The Man Who PREDICTED ChatGPT Has A New SHOCKING Prediction!

So is the AI hype going too far? That's what one MIT roboticist pioneer Rodney Books, thinks, that people are vastly overestimating generative AI. This video is going to dive into someone who predicted the rise of large language models and what he's predicting next about AI and how it's going to shape our opinion of what's to come next. And honestly, this genuinely might surprise you. So we can see here by this article on TechCrunch. MIT roboticist pioneer Rodney Brooks thinks people are vastly overestimating generative AI. Now this article is rather fascinating because it dives into the nature of his predictions. He's someone that has been in the space since 1997, or in fact I do believe a little bit earlier than that. And since then he actually has made predictions about the future of AI and, and he also keeps a scorecard on his blog about how well he's doing. And essentially the article states that when Rodney Brooks talks about robotics and artificial intelligence, you should listen. Currently the Panasonic professor of Robotics at emeritus at MIT, he also co founded three key companies including Rethink Robotics, iRobot and his current endeavor, Robust AI. He also ran the MIT Computer Science and AI Laboratory for a decade starting in 1997. Basically, he knows what he's talking about and he thinks maybe it's time to put the brakes on the screaming hype that is generative AI. Brooks thinks it's impressive technology, but maybe not as quite as capable as many are suggesting. I'm not saying the LLMs are not important, but we have to be careful with how we evaluate them. And we're going to get into some of his predictions later. But just take a look at what he says about the current state of AI. He says that the trouble with generative AI is that while it's perfectly capable of performing a certain set of tasks, it can't do everything a human can. And humans tend to overestimate its capabilities. When a human sees an AI system perform a task, they immediately generalize it to things that are similar and make an estimate of competence of the AI system. Not just the performance on that, but the competence around that. And they're usually over optimistic because they use that to model of a certain performance of a person's performance on a certain task. Basically, they think if this AI system can do this one thing well, they, they think that it can do many other things well. He also added that the problem is that generative AI is not human and not even human like, and it's flawed to try and assign some human capabilities to it, he says. People see it as so capable, they even want to use applications for it that don't make sense, he says. Brooks offers his latest company, Robust AI, a warehouse robotics system, an example of this. Someone suggested to him recently that it would be cool and efficient to tell his warehouse robots what where to go by building an LLM for his system. In his estimation, however, this is not a reasonable use case for generative AI and would actually slow things down. Instead, it's much simpler to connect the robots to a stream of data coming from the warehouse management software. Now, one thing that I do think that is really interesting is his position on humanoid robots. He essentially thinks that what we are currently building in the common world with humanoid robots is quite, quite wrong. He explains that it's also about robots and humans also working together. So his company designed these robots for practical purposes related to warehouse operations as opposed to building a human looking robot. In this case, it looks like a shopping cart with a handle, he says. So the form factor that we use is not humanoids walking around. Okay, not humanoids walking around. Even though I have built and delivered more humanoids than anyone else. Okay, these things look like shopping carts, he said. It's got a handlebar, so if there's a problem with the robot, the person can grab the handlebar and do what they wish with it. After all these years, Brooks has learned that it's about making the technology accessible and purpose built. I always try to make the technology easy for people to understand and therefore we can deploy it at scale. Basically, he's stating that in his current company, the form factor is not most effective for humanoids because you're not able to essentially use it and fix it in certain scenarios. Like he said, if there's a problem with a robot, a person can easily grab it and do what they wish with it. And he does have a weird example here. And there was a comment that kind of, you know, I guess you could say explains why this is somewhat of a weird comment. But at the end of the day, this is a very reputable source for someone that's been in the space for over two decades, he said. He adds that there's a mistaken belief, mostly thanks to Moore's Law, that there will always be exponential growth when it comes to technology. The Idea that if ChatGPT4 is this good, imagine what ChatGPT5, 6 and 7 will be like. He says that this is flawed logic and that tech doesn't always grow exponentially. In spite of Moore's Law, he uses the ipod as an example. For the first few iterations, it did in fact double in storage size from 10 all the way to 160 gigabytes. If it had continued on that trajectory, he figured out we would have an iPod with 160 terabytes by 2017. But of course we didn't. The models being sold out in 2017 actually came with 256 gigabytes or 160 gigabytes, because as he pointed out, nobody actually needed more than that. And I do think that whilst, yes, this is a point that Moore's Law doesn't always work in every single, I guess you could say, analysis of technology, I don't think that this example is right. Forgive me if I am making a completely moronic statement here, but I think the more applicable comparison with would be processor speed, because that's essentially what, you know, I guess you could say computing power, LLMs. I guess that that is essentially what you could compare it to. Now, he says here that Brooks acknowledges that LLMs could at some point help with domestic robots, where they could perform specific tasks, especially