Before AI: Humans Were Building Robots Thousands of Years Ago

SPEAKER_02 0:03 AI is becoming more dystopian by the day. You've got creators showing how sycophantic it ha it can be, how it can manipulate you, and how it can make you believe your own BS.

SPEAKER_00 0:16 Yeah, you've got writers who've suddenly realized their brains aren't working quite how they used to, because ChatGPT has been work helping them a little bit too much. And you've got Coke making their usually heartwarming ad entirely with AI, left feeling a little hollow.

SPEAKER_02 0:32 So it got us wondering: if we could delete AI tomorrow, would we? And then we got thinking about the history of AI. I'm Ollie. And I'm Matt.

SPEAKER_00 0:44 And this is What We Did Before, the podcast where we explore the history of modern inventions.

SPEAKER_02 0:51 Now it may seem silly to be discussing what came before AI, because for many of us, its rise really began with the advent of ChatGPT, and that was only what, late 2022, early 2023? So surely we can all remember what it was like without it.

SPEAKER_00 1:10 Well, I'm not so sure, because so many of us become really reliant on it. And besides that, there's another very important point to discuss, and that is what we call AI has been around a lot longer than ChatGPT.

SPEAKER_02 1:26 Yeah, exactly. And the groundwork for AI actually began in the early 1900s, and mankind has actually been experimenting with machines that can act by themselves for a lot longer than that. Take automatons, for example. These are basically self-operating machines which often resemble humans or animals. Like the cuckoo on a cuckoo clock. Yeah, exactly. We've got some great examples of early automata dating as far back to uh 300 BC, and that was all thanks to a guy called Tusebius. So he was an inventor, a mathematician, and he was probably the first head of the great library of Alexandria. So he's well known for creating something called the hydrolis, which is a water organ considered to be the precursor of the modern pipe organ. But in terms of automata, he was also the person who perfected something called the water clock. So water clocks had been in existence since ancient Egypt, uh 1417 to 1379 BC, so a lot longer uh further back in history than Tusebius. But the water clocks from those times were really inaccurate. And then Tusebius came along and he basically spruced them up. So water clocks they told the time by letting water drip at a constant rate into a vessel. And then the rising water level uh lifted a float that connected to gears that then triggered sounds and moved little figurines at certain intervals. And his version uh of this water clock, his perfected version, was actually the most accurate way to tell the time for 1800 years.

SPEAKER_00 3:13 But around the same period, we also get the earliest known analog computer as well.

SPEAKER_02 3:18 Yes, the Antikythera mechanism, is that right?

SPEAKER_00 3:21 Yeah, well, I think that's how you pronounce it, yes. Something like that. Um so it was found in 1901. Um, and at the time it just looked like this corroded um lump of bronze and barnacles. Um, but with some imagination, and they uh broke it apart into multiple pieces. I think that was actually not intentional. Um and then they also did more recently some uh x-rays and um scans of the device, and they found that it had um something like 37 meshing gears and dials, and it was capable of um modelling the movement of the uh the stars in the sky and predicting eclipses, and this was all over 2,000 years ago.

SPEAKER_02 4:11 Yeah, I find it amazing because I'm not even entirely sure what an analogue computer is or how it would work. Like we're raised in the digital age, so I kind of get digital now, but it's quite amazing that this was at one point the prevailing technology, and digital was like a complete mystery, and analog was the way that they were doing it.

SPEAKER_00 4:31 Absolutely. I mean, I find with these kind of things actually it's more impressive to me from it uh being born in the digital age to look at analog technology and kind of the ingenious ways that they have come to solve these things because for us it's kind of like, oh yeah, you just get a computer program or you just make some electronics that will do that. Um but these are people that are inventing a whole mechanism of doing a thing that has never been done before um without electronics, without um prior knowledge or science. So yeah, it's it's crazy.

SPEAKER_02 5:10 Yeah, and what I find amazing is that given that we didn't even have a word for robots back then, even in Greek mythology, descriptions of automata are quite common. So, for instance, we got Talos, who is an artificial man of bronze who protected Crete from pirates and invaders. And if you look at this picture here, uh like he's henching. But this tradition it carried right through to the medieval times with the creation of singing birds, roaring and moving lions, and uh even wind-powered statues that could turn with the breeze. So this was something we were doing throughout history, it's quite incredible.

SPEAKER_00 5:54 I guess we were all just fed up of doing work ourselves, and we were just looking for something else to do it. Um, but this did all sort of die out for a while until the creation of Automata saw a massive revival during the Renaissance. So one very famous example is a humanoid mechanical knight designed by Leonardo da Vinci, and it consisted of pulleys that could mirror human movement. And one of my favourite examples um similar to this was the Mechanical Turk, a chess playing automaton controlled by levers and magnets. And this was in 1770. So the way it worked, if you look at this picture, there was this um kind of wooden puppet man at the top with a uh chess board on top. The idea, or I guess the scam was, uh, because of course it was used for scamming people out of money, um, was you know, can you beat the autonomous uh mechanical Turk at a game of chess? Um and so people would come and they'd play this uh puppet at uh uh chess, and they would be amazed because it would be moving and picking up pieces of and actually playing the game to the rules. And how it actually worked was um that there was a person inside the big box who had a chessboard in front of them, and it was effectively duplicated there, and all of their movements were controlling the hands of the um mechanical turk on top. Really smart, really well calibrated to be able to do that and not just accidentally uh you know rip someone's hand off or something, like or pick up, you know, the wrong piece or or whatever.

