Don’t Believe AI Hype, This is Where it’s Actually Headed | Oxford’s Michael Wooldridge | AI History
What this covers
Michael Wooldridge, a prominent AI researcher at Oxford, sits down with interviewer John Jang to challenge the dominant narrative around artificial intelligence risk and to trace how we arrived at the current moment. The conversation spans the full history of AI from its inception in the 1950s through today's large language models, using that historical arc to argue that existential risk scenarios are implausible and distract from concrete harms already unfolding. Michael Waldridge contends that singularity narratives rest on assumptions—unguarded AI systems with autonomous goals and unchecked self-improvement—that would represent a catastrophic failure of basic safety design rather than an inevitable outcome. He locates the appeal of these scenarios not in empirical AI capabilities but in archetypal human fears, the Frankenstein narrative, and a psychological hunger for cosmic meaning akin to religious apocalypticism.
The interview reconstructs how repeated cycles of hype and disillusionment have shaped AI's development: the optimism of the Golden Age, the winter that followed symbolic AI and expert systems, the pivot to embodied and agent-based approaches, and now the current dominance of deep learning and transformers. Throughout this history, Wooldridge identifies architectural constraints repeatedly overlooked by each generation—pattern matching capabilities mistaken for reasoning, toy problems mistaken for real-world competence. He argues that today's large language models, trained on vast digital datasets, excel at low-stakes tasks where training data is abundant but remain fundamentally disembodied pattern-matchers incapable of embodied interaction, genuine planning, or the evolutionary-shaped intelligence that grounds human cognition. The near-term risks he emphasizes instead—AI-generated content making truth indistinguishable from fabrication, algorithmic fragmentation, and state-level deployment of synthetic disinformation—are tractable and addressable but chronically underfunded relative to speculative existential concerns.
Waldridge argues that the Singularity narrative is implausible and distracts from real AI risks; studying AI history reveals that symbolic AI's supposed failures were premature, that current foundation models are architecturally limited to pattern recognition rather than reasoning, and that the field's progress comes from engineering scale rather than theoretical breakthroughs.
- The Singularity requires implausible conditions (ungoverned superintelligence with recursive self-improvement) that contradict how AI systems actually work
- LLMs succeed only on tasks with abundant data and low real-world stakes; they cannot solve novel problems through logical reasoning, only pattern-match against training data
- Progress in AI has come from throwing compute and data at problems, not from discovering elegant principles—a 'sobering lesson' that mirrors the messiness of human intelligence itself
This asset isn't compiled yet
You're seeing its claims, ranked. Compile it to build the argument threads, weight them, and check each claim against your library — the full view.
Science is not orderly progress from ignorance to truth but rather a messy process involving false turns, ideological crusades, and paradigm shifts; it is not purely rational but contains elements of belief, commitment, and struggle similar to religious or ideological debates.
“I think one point in the book I say you know if you if you think that science is about orderly progress from ignorance to truth absolutely is not it's messy false turns um almost kind of like ideological Crusades I mean and it really is ideology religious this rounds in a full circle the apocalyptic uh mentality of the of the ex risk people”
The Frankenstein narrative—creating something that turns on you—resonates deeply in human consciousness as an archetypal fear; this primal narrative shapes current AI apocalypticism more than empirical AI capabilities.
“you create something you have a child and they turn on you you know that kind of the Ultimate Nightmare for parents you know you give birth you nurture something you you create something exactly so uh or you know and this that narrative that story is very very resonant and for example you go back to the the original science fiction text Frankenstein that literally is the plot of Frankenstein you use science to create life to to give life to something to create something and then it turns on you and you've lost control of that thing so it's a very very resonant idea I think and so very easy for people to latch onto right”
Philosophical questions about AI (does it think? is it conscious? does it understand?) have shifted from pure philosophy into experimental science because we now have systems to test; questions that were purely metaphysical are becoming empirical investigations.
“we've gone from a period where a lot of questions in AI were purely philosophical questions they were literally reserved for philosophers until a few years ago uh and suddenly it's experimental science you know uh are large language models conscious well let's roll up our sleeves and do some experiment and find out no by the way they're not uh but you know these are now practical handson questions”
Waldridge anticipated that networks would be ubiquitous (correct prediction) but failed to anticipate the specific form: the World Wide Web, centralized search (Google), and e-commerce (Amazon), which represent symbolic/search-based AI rather than multi-agent delegation he expected.
“by the way having realized that networks were the future I completely failed to anticipate the worldwide web or Amazon or any of that I look I look at the missed opportunities in my life um for for doing transformational work I totally got that networks were going to be the future but I still didn't understand exactly what that future was going to look like”
Early in his career (late 1980s), Waldridge realized that computer networks would become ubiquitous and that this would intersect with AI; the natural question was: what happens when AI systems can communicate with each other across networks? This motivated his work on multi-agent systems.