with an aging population and not enough people to take care of them. But even that, he says, could come with its own unique set of challenges. People say, oh, the large language models are going to make robots do things they couldn't do. But that's not where the problem is. The problem is being able to do stuff is about control theory and all other sorts of hardcore math optimization. Now, the most useful thing that I found about this certain post was not this entire post itself, it was very insightful. But the most important thing was the link to his blog post where he keeps a scorecard on his blog of how well he is doing. So this is what I wanted to show you all as well, because this actually shows you guys what he does believe about the future of AI on robotics and where we're going to go in certain things. You can already see his opinions on generative AI and what he believes about humanoids. But now we're going to take a look at some of his predictions because whilst, yes, some of them have come, you know, to be true in the past, I wonder if this is going to be the case once again where someone is underestimating the abilities of AI and the generative landscape that we're currently in. So here's where we have his predictions on robotics, AI and machine learning. He says it's worth reading this story. The story he's referencing is one actually about Sam Altman and he says that with the increasing number of CEOs in Silicon Valley ending up in jail for overhyping their businesses to the point of fraud, many more walk that line, some for multiple companies at the same time. And from the story, governance got a bit loosey goosey during the bubble. Vice president of financial strategy at Kruze Consulting, a provider of financial strategy services for startups, lately Mr. Jones said he has noticed venture firms doing more due diligence on potential investments, but they probably shouldn't get a gold star for fulfilling their job description. So basically with this article from the New York Times, you can see here that he actually is referencing the startup fraud and how it's unraveling. If you haven't been paying attention to the San Francisco startup scene, and it's not just something that is unique to San Francisco, there are many different startups that essentially, I wouldn't say it's not their fault, but I would say that startup culture breeds a situation where people always want to appear more successful than they are. And in doing so, they end up making ridiculous claims. And in doing those ridiculous claims, people will go to many attempts to have their egos, you know, satisfied and in some cases as far as jail. So you can see basically the absence of controls, A headspin is a part of an increasingly noticeable pattern at Silicon Valley startups that have run into trouble over the past decade. Investor in tech startups were so eager to back hot companies that many overlooked reckless behavior and gave up key controls like board seats, all in the service of fast growth and disruption. Then when the founders took the ethos of fake it till you make it too far, investors were unaware or helpless. Basically this entire article is stating that, look, there is this fake it until you make it attitude in Silicon Valley. People want to be the next big hotshot on the scene. And in doing so, this results in a remarkable level of startups that have, you know, been shown to be just simple frauds. I mean, Theranos, if you've ever heard about this story, this was the biggest fraud ever. You know, there's like a 30 minute documentary on it. I could go into all the details, but long story short, this person was pretending to be this, you know, genius level CEO who developed this revolutionary product that could take a prick of your blood and you'd be able to get all of your diagnoses within a matter of seconds. And the thing was just a terrible machine and that didn't even work. And it basically, you know, allowed people to look okay and say, take a step back for a Minute, let's actually analyze what on earth happened here, because we had millions of dollars of backing, everyone saying that this is a good product, millions of people vouching for her, and now this woman is in jail and is convicted of fraud. And the similar thing happened with ftx. But the point is, is that he's stating that, look, when things are getting really, really hypey, we always need to have a rational voice in our heads because sometimes we can lose sight of what is actually happening at the ground level. He also says that you can see here that this is what happens when everyone is entitled to their own alternate facts. Current icons, young, brash, worshipped CEOs are not immune to forcing their own alternate facts upon eager investors. And second, people willing to set aside critical judgment or when promised a magical, rosy fortune. And basically, this is the article here where he's actually referencing how Sam Altman, you know, how he's been going through some trials and tribulations at the board seat, or I guess you wouldn't really call this trials and tribulations, but I guess you know how he's had a history of rebounding from leadership crises with the support of powerful allies. After being fired by OpenAI's board for alleged dishonesty, he quickly rallied support from influential figures like, you know, Airbnb's CEO Brian Chesky and Microsoft CEO Satya Nadella. And this pattern of resilience has marked his career since his first startup Looped, despite calling, you know, many people calling for his removal and the employees there at Y Combinator, where he was asked to resign for prioritizing personal products. So basically they're stating that his ability to leverage a strong network has consistently helped him climb into more powerful roles. However, this leadership is remaining under scrutiny and investigation into recent turmoil at OpenAI begs the question, what is really going on? Now, the reason I've included this information into this video is because it sets the tone that gives you guys a realistic view on how the market is. Because whilst, yes, you're on this channel and you're viewing this content, you're kind of somewhat in a bubble. Now, it is good because I