SPEAKER_02 7:45 So yeah, I thought this was um going to be quite small, but there's an entire human inside the box mechanism.

SPEAKER_00 7:52 Yeah, so it was a whole it was a whole kind of um desk arrangement, big th it was just you know, there was this big box and then uh the mechanical Turks sat on the side of it with a chess board in front of them. You would sit on the other side and play chess with them, but actually it wasn't just a desk, it was a big box with a person inside playing chess against you.

SPEAKER_02 8:16 Well, we'll come to it in a bit, but that's actually not too dissimilar to some of the recent robot announcements that we've had in the modern age. But yeah, does that count as automata if it wasn't actually uh autonomous, if it if it had to be sort of activated by a human to be used? I'm not sure. Even still, you know, back then it must have been mind-blowing to be beaten in a game of chess by a wooden man. And this actually happened even before the steam train was invented.

SPEAKER_00 8:49 Yes, it I mean, if you I they never actually explained how this worked to anybody, and it wasn't really revealed until it was out of service many years later, I think it was in the 1830s or something. Was that just one? So it looks like to me that it was that it was the only one. The Turk was a chess playing machine displayed in 1770, appeared to be able to play a strong game of chess autonomously, but whose pieces were in reality moved by levers and magnets by a chess master hidden in its lower cavity.

SPEAKER_02 9:26 The machine was how disappointing to be a chess master and then locked in a box.

SPEAKER_00 9:36 Could you imagine? Yeah, Magnus Carlson, sorry, you're gonna have to just spend your life in a box. Um so the the ma it was taught and exhibited for 84 years as an automaton and continued giving exhibitions until 1854 when it was destroyed in a fire.

SPEAKER_02 9:55 Right. Amazing. Um, so just to wrap up on automata, since you know we've got so many more sophisticated ways now to mimic the human form, um, automata of old, they've kind of fallen out of favour, and they're usually only created as pieces of art these days.

SPEAKER_00 10:14 Yeah, in fact, you can find a lot of these modern works at the MAD Museum in Warwickshire in England.

SPEAKER_02 10:21 Yeah, I've been on their website and I've seen some cool um artifacts, and it really is amazing that they've managed to keep the spirit of that alive. But I really want to move on to robots now, like proper robots that you would imagine now, you know, electrical machines, not automata. And, you know, I wonder um how long the idea of robots have been about, uh, how long sort of AI or machine intelligence has been a part of the conversation when it comes to robots. Like how deep does this run?

SPEAKER_00 10:52 Well, in the early 1900s, the idea of robots and artificial humans began to become popular. So the word robot comes from the Czech word for slave or forced labor, and that's robota, I think. And that was coined by um in 1921 by Czech playwright Karel Chapek, um, whose science fiction play Rossum's Universal Robots introduced the idea of artificial people. So these robots weren't what we call robots now per se. I think they were biological but artificial, not human. Um, and they were built to work on factory assembly lines, and they later rebelled against their human masters. So we've had this science fiction terminator obsession much, much earlier than uh the 1980s.

SPEAKER_02 11:49 Yeah, it's amazing that this has been such a big staple for the genre ever since even the word robot was coined. So for as long as we've been thinking about artificial people, we've been worried about what might happen if we actually create them. And you know, right now people are more worried than ever, I would say, especially with the possibility of artificial general intelligence headed our way.

SPEAKER_00 12:14 Yeah, absolutely. I mean, but it was always consigned just to science fiction. I think even when we were kids watching Terminator, I think nobody would have taken it as a serious thing. Maybe some people would, but uh they'd be labelled as conspiracy theorists. Um and now I think people are starting to see it as uh a legitimate possibility in some in some ways. Anyway, uh so among the first real-life autonomous robots were um these tortoise robots. Um they were called Elma and Elsie, and uh they were created by a neuroscientist called W. Gray Walter, uh, and he wanted to show how a small number of brain cells can give rise to complex behaviors. This was between 1948 and 1949.

SPEAKER_02 13:10 Yeah, these tortoise robots they used purely analogue processes right. So these were an example of analogue before digital, I guess really close um in time from when that shift happened. But yeah, this was analog.

SPEAKER_00 13:27 Yeah, yeah. So Elmer and Elsie relied on analog signals, like just changing voltages or current and using valves instead of transistors, etc., rather than digital processing. But even still, these were really impressive machines. Um, and they could even find their way back to their charging station when they got low on battery. So kind of a little bit like the modern autonomous vacuum cleaners that we have.

SPEAKER_02 14:54 Yeah, oh yeah, or lawn mowers. I really wanted to buy one of those, but they're like super expensive. Um But the first digitally operated and programmable robot was invented by George DeVall in 1954, and that was called the Unimate. So this is um a really cool industrial robot which ended up working on a General Motors assembly line.

SPEAKER_00 15:17 And did you see the recent announcement of um Neo, the humanoid robot that does stuff for you around the house?

