually put it in a way that you could express yourself.

And people started getting the sense that they were talking to an intelligent being. In the 70s then, we started having a different programing language that people started to, to glom on to, for doing AI programming, and ah I I really began ah to start using it in the 80s. And the the name

of the language is called Prolog. It was a short for programming in logic. And the idea was instead of having these recursive systems that that we had with Lisp, with Prolog, we had a bunch of rules.

And you would set down a whole bunch of rules, maybe relationships or things like that, and then have it run inferences against those things. But again, with both of these systems, one of the major hallmarks was if you wanted to make your system more intelligent, add more capability to it. You

had to go back and add more code. So you were programming these systems. They were not really learning in the in the sense that we think of it today.Ah. Then in the 80s, this is when we started having a boom in the area of expert systems. The idea was that we could have systems that would learn a certain amount of things. We could put certain kind of constraints in it, and then it

would be able to figure out ah certain advice that it could give us in particular context. Businesses

were really big on the potential and there was a lot of hype, a lot of expectation, but it never really delivered on that expectation, not in the big way that everybody was thinking. So, this kinda went through ah, ah, if if people were getting a little bit interested. Then they started getting a little less interested when they saw that the expert systems were kinda brittle. They were not

able to really be malleable and learn as quickly as we'd like them to. Then there was a big milestone that occurred in 1997. IBM built an AI system called Deep Blue.

And what Deep Blue did was for the first time in history, we had a computer that beat the che the best chess player in the world, Garry Kasparov. Now, it had been thought that you could write a computer program that would be able to beat an average chess player, maybe even a very good chess player.

But to overcome the ah intelligence, the expertise, the planning skills, the strategy, the creativity, the just sheer genius of what it would take to be a chess grandmaster, it was thought no computer would ever be able to do that. Well, again, that happened in 1997. That was

actually a a good while back. And when it happened, it really signaled again a resurgence in the thoughts around AI and what this thing might be able to do. Then, ah, we moved on to in the in the 2000s on. Now, this technology had actually been around in research for a

while, but it's when it really started to catch people's imagination that we started to see the growth of machine learning and deep learning algorithms, where machine learning was now doing pattern matching and deep learning was simulating human intelligence through neural networks. So, this

thing then started to grow across. And in fact, we're still using that technology today as the basis for how we're doing AI. But this was a big departure from the Prologs and the Lisps where we were programing a system. In this case, the system was learning. We would show it a lot of different things and then ask it to predict what the next thing was, or I show it a bunch of things

and ask it to tell me which one doesn't belong in this group. So it was pattern matching on steroids.

That was machine learning, and it was learning through seeing these patterns and recognizing them. But it could do it on a massive scale that would be very hard for humans to be able to

them. But it could do it on a massive scale that would be very hard for humans to be able to accomplish. Then we took machine learning and deep learning capabilities, and there was another huge

accomplish. Then we took machine learning and deep learning capabilities, and there was another huge milestone that happened in 2011, when the TV game show, ah, IBM used a computer called Watson to play Jeopardy! And Jeopardy! is a game, if you're not familiar with it, asks a lot of

trivia questions in a lot of different areas. This was actually a very difficult problem to solve for a number of reasons. One, because the questions come in natural language form, and the the way we express ourselves with language can be varied, ah, in the great degree. There are things that we use

like puns and idioms, figures of speech. If I say that, ah, it's raining cats and dogs outside, you know I don't mean that that there are small animals falling out of the sky. But those are the kinds of things that go into the clues that are in Jeopardy! And we had to have a computer that would understand those vagaries of human language and understand what to take literally and what not to.

You couldn't just program rules or, ah, some sort of list processing that would know and anticipate all of those. You can't even list all of those that you know, those idioms. So this was a really hard problem. IBM had, ah, a, a case where we use our Watson computer to play against two of the

hard problem. IBM had, ah, a, a case where we use our Watson computer to play against two of the all-time Jeopardy! champions. That was again in 2011, and we beat them both, ah, three nights in a row.

all-time Jeopardy! champions. That was again in 2011, and we beat them both, ah, three nights in a row.

This was another big milestone in AI. And it's interesting to me that this actually came along much later than winning at chess, ah, where, ah, it's because there's so much variability in this and the subject matter is so broad. So you had to be an expert in this, this system, and it couldn't

just be going out to the internet and querying all these things. It had to be coming up with answers very quickly because, you know, if you've ever seen the game show Jeopardy! if you don't answer quickly, then someone else will answer it for you. And, if you answer if you're the first one that answers and you're wrong, then you lose points. So you had to calculate how confident am I

in my answer? So, this was, ah, a lot of really important work that showed the possibility again for AI, after there had been a period of kind of disappointment and people hadn't really seen much come out of all of this. Around about 2022, we had another major inflection point where AI

suddenly got real for a lot of people, and that was when we introduced this idea of generative AI based on foundation models. And here is where we started to see the rise of the chatbots. And

that's what got everyone's imagination, because now we were seeing not a a fairly stiff natural language processor like ELIZA was. It was very limited in terms of what it could talk about. Now

we had something that acted like an expert, and it would do all kinds of amazing things. seemed to

know the answer to everything, be very conversational. And this is when for a lot of people, it felt like AI finally got real. And it generates more than just text. You know, we could have it write a report for us. We could have it summarize emails or documents, things like that.

Also, we could use it to generate images or generate sounds And from that we could also generate deepfakes. So I could have something that is an impersonation of a real person that looks

generate deepfakes. So I could have something that is an impersonation of a real person that looks realistic enough that it would fool someone. So, a lot of good, a lot of bad, a lot of all of this happening, but a lot of excitement. And as I said, for a lot of people, this is when AI suddenly

got real even though it had been happening for a long time. And then where are we going with this?

Well, we're already seeing 2025 I think has been the year of the agents. This is when we start seeing agentic AI coming in, where we're taking an AI and giving it more autonomy, where it's able to operate on its own. We give it certain goals and things that it's supposed to accomplish, and then it uses different services in order to accomplish those things for us. So. we're gonna

see a lot more of this happening as well. And now where does the future head for us? Well, the short version is if all of this is a sort of artificial, narrow intelligence where the intelligence is specific in particular areas, things that it can do, well, the next thing to be would be artificial general intelligence, where we have something that is as smart or smarter than a person in

essentially every area that we could imagine. And then the next area would be artificial superintelligence, where we have something that far exceeds human capabilities in terms of intelligence across a wide variety of things. So you can see with this, basically, we've it's been a what felt like a snail's pace of progress as we move from these early days until we started

adding more and more capabilities with machine learning. And then we started introducing generative AI, and now we're off to the moon. For decades, it felt like AI was just a pipe dream. Then suddenly it seems like AI can do everything. But can it really?

Well, in the next video, in this two-part series, we'll take a look at what are the limits of AI, both in terms of what it can do and what it can't do, at least not yet.