would say that the majority of people are not paying attention to AI technology and they usually just wait till it appears on the news or someone is showing off a magical new demo that is going viral. But the point here is that we can see that if we actually take a fundamental view of what is going on. There are a few things that do seem a bit strange about the current startup scene. You can see here that he says, I only made three comments in the table this year and only one of them is directly about a predicted milestone being hit. And as you guess, they are all about generative AI and large language models. No question that 2023 was the year when these topics hit the general consciousness of the scientific, cultural and political worlds. I've officially been an a researcher since 1976, and before that I was a high school and undergraduate hobbyist. But this is the first year that I've heard politicians throughout the world say the words artificial intelligence. And when they have said those words, no one has been surprised and everyone knows what they are talking about. And he says, I had not been bothered to predict a rosy future for humanoid robots. As when I made my predictions, I had been working in that space for over 25 years and had built both research humanoids and thousands of humanoid robots that were deployed in these fact, the extra difficult challenges requiring the fundamental research breakthroughs were clear to me. There are plenty of naive entrepreneurs saying that work will be changed by humanoid robots within a handful of years. They are wrong. I'm guessing that this statement is referring to many, including Elon Musk and maybe Nvidia's Jensen Huang, stating that, look, you know, these humanoid robots are going to change the way work is going to be. Elon Musk did say that there's going to be hundreds and thousands and even millions of humanoid robots in, you know, in the 2030, 2040, 40s. So I'm wondering if he's stating that since he's worked hands on, you know, and been an AI researcher since 1976, that, you know, his entire contribution to the field is more relevant than some of these quote unquote, newer entrepreneurs. So he says that my lack of predictions about humanoid robots were based on my expectations that they will not play any significant role for at least another 25 years. That is an incredible prediction. Okay, I'm going to state this again. My lack of predictions about humanoid robots were based on my expectations that they will not play any significant role for at least another 25 years. So essentially he's stating that if we take today's current date, which is 2024, that means that until 2050 we won't see humanoid robots in any significant roles. And I don't think that's true just based on the things that we've seen. But of course I could be completely wrong. I could be someone that's just completely hyped up on things, tech demos and, you know, maybe just all the demos that we've seen have been completely falsified because stuff like that does happen. I mean, I'm not at the companies, I'm not there. I'm not working behind the scenes. But of course these things can't happen. He says, here are some of the humanoid robots that I and the teams that I've led have built. Of course, he shows some of these humanoids right here that you can see and what they've been able to do that are really interesting and really cool. And of course, robotics is much harder than you can think. And yeah, it's a pretty, pretty crazy statement. But here's where we get onto his AI stuff. He says that I had predicted that the next big thing in AI beyond deep learning would show up no earlier than 2023, but certainly by 2027. And I also said in the table of predictions in January 2018 that for sure someone was already working on the Next Big Thing and that papers were most likely already published about it. I just didn't know what it would be. But I was quite sure that of the hundreds of thousands of AI projects that groups of people were already successful and working on, one would turn out to be the next big thing that everyone just hopes is just around the corner. I was right about both 2023 being when it might show up and that there were already papers about it before 2018. Why was I successful in those predictions? Because it always happens that I just found the common thread in all the next big things in AI and their time constants. Now, here's where we get into generative AI, which is rather important because this is where we get to certain dates. He says the next big big thing, generative AI and large language models started to enter the general AI consciousness last December. Then he talks about. I talked a little bit about it in last year's prediction update. I said that it was neither the savior nor the destroyer of mankind, as different Camps has started to proclaim right at the end of 2022, and that both sides should calm down. I also said that perhaps the next big thing would be neurosymbolic artificial intelligence. This is basically just combining the different levels and reasoning capabilities of multiple different AI systems into one that is able to kind of think like a human. And I do believe that you're going to see this term neurosymbolic AI be more prominent in the next decade, because this is more, you know, I guess you could say it more resembles what humans are in the sense that they're like an AGI system. So neurosymbolic AI is essentially an approach in artificial intelligence that combines neural networks, which are often used in deep learning, with symbolic AI, which focuses on high level human readable knowledge and reasoning. This hybrid method aims to leverage the strengths of both paradigms to create more robust and versatile AI systems. So basically you have the neural networks which are deep learning, which are excellent in packing recognition, learning from large data sets, handling unstructured data, and they're particularly good at speech recognition and natural language processing. But their weaknesses, things like LLMs, the black box nature, our difficulty in incorporating explicit rules of knowledge and challenges in reasoning and understanding. Now, symbolic AI is good at representing and