SPEAKER_02 15:25 Yes, I did. I watched uh Marcus Brownlee's review of it, and hilariously, so this is what I was talking about before. This autonomous robot has to be remote controlled currently, or at least it was remote controlled in its demo, uh, the first demo that it did. So I wouldn't say it was quite the landmark revolutionary thing that we were hoping for. Um, and it's I think it's priced at $20,000 as well for some someone else to control it badly and for it to like trip up around the house and load your dishes and potentially smash them. So I'm hoping they do make some some some small improvements.

SPEAKER_00 16:05 Yeah, uh well, I I think this isn't the first time that uh these robot uh manufacturers have pulled something similar. I think there were rumors at least about the Tesla robots that were effectively just human-controlled. But anyway, with the Neo, something and the a lot of these robots anyway, but particularly the Neo, it's something striking to me is how the robot looks. It's not designed to simulate an actual human and look super accurate uh like we have in many movies. Um, I think a lot of people would feel a bit kind of uncanny valley about that, especially if you've seen any of these clips of like hyper-realistic kind of robot faces that are talking. It's weird.

SPEAKER_02 16:52 So um yeah, and in sci-fi when like a part of their skin gets peeled back and it's just a little you know slightly robotic underneath, and then suddenly that's the moment when you uh that's usually the frightening scare moment, and you're like, oh these robots are these humans are actually robots. But with Neo, what they've done is it's kind of got a blank face, and there's no nose, there's no mouth. I guess he doesn't need that. Um he's just got the two little eyes, which I assuming are the cameras, and then sort of big round pads for ears, but otherwise very plain, very basic and stripped back. I wonder if that will make it hard for robots to integrate into our daily lives if they keep going down that route, or if it's the only way that they can make robot uh robots integrate into our lives because it would just otherwise be really, really strange.

SPEAKER_00 17:42 Yeah. Well, actually, that's a good point. There's um when we look at science fiction movies and when we look at video games, um, and the YouTube comments of the robots that look a little bit too realistic, I think from most people's perspective, it's probably easier to not see them as a human.

SPEAKER_02 18:04 Especially if they're going to be serving us, like uh not uh they would don't have a will to serve against, but it is against their will. Like they're not opting in.

SPEAKER_00 18:14 Well, this is the thing, ethically, it's a lot easier, isn't it, to kind of dehumanize the machine that's being a servant for us than to have this like super realistic human and just demanding that it does stuff for you. It doesn't probably feel right to a lot of people, especially if you know just interacting with an LLM feels like you're talking to a human to them.

SPEAKER_02 18:39 But even though we are still far away from making them look properly human-like, if we go back to the first robots that we spoke about from the 1940s, so Elmer and Elsie, they were way less human-like still. In fact, they they weren't even intended to look like humans. Um so Elsie looks like a tiny mechanical creature on wheels and is built from what look like radio parts and bicycle scraps. So you've just got a load of exposed wires, gears, valves, and a chunky battery strapped to this sort of metal frame that in no way intended to look um even like an animal. I know they're called tortoise robots, but still, you know, it's got wheels, there's nothing animalistic about it. And then you've got Unimate, which we mentioned from the 50s, and this is just a big boxy industrial robot arm. Uh, it's got a heavy metal base, it's got this long hydraulic limb ending with a little simple claw at the end. So in no way were they trying to make it look like um they were biological beings. So it's quite interesting that we we sort of assume robots are these entities that are human-like, but it's I think probably quite a recent invention that we have decided that that was the case.

SPEAKER_00 19:57 Yeah, I guess um there's probably a couple of reasons for that. It's quite difficult to make a walking kind of humanoid compared to something with wheels. But in the modern era, I guess one thing that makes kind of sense, or one debate that you can have is um when you look at like the Unimate, it's perfect for the task that it's made for. I would imagine that humanoid robots probably aren't masters of any task, but because they're in a human-made world that's designed for humans, they can do a lot of tasks kind of quite well, whereas it makes them more uh flexible, let's say.

SPEAKER_02 20:38 Yeah, it's quite self-aggrandizing if you think about it, that we have assumed uh the ultimate direction for robots should be the human form, as if it can't be improved upon. That's true.

SPEAKER_00 20:51 Yeah, I guess just um if we want it to move around our houses, you know, one of the problems I had as a lazy person that lived in a um house that had three floors was I couldn't get uh one of these autonomous um vacuum cleaners because it couldn't go up and down the stairs. So if there was a humanoid robot that actually pushed my Henry Hoover around and took it up the stairs, then it could do it for me.

SPEAKER_02 21:21 I'm glad you've got a Henry too. That's as autonomous as I want to get. It's got a face, I'm happy.

SPEAKER_00 21:26 Yes, yeah, yeah. None of this, uh none of these actual evil robots. Autonomous Henry would be the peak of civilization.

SPEAKER_02 21:34 Oh god, going around just sucking people up. There should be a B movie.

SPEAKER_00 21:42 Yeah, absolutely. Um so anyway, we're um we are inching closer to human-like robots, um, and we've come a long way since then, obviously since the 1950s, just uh link back to what we were saying before that tangent. But so far, no mass-produced or commercial robot has really taken on the guise of a human completely. There are these uh clone robotics androids that have synthetic organ systems and 3D printed skeletons, and you can't buy those yet, thankfully, because look at them.