“as an undergraduate in the 1980s uh I was fascinated with AI uh but I also became fascinated with computer networks and you have to remember at the time computer networks were not common...I got the opportunity to work on the UK's extension of that called Janet The Joint academic Network and I had a kind of moment of Revelation at which point I realized this is going to be the future networks they're just going to be everywhere”
Studying the history of AI teaches that it's very easy to get over-excited and read too much into technical breakthroughs; past cycles of hype—the Golden Age, expert systems, Psych—show repeated patterns of unrealistic expectations that eventually disappoint.
“because the history of AI still has lessons to teach us and one of the big lessons that it teaches us is that it's very easy to get OV excited and to read too much into what you're seeing in Ai and people have done that on multiple occasions in the past”
Computers were invented as a byproduct of Turing's attempt to solve the Entscheidungsproblem (decision problem), not as a primary goal; he needed the mathematical abstraction of the Turing machine to solve whether mathematics could be automated, and only later did codebreaking work and others realize physical machines could be built based on this model.
“it's kind of one of the great ironies of mathematical history that computers get invented as a byproduct I mean he wasn't setting out to invent machines that could do things he was setting out to solve the idun's problem and he had to invent computers in order to do that”
Paradigm shifts in science result in the loss of important moral intuitions and techniques from previous paradigms that should be recovered; Thomas Kuhn showed that STEM innovation often involves paradigm shifts where valuable methods from prior frameworks are forgotten and may be worth rescuing for future breakthroughs.
“in philosophy there's an idea that a lot of moral intuitions good moral intuitions are lost through Paradigm shifts so we gain things in this whole Christian worldview but we moved away from the Roman world and the Pagan world and there's things to be rescued from that world that have been forgotten and my my training early training was in stem stem usually doesn't study historical stuff right you usually just study the latest physical theories but there is a view of even stem Innovation Thomas [ __ ] being the biggest proponent as being these Paradigm shifts that there are things that are important that are lost in previous paradigms”
AI history teaches that it is easy to get over-excited and read too much into breakthroughs; people have repeatedly believed that a single technique (search in the 1950s, deep learning a decade ago, now Transformers) is the 'magic ingredient' for AGI, but each time the field discovers there are missing ingredients we don't yet know about.
“the history of AI still has lessons to teach us and one of the big lessons that it teaches us is that it's very easy to get OV excited and to read too much into what you're seeing in Ai and people have done that on multiple occasions in the past um now I think with the current wave of AI I think there is real substance here I think we are at a breakthrough moment uh but I'm not convinced that we're at the end of the road in AI or that the Transformer is the magic ingredient um a few years ago people were saying deep learning alone is the magic ingredient for AI now it's the Transformer architecture and so on I don't think either of those things are the magic ingredient I think there are some ingredients that we don't yet know about”
Computers were invented not as a primary goal but as a byproduct of Alan Turing's attempt to solve the Entscheidungsproblem (the decision problem) in the 1930s—one of the great ironies of mathematical history.
“one of the great ironies of mathematical history that computers get invented as a byproduct”
The philosophical question of whether machines can understand or think became a practical question once early computers demonstrated sophisticated intellectual feats, causing people to ask: if a machine can do mathematics faster and more accurately than any human, isn't that evidence of intelligence?
“those computers those very early incredibly crude computers there's there's less than a handful in the whole world but they're capable of what seem like incredible intellectual Feats they can do huge quantities of mathematics very quickly and very accurately much more quickly and accurately than any human being could do and people start to think are these machines intelligent and that puts the idea of AI in the air”
Research on AI risks should distinguish between existential (remote, speculative) and near-term risks; funding and intellectual effort dramatically skews toward existential risk despite near-term harms (fake news, social fragmentation, surveillance) being more concrete and addressable.
“there are real risks associated with AI it tends to suck all the oxygen out of the room in in in the the phrase that my colleague used and it tends to dominate the conversation and distract us from things that we should really be talking about right”
A more imminent and real risk than the Singularity is that within one to two decades, most of what we read on social media and the internet will be AI-generated, we will not know what is real, and society will fragment as AI systems are optimized to feed users content matching their existing beliefs, while autocratic states and populist politicians exploit AI-generated disinformation.
“we are heading into a world where basically within a decade two decades think at the most um pretty much everything we read uh and see on social media and the internet is going to be AI generated and we're not going to know what's real and what isn't real in that world and there are many many risks associated with that that Society just fragments because there is no Common Core of beliefs anymore that we're all obsessed with some particular issue and that social media and the Internet is just driving us around that one particular issue because AI is programmed to pick up on the issues that you care about and to feed you stories emphasizing those risks”
Current research is trying to determine what specific role large language models should play in AI agent architectures—whether they handle only natural language and human interaction, or can be leveraged for problem-solving and planning, an open architectural question at the frontier of research.
“we have large language models and they are not sort of full general intelligence but they are nevertheless very capable how do we actually deploy those in our agents do they just handle the natural language part the conversational part or could we actually leverage them to do