manipulating explicit knowledge using symbols and rules, excels in tasks in logical reasoning, planning and understanding. And essentially it's limited in handling unstructured data and learning from raw data, which is pretty difficult to do. So this is where neurosymbolic AI comes in, combines the best of both worlds and, and we get, you know, really useful system. So he says, By March of 2023, it was clear that the next big thing had arrived in AI and that it was large language models, the key innovation, had been published before 2018, in fact 2017. So I'm going to claim victory on that particular prediction with the bracketed years. Okay, maybe I was a little lucky. And that paper had, you know, and that major paper for the Next big thing had already been published by the beginning of 2018. Okay, so I was even luckier. I think this might be sarcasm. So you can see here that here's where we get into the actual predictions. And he's already made some startling claims about the entire space. But I do honestly understand what he's trying to say here. So he says, so we can see here in the table that these are some of his 2018 comments. And then of course, these are some of the updates. So these are some of his 2018 predictions. Okay. And we can see here that the emergence of generally agreed upon the next big thing in AI is beyond deep learning. Of course, this was what he predicted. And if you want to know, you know, his, I guess you could say acronyms, you can see that net means it's not going to happen before that year. So not earlier than this year. Then by this year means it's going to happen by that year. And not in my lifetime means not before 2050. So the colors just mean the green mean, you know, these are accurate predictions that he made. So these are correct. The red ones were predictions that were too pessimistic, meaning they happened earlier. And the Blue ones were, you know, too optimistic. So we can go back and take a look at this. So if we come back to here where he said emergence of the generally agreed next big thing in AI beyond deep learning, this was a correct prediction in 2018 of no earlier than 2023 and at least by 2027, which is a pretty accurate date. He says whatever this turns out to be, it'll be something that someone is already working on and there are already published papers about it. There will be claims on this title earlier than 2023, but none of them will pan out. He said it definitely showed up in 2023. It was in the public mind in December 2022, but it was not yet that the big thing became during 2023, a year ago, I thought that it perhaps would be neurosymbolic AI, but it is clearly LLMs and ChatGPT and its cousins. And as I predicted in 2018 something was already being worked on as the attention is all you need paper, of course here he talks about the press and researchers generally mature between the so called Turing Test and Asimov's three Laws and valid measures of progress in AI and machine learning. And he said that this would be not earlier than 2022. And of course you can see here he was way too optimistic about this, that the Turing Test was missing from all the breathless press coverage of ChatGPT and their friends in 2022. And their performance, though not consistent, pushes way past the old comparison. Now here's where we get into some humanoid robot predictions. He says that the dexterous robot hands generally available are not going to be available essentially no earlier than 2030 and by 2040, he hopes, he says despite some impressive lab demonstrations, we have not actually seen any improvement in widely deployed robotic hands or end effectors in the last 40 years. So he believes that this is not going to happen by 2030 and will happen by 2040. Or at least he hopes also what's here, and this is a pretty crazy one because we're about to see in the next couple of years, we're truly about to see here. He says a robot that can navigate around just about any United States home with its steps, its clutter, its narrow pathways between furniture examples. He said a lab demo for this is not going to come before 2026. An expensive product that you'll be able to buy is not going to become before 2030, and an affordable product that you're able going to be having in your home is no earlier than 2035. Of course, what is Easier for humans is very hard for robots. So this is something that we still have I think about a year and a half to do. I mean one of the companies that I think is either going to do this is probably either Tesla or figure robot is going to do this. I think that one of those companies could do this, navigate around the US home, maybe they're going to show us a different demo. But it will be interesting to see what happens there. And of course he says a robot that can provide physical assistance to the elderly over multiple tasks, for example, getting into and out of bed, washing using the toilet, rather than just point to a solution, this is not going to happen before 2028. He says there may be point solution robots before that, but soon the houses of the elderly will be cluttered with too many robots. So an interesting prediction there that soon the elderly, they're going to be having so many robots and I do see that that as being an application because robots that move slowly and are strong and can help someone around the house, that is certainly a real world use application. And he says here a robot that can carry out the last 10 yards of delivery, getting from a vehicle into a house and putting the package onto side the front door. The lab demo he says is going to be no earlier than 2025. And deployed systems, you know, this is actually interesting. So he says deployed systems could actually be here by 2028. So he says that, you know, in the near future we could have a situation where we get, you know, robots that are from Amazon that are basically rolling around in neighborhoods and after they get on the truck, they essentially go over and they deliver the package. And this is not going to happen before 2028. But that is the time frame of where it could happen. And I think that that's a rather interesting prediction because we're seeing that robots, you know, are already doing