SPEAKER_02 22:24 At least from the inside of a human as you're gonna get, at least right now. Um, I have seen the images of them sort of covered up, and even then they've got the humanoid form without their features, so they're still far from being like totally humanized. But if you look at the pictures of the of the clone robotics androids, yeah, I mean you can see that they've tried to emulate even just like individual muscle fibers in the chest. You've got a rib cage as well. Uh I assume protecting some sort of inner workings, because otherwise, what's the ribcage for? But yeah, that I think they're probably a little way off making. Making anything from that. I think on their website it does just say something like coming soon. No, it says Androids have arrived. But then if you try to um scroll down, it's sort of like you know that there's you can't buy anything. Yeah. So Androids have not yet arrived. You'll be thankful.

SPEAKER_00 23:17 Androids are currently on their way.

SPEAKER_02 23:19 Yeah. But although we invariably want to integrate AI into robots, robots aren't AI. So let's track our history with artificial intelligence, proper AI. So over 180 years ago lived a woman who became known as the world's first computer programmer, and that was, of course, Ada Lovelace. So she was working on the analytical engine by Charles Babbage in 1842, 1843. And she noted something remarkable. Firstly, the analytical engine was quite remarkable in itself. It was designed to be the world's first general purpose computer. And this was a full century before electronic computers existed. It didn't actually get made, but it was sort of designed at least. But anyway, Ada realized that a machine like this, like the analytical engine, could manipulate symbols, not just numbers, and as a result of that, could potentially create things that no one directly programmed into it. And she even speculated that the engine, and this is a quote, might compose elaborate and scientific pieces of music. Now, we've got generative AI making pop tunes, and actually even heavy metal tunes, rightly or wrongly. And some argue that Ada Lovelace actually predicted this very outcome.

SPEAKER_00 24:50 Yeah, but this is still um hypothetical AI. It still wasn't an implicitly stated goal at this point. Where we take a real step forward is in 1935, thanks to Alan Turing. His concept of the Turing machine proposed that machines are able to imitate any formal process of reasoning. And by the nineteen forties, Conrad Zeus invented the world's first programmable digital computer. That led to Turing publishing his 1950 paper titled Computing, Machinery and Intelligence. The extract of the paper says, I propose to consider the question, can machines think? And this work, of course, later became the foundations of the Turing test. But it's John McCarthy who is regularly regarded as the father of artificial intelligence.

SPEAKER_02 25:45 Yeah, and it was John McCarthy who actually coined the term artificial intelligence, isn't it?

SPEAKER_00 25:50 Yeah, yeah. He did that in 1955, so a year before he organized the Dartmouth conference in 1956, which was considered the birthplace of AI. And that brought together the likes of Alan Turing, Marvin Minsky, and this marks the foundations of AI research. He also developed a programming language called Lisp, which was essential for AI research and is still used to this day.

SPEAKER_02 26:19 And as for the first working example of AI, that accolade is typically given to the logic theorist. So this was the first program deliberately engineered to perform automated reasoning. And this was as far back as 1955. And the logic theorist, it was capable of proving mathematical theorems, which we used to think required human intelligence.

SPEAKER_00 26:43 It could prove mathematical theorems in 1955. Yeah, and I can't do it in 2025. No, um I'm just really curious how that works. So you're not talking about just uh a glorified calculator, but something capable of reasoning.

SPEAKER_02 27:03 Yeah, so this really is the world's first AI program. So it ran on one of the world's first digital computers, actually, which was the um Johnny Acc made by the Rand Corporation. So the logic theorist essentially solved maths problems by thinking them through step by step, almost like a human would do. And then instead of trying every possible answer and just, you know, throwing all the eggs and just giving it a go, it actually used smart shortcuts and uh rules of thumb to choose the most promising steps and then build a logical proof.

SPEAKER_00 27:37 That does sound quite a lot like how the um reasoning sort of systems work in modern um AI, actually. Um so more recently, ChatGPT added this uh reasoning function. In around 1966, we got the precursor to ChatGPT in the form of ELISA. This was a natural language processing computer program made by Joseph Weisenbaum that you could chat to. So it used pattern matching to simulate conversation, kind of like modern LLM's work, um, playing the role of a psychotherapist, and it seemed to understand its users, uh especially in 1966. Obviously, it was very it's hard enough to explain to people uh that it doesn't really understand you now. Uh so although it doesn't actually understand anything, people were projecting human-like traits onto it. Uh, some were even convinced it was a human. Even Weisenbaum's own secretary began to grow very attached to the thing. This was known as the ELISA effect.

SPEAKER_02 28:48 Yeah, I I don't know if you've ever used ELISA. Um I mean, it's very obviously not a human to us now. But I can imagine being an emergent technology, it's like, you know, you've never experienced anything like that responding to you in real human language. Largely if you use it now, it just kind of repeats your question back at you, maybe a bit like a psychotherapist might actually do. Um, but it's not that convincing compared to ChatGPT, but you can imagine, yeah, uh in the 1960s, that was that was incredible. And this phenomenon, the ELISA effect, it it did have an impact on a lot of people. And I think we're seeing a similar i it effect with people marrying ChatGPT or um having you know relationships with it. Yeah, but anyway, by 1972, there was an early form of AI developed at Stanford University, and it was called Mycin. And it was actually made using McCarthy's programming language Lisp, and it was made to help diagnose bacterial infections and recommend antibiotic treatments. Uh Matt, as a computer programmer yourself, maybe you can explain how Mycin worked? I can try.