similar things like this. If you've seen things like agility robotics, what they've been able to do, it doesn't seem too incredible. Of course he's talking about, you know, a conversational agent that carries long term context and does not fall into easily recognizable patterns. Of course this is already there. You know, he's saying deployed systems are going to be 2025 and it's not hard to get ChatGPT to connect it to speakers and microphones. But of course it doesn't have any updatable memory part apart from its token buffer which is what just been said. And of course long term memory has been, is going to be coming and now he says here an AI system with ongoing existence, not before 2030. You know, no day is the repeat of another day. So an AI system with an ongoing existence. This is pretty crazy because I think this is actually really hard to achieve and this is something that I haven't really heard about or seen any research papers about. And essentially here no day is as a repeat of another day as it is currently for all AI systems. So you know how currently when you have ChatGPT and you ask it exactly what's going on and it's like, oh, this date is 23rd December 2023, that's the current date. So that's going to be something that's really interesting because I'm wondering if AI systems are going to be like, wow, how are you doing today? You know, we got this new president, what's going on, yada yada, yada. But of course he's going to have an entire blog post to explain this. And here's where we get into some similar predictions. Like Yann Lecun and the likes of Gary Marcus. A robot that seems as intelligent and as attentive and as faithful as a dog. This is so much harder than people imagine. Many think we are already there. I say we are not all there. And he says that this is not going to be, you know, as earlier than 2048, which is very, very far away. Like 2048 is 24 years from now. And if we look at life, you know, 24 years ago, you know, technology has improved so much since then. And of course he says here that, you know, this is never going to happen in his lifetime. So not until 2050. A robot that has any real idea about its own existence or the existence of humans in a way that a six year old understands humans. I'm guessing that this is not like how an LLM understands humans with the sense that, you know, it's just got a textual based understanding of what humans are and what they do. But, but like, you know, a six year old's understanding of humans and their place in the world and how they interact and what humans can actually do. He says not in his lifetime. So I mean, one thing that I will say is that, you know, this is a prediction and predictions about the future are extraordinarily hard to do. And these predictions are extraordinarily hard because the future is very hard to predict. It's just the nature that it's the future. There are essentially unknown unknowns and if there is an unknown unknowns, those unknown unknowns can change the course of what might happen. So the, these predictions are extraordinarily hard. And I know some people might say, okay, this guy's going to get proven wrong. I mean, it's going to be interesting to say because this guy has decades of experience in the field. He predicted ChatGPT and the rise of LLMs, and he's got so much experience in humanoid robotics that I would argue it's pretty hard to find someone who's as well versed as him. So the point here is that, do you guys think that this is a prediction that reframes your mindset on where we're going to be with certain predictions? And it's also important to note that he doesn't just say that, look, AI is dead, nothing's coming, yada, yada yada. He states that, look, while these tools are useful, some of these things are going to come soon, and some of these things are going to happen in a pretty long lifetime. Of course, there can be breakthroughs that can accelerate these, I guess you could say, predictions, but it's also kind of difficult to do that without the proprietary information. One thing that I would like to know is what is really going on in these labs and if he knows exactly what's going on in these labs, because research is no longer being published. And I think this can kind of change the kind of nature of what kind of predictions people are going to make. Essentially what I'm stating here is that, you know, prior to 2018, 2019, 2020, all of the research that was done by all of these labs were essentially published and it was shared and everyone was advancing at the same pace. But now that, you know, we've got such this terminal race condition where all these companies are trying to compete, we've got like all of these researchers that are, you know, having breakthroughs, but they're not sharing the breakthroughs that they're making. Which means that currently we have a system where it's hard to make predictions because the breakthrough that are being made internally, we only get shown when we're shown a demo. So we won't know what's truly capable until they kind of show us exactly what's there. I'm guessing that he probably likely has factored this in, considering he's working in such an advanced field. But it is something that I do think has fundamentally changed this, which is why I think in the future now, it's going to be even harder to predict what is going to happen. Because like I said, With GPT4, they came out and shocked us. And of course, with Sora, they came out and shocked us. So it's going to be interesting to see what kind of things we're going to see here. So will be interesting to see if some of these predictions come true. And hopefully you guys enjoyed this video. And let me know if you think these predictions are going to be coming true, because, of course, some of them do, you know, are a bit contrasting with Ray Kurzweil's Singularity. But I think some of them are pretty reasonable, like a robot that can navigate around just about any United States home, a Lab demo by 2026. That is still pretty early. Okay. And he's saying that between 2026 and 2030, that's a pretty reasonable deadline. So this isn't something where he's saying it's never going. I just think that this is a document worth reading.