SPEAKER_00 30:00 So um so it probably used a uh rule-based program that just used if-else or if-then rules. Um it would make decisions based on reported symptoms and medical test results from patients, and could even explain its reasoning to physicians. And amazingly, it operated at roughly the same level of competence as human specialists diagnosing blood infections and better than general practitioners. This was called um an expert system, or something that we'd perhaps call narrow AI these days. Kind of like um AlphaGo, which is this program developed by Google DeepMind. Um, have you heard of it?

SPEAKER_02 30:47 Yeah, it it beat the world's best Go players. Although before AlphaGo came about, I had no idea what Go was. Yeah. But it's it's this intricate sort of game for smart people, uh kind of like chesses in a way, isn't it?

SPEAKER_00 31:01 Yes, yeah. I think um I think it's more popular in Asia, possibly. Um I'm not sure. We'll have to fact check that one. So AlphaGo is very smart, obviously, but it was only good at that one thing, um, kind of like deep blue, I think it was called deep blue for chess, um, back in the 90s. It beat Kasparov, I think. Still, Mycin was a landmark in the history of artificial intelligence and medical AI. Now, although a lot of money was poured into the field after 1956, progress was much slower than anyone had anticipated, and interest began to fade. So by 1974, much of those initial funding streams were just pulled.

SPEAKER_02 31:46 Yeah, but this wasn't the end of AI development, obviously. Although this is where things get a bit hard to define, right?

SPEAKER_00 31:55 Well, I think it's entirely hard to define artificial intelligence, but the history is obviously quite intertwined with computing generally. So around this point it was still in its infancy, and AI, as we know it, can only exist today because of the market growth and the technological advancements in the computing space generally.

SPEAKER_02 32:17 Yeah, and I think this is where we come to the crux of the matter, and that is the fact that what we call AI isn't really intelligence. Ultimately, it is just computer code. So we can include the development of expert systems as a part of this history, really. Like it it's fair to do that, I think.

SPEAKER_00 32:37 Yeah, yeah, and obviously what we did before computers or what we did before the internet could and probably will be episodes in themselves. Um, it would be remiss of us not to mention these points as a very important chapter in AI's history.

SPEAKER_02 32:54 Yeah, so with personal computing and then um the computerization of workplaces ramping up in the 80s and 90s, there were many systems that came about to do tasks traditionally done on pen and paper, uh, from stock management to accounting.

SPEAKER_00 33:09 Yeah, and we're talking about a time before the internet was a big thing. Once that started taking off in the 90s and the World Wide Web was invented, investors weren't so interested in AI. There was this period of interest waning and criticism of the field at the time.

SPEAKER_02 33:29 Yeah, this is actually called the AI winter. But interest did start to grow again in the 2000s, and this was because of a rise in the use of machine learning. So this is probably a good time now to define what AI really is. What is the definition? Because the term is kind of vague.

SPEAKER_00 33:49 Yeah, I think the term is actually quite dangerous, to be honest. Um we call it artificial intelligence, which evokes this idea of a thinking, feeling computer brain, a sentience, like we said before, that can uh has an experience of being. The reality is that AI during this time looked like very narrow specialized tools, things like spam filters, search ranking algorithms, and uh early medical diagnosis systems. Even now, what we call AI can be summarized as a really good prediction engine, made possible just purely by really impressive machine learning and deep learning techniques, a lot of training data that we've provided to it from the whole growth of the internet. So, how did we get here from a time when Amazon and Google were startups? YouTube didn't exist, and most people who use the internet used Internet Explorer. No one uses that anymore.

SPEAKER_02 34:57 Um God. Although they keep on trying trying to make me set it as my default browser. Oh, Edge, yes. Or Edge, yeah.

SPEAKER_00 35:10 It's still just as bad.

SPEAKER_02 35:11 No, it's it's actually all right, but uh I that can be for our what we did before computers episode. But um, back to machine learning. So that was really the first thing that took off in both academia and industry. So machine learning was being applied to a a wide array of problems through general technological improvements and the computerization of data. Hardware was powerful enough and affordable enough to start being used to crunch through immense quantities of data and find patterns that would take humans decades.

SPEAKER_00 35:45 And not to mention advancements in algorithms that facilitated this type of technology. You have sort of two different approaches here, right? You have an improvement in the maths and the actual code behind it, but also the power of the hardware that can process it, both pushing forwards. And this advancement continued somewhat quietly from the general public's perspective, anyway. Uh, we were all just so focused on the internet, you know, Web 2.0, when uh you could finally start sharing things on YouTube and social media, um, and then the hardware improvements in video games and CGI in movies, smartphones, all of this uh took our attention while quietly AI kept progressing. And to many of us, ChatGPT springing up in 2022 seemed like it came out of nowhere.

SPEAKER_02 36:42 Well, uh it kind of did to some extent. Um there were a few technological breakthroughs that led to ChatGPT, and there was a really big leap, and that was deep learning. So instead of giving the computer rules, we let it we started letting it learn patterns from these huge amounts of data that you you mentioned. And that's what made things like speech recognition and image classification suddenly get way better. And because Transformers had come along a few years before in 2017, this new architecture let AI pay attention to different parts of a sentence at the same time. And so that meant it could understand context, memory, and long-range meaning much better. So transformers are basically the reason ChatGPT can produce paragraphs that uh actually make sense.

SPEAKER_00 37:32 And uh there were also some pretty public forays into AI before ChatGPT. So I remember using this thing called Murphy bot. I think this was back in 2016. It was this Microsoft image generator where you could generate or modify images with a plain text prompt. So it seemed amazing at the time, obviously, in comparison today, it's uh nothing like what we have now. But you could upload a picture of yourself and say, like, what if I was flying a plane and it would um find a picture of a plane and find a pilot and put your face on it, and it was quite it was quite clear. It was bad photoshop simulator, yes. Um and and that was 2016 that you could do that. Something like that. Around 2016-2017. There's also this image generator from NVIDIA that got a bit of interest called um Gogan. Uh, and you may have seen some pictures from around this time and so that were generated and some videos. It used to find faces in things that it would generate. It was like an LSD trip, um, which is how early image generation looked. It was very, very interesting. And there were chatbots like Tay from Microsoft, which learned from internet data and the conversations it had with users.

SPEAKER_02 38:59 Yeah, but the these attempts didn't take off for a number of reasons, like the fact that Tay turned into a Nazi after hours of learning from the internet and it had to be shut down by Microsoft only days after it was launched.

SPEAKER_00 39:12 Yeah, which unfortunately is a little bit unsurprising, really, if you think about the sort of 4chan-esque culture that was around the internet in that time. And these early AI tools were impressive, but mostly as tech demos, you wouldn't uh seriously as a company want to uh unleash Tay as your as your customer service representative. Um probably not a good look for you.

SPEAKER_02 39:44 Yeah, not a good look. But you know, things like Tay for all its faults, we learnt from it, and then we had the transformer technology as well, which allowed computer generation of really impressive long passages of text that made you know grammatical and contextual sense. All of this is what took the world by storm, kind of combined.

SPEAKER_00 40:06 Yeah. At the time when ChatGPT came out, it kind of felt, at least to me, like we were in a bit of a period with stagnation with technology, um, you know, just uh iterating on the things we already had. Uh, and suddenly that wasn't the case anymore. We just didn't hear about these things, or we weren't taking them seriously, and suddenly Chat GPT was here.

SPEAKER_02 40:30 Yeah, and now it seems like AI is everywhere. I mean, I don't go a day without hearing the term or interacting with it in some way. It's actually quite exhausting. Yes. And there's really there's really been a huge investment into the industry, potentially unsustainably so. And it's I've really it's generated this enormous amount of hype in a very, very short space of time.

SPEAKER_00 40:55 Yes. So you were right. Uh despite it being so recent, given the impact, it's almost becoming harder and harder to remember what the world was like before it.

SPEAKER_02 41:07 Now, Matt, we're both gamers, and a place where AI has been integrated into our our lives for ages, at least seemingly, is in video games. And I've always called non-player characters or enemies in video games AI. You know, the AI was rubbish in that game. Um, and that was that was way before the term was associated with ChatGPT and Sora or ge Google Vio, things like that. But that kind of AI doesn't count as true AI. So, how long has actual AI been a part of our lives?

SPEAKER_00 41:41 Well, it's been a lot longer than people might think. So, spam filters have used machine learning since the 90s. Search engines have used AI to rank results since the early days of Google. Uh, Netflix and Amazon rely on these recommendation algorithms which will pull from big data in a similar way using machine learning. Your phone has been using AI for autocorrect, face recognition, photo enhancement for over a decade. Um, you can even see it if you zoom in on some of your pictures. You can see these weird artifacts where it's tried to improve the text and stuff. Even fraud detection, speech recognition, translation, all powered by proper machine learning. And they've been quietly running in the background long before Chat GPT arrived. Does Clippy count as AI? I miss Clippy. Uh might be one of the only people that does. But um yeah, the little uh paperclip assistant introduced in Microsoft Office 97. He was amazing. Um, he was just a pre-programmed little helper.

SPEAKER_02 42:53 Yeah, I mean you say you might be one of the only people who misses Clippy, but he's interesting actually, because there's been a bit of a Clippy revival. I don't know if you've uh noticed this, but there was the YouTuber um Lewis Rossman. Yes. He started a bit of a movement that got loads of people changing their profile pictures on YouTube to Clippy, and this was in protest of shady marketing and data collection tactics which were being used by big companies. Yes. But uh in my mind, it's also got I don't know, sort of undertones of protest against this uh AI being shoved down our throats in a way. Yeah as well. So I think Clippy's quite an interesting like uh callback to potentially what we thought, you know, something something that we hated in the past and now we long for.

SPEAKER_00 43:42 Yeah, well, we're old enough to remember what it was like when Clippy was the worst thing that Microsoft did, if you remember. Um everyone liked to hate on Clippy. Um but yeah, you're you're right. I mean, there's this debate going on about whether a copilot or any AI should be part of your operating system. Should it be part of your word processor or Excel or um should it be everywhere? Maybe that's something that um we should be able to choose. And actually, you could choose not to use Clippy. You you if you wanted Clippy to go away, you could. He was sometimes annoying, sometimes he came when you didn't want him to. Uh looks like you're writing a letter, uh, whatever. But yeah, I I really guess.

SPEAKER_02 44:36 He wasn't being aggressively marketed to you on a daily basis.

SPEAKER_00 44:40 He wasn't he wasn't stealing your data. Um and you know, he was actually pretty cool, to be honest, in that context. So thinking um ahead from Clippy to today, what tasks have we largely replaced thanks to AI? Well, we don't need to remember directions anymore. And even further to that, in satinavs and Google Maps, it will detect your the best route to take. It will take into account things like traffic and road closures and determine this on the fly. We don't need to hand sort photos because facial recognition groups them. And we don't need to manually back up files, manage inboxes, tune our cars or adjust our thermostats. Algorithms do all of that for us automatically. We've outsourced our music choices to recommendation engines.

SPEAKER_02 45:36 Yeah, I yeah, yeah, I'm not sure how I feel about that. Yeah, me neither. And then if we think personally, like what did we do before AI? So for me in my working life, you know, writing podcast descriptions, doing research into topics and guests, editing transcripts, that was all done manually by myself. And Chat GPT, things like that are saving time on some of those tasks. Although the caveat is it's not doing things that were impossible to me before. I think that should be uh that should be an important point. Um, and actually personally, I make a very concerted effort not to offload all of that stuff to AI anyway, because I want to keep my brain sharp and my content original, because it doesn't produce original content.

SPEAKER_00 46:22 No, well, this is the thing. Um LLMs and other generative AI feels like a huge leap, and don't get me wrong, it is. But if we take a wider look at it and the history of computing generally, it's just one of the many steps that we've been making in the name of efficiency. And chat GPT doesn't actually think or understand, it just predicts what words usually come next or what should come next. That's why it hallucinates regularly and makes stuff up. It looks intelligent, but under the hood it's basically just statistical pattern matching, not conscious reasoning.

SPEAKER_02 47:00 Maybe discussions about AI these days aren't so much about whether it's good or bad for us, because honestly, given the timeline that we've just shared with you, and it's rich history from automaton all the way to what we've got now, and all the things that AI has and is still helping us to do, is amazing. I think the question now actually is how much AI is too much, and can you have too much of a good thing?

SPEAKER_00 47:27 Uh yes. Um I think you can, and it's a under debate whether it is a good thing. So let's just talk for a moment about how dystopian AI videos have become. You've seen the comparison of Will Smith eating spaghetti in 2023 versus 2025. Yes, I've seen it. Yeah. Um, it was crazy, right? The the initial stuff was impressive just two years ago, where it was kind of obvious it was Will Smith. Now it could well be Will Smith. And it's just incredibly quick that tech has developed.

SPEAKER_02 48:05 Yeah, maybe it is just Will Smith and he's he's trolling us, but um, he's just come along and eaten some spaghetti just to make it look like AI's advancing quicker than it is. But it is it is it's insane. Like not long ago, President uh Donald Trump posted this strange AI-generated video of himself playing football with Ronaldo in the White House.

SPEAKER_00 48:28 Yeah, and I think there were debates about whether he some of his announcements on true for Twitter were AI generated as well. Um, and I think there was some protests going on in the US, and he posted this AI video of him flying a jet and dumping human shit on the on the protesters. It's just obscene what kind of world we live in. Anyway, sorry.

SPEAKER_02 48:59 It is weird, yeah. Like, I mean some UK Labour politicians have actually been accused of using AI to write their speeches as well, and because loads of them included this one phrase specifically, I rise to speak, which is something that we don't say in Britain typically, but I think it's more of an American thing, is it?

SPEAKER_00 49:17 Yes, yeah, I heard about this. Um and one MP was spotted using Chat GPT to respond to constituents on the train. Um, so it it shows how normalized AI-generated text has become even in politics. And um it's it's strange because actually just a couple of years ago I think we would have found that like absurd that you someone would write something um supposed to be kind of heartfelt or at least authentic, and they are outsourcing that to a machine. Now it kind of seems normal.

SPEAKER_02 49:52 Yeah, people are writing wedding speeches with Chat GBT. Yeah. Um and I can't think of anything more sad.

SPEAKER_00 50:00 Yeah, it is sad really that the sort of poetry of language will get uh automated away for people. Then there's this more worrying trend you said about uh wedding vows. Uh there's this more worrying trend of people falling in love with chatbots, and even one teenager whose parents say he was encouraged to take his own life by one. But to your point, Ollie, from AI being used in medical applications to identify signs of serious illnesses earlier, helping to progress science in ways that humans previously struggled with, such as um Alpha Fold, which was used to solve some uh protein folding problems, um, which can have these massive impacts on how we maybe approach kind of previously very difficult to uh solve diseases, to simply making video game enemies more challenging and complex, uh, or even to just give you good recommendations of what to watch on Netflix. There are plenty of wins.

SPEAKER_02 51:08 Yeah, there are. So I think it's not as simple as saying it's like a good versus bad scenario with AI, but I do think using AI to write for you to the point that you can't even remember the information that you wrote or communicate it to a friend. There was a recent study on uh students who used AI to write um, they basically monitored students who used AI fully to write their dissertation, I think it was, um, those who used AI in conjunction with their own writing, and then those who did the writing completely manually. And the students who did all the writing manually, as you might imagine, they were able to retain the information for a lot longer and generally their just uh ability to write was unaffected, whereas um those who used AI completely did start to lose their ability to write. So, you know, that does seem bad. Or, you know, perhaps the fact that we're creating soulless art. We've got uh videos and music being created that kind of suck all of the fun out the creative process. Actually, I heard something a while ago that it reminds me of, um, and that's the fact that we all hoped AI would automate the boring parts of life so that humans can be free to create art, but instead, AI is creating art while we do the boring work. Yeah. So maybe we have just pushed it a little too far.

SPEAKER_00 52:33 Yeah, I think so. Um, I think in some ways it seems like Wally, the movie, was a warning. Uh and you've probably heard of the dead internet theory, right? Yes, yeah. It's this idea that most of the internet is now run by bots and automated comment content, with only a tiny fraction still created or used by real people.

SPEAKER_02 53:01 Yeah, and I can see that happening. I mean, I even receive like 20 requests for people to feature on I make podcasts and I make podcasts for loads of different people, and so I get loads of requests for guests to feature on those shows, and I'd say 99% of them now are like clearly written by AI, like really obviously written by AI. So I'd just ignore it. And to be honest, I think the golden age of the internet was the early 2000s. I I think if we're gonna go back to any time, we had it right then. Uh you had forums where people wanted to share good information just because you had video games that felt new and exploratory because they hadn't been solved already, and you couldn't just go online and find out somebody who's already solved it. I remember back in the days when uh if you really were having trouble on a video game, you'd have to buy a book, you know, the puzzle solver book, and then follow it to try and help you with like the new legend of Zelda or whatever. Uh and back in the 2000s, you know, early 2000s, we had MSM Messenger as well. That was so good. And then we yeah, it was after that we took it too far, and then you know, it really feels like we're doing that with AI too.

SPEAKER_00 54:10 Yeah, and I mean obviously it depends on your viewpoint. I tend to think that although the connectivity that we have now does have a lot of benefits, I really liked this kind of concept with MSN uh and similar messages where you were going to your computer to sort of consent to messaging with people, whereas now we are expected to be reachable almost 24 hours a day because we have this device that pings every time someone messages us. But it's different for everyone. I think we probably look back with a little bit of nostalgia, and the internet has its faults or had its faults in the Wild West days, but there was a lot to be said for the place where passion and excitement for being connected was the main objective, and the technology was just really fresh and interesting, and people were using it to share the things that they as huhans were doing, and now the ability to connect to us and use our attention is just being exploited.

SPEAKER_02 55:18 Not to mention that the relentless push towards accelerating into an AI future that you know, frankly, I don't think we're ready for as a society, is um pouring billions of dollars down the drain into data centers. Um we're using insane amounts of power and fresh water to do that. And then even despite the money pumping in, instead of excitement for the future, we just hear of the looming threat of AI-related job layoffs. So we're doing all of this, we're spending all this money, we're creating this potential damage to the climate, all for us to be like more frightened and worried than ever. So I guess with all that being said, should we should we be scared about the future? So perhaps this is where we could consider what we do after rather than what we did before. Like, what do we imagine a post-AI world might look like?

SPEAKER_00 56:11 Oh that is a tough question because um it's easy, I think, to at the moment be very pessimistic. Obviously, the technology's impressive, and we've said that there are some real benefits of this. Um, but what we have um is also a lot of new challenges. We have uh the greatest system for scamming and propaganda that we we have ever created, just to name one thing that we'll have to deal with. Um we have the imminent case now where we won't be able to trust visual evidence or audio evidence of anything, and we have a lot of processing power and power generally in private companies, uh, which was already a problem before. However, I don't think we should be scared. I'm sure we will have to adapt. There are things that we'll have to work out along the way, but there could be positives. Perhaps the sheer amount of generated content could mean that people will ultimately learn not to trust everything that they see on the internet so readily, which obviously I think that's probably something that we could say over the last decade or so has been a bit of a feature of sort of our politics. Um so uh maybe there are some positives to be had from this.

SPEAKER_02 57:44 Yeah, I personally think we're gonna see, and I think we are seeing this return to uh people seeking out more human experiences, either in real life or in like actual, you know, valuing actual human-created art and videos and podcasts and music a bit more. And I think that's where a series like this is quite interesting because we can actually look into or back into our history at what we did before, take stock of what led us to where we are now and and maybe choose a point in time where we did have things right. I sort of facetiously said that was the early 2000s, but I do gen, you know, genuinely think for all the progress that we've made, I think we should keep that, but look back at where perhaps we got other things right, even if we didn't get everything right. Because I think as a society we have come a lot further. But we can take little elements from the things that we did in our past and integrate them because there's no reason why we can't. I mean, there's no rules saying that we have to accelerate into a future of AI slop. Maybe some people don't agree, um, but I don't think all progress is positive or necessary. Exactly.

SPEAKER_00 58:58 Yes. Um we don't have a crystal ball, but it's clear that AI is already reshaping the world around us. And it's only been three years since ChatGPT was first launched.

SPEAKER_02 59:11 So, given everything that we've outlined here about AI, its history, beyond the very obvious ChatGPT stuff, I want to put the question to you listening: would you delete AI tomorrow if you could? Or perhaps you'd delete just some parts of it? Let us know in the comments. This is what we did before. I'm Matt. And I'm Ollie. And if you've got any ideas for a future episode, please do let us know. And if we choose your idea, we'll credit you in the outro of that episode. So please like and subscribe if you enjoyed the episode, and we will see you next time.

SPEAKER_00 59:48 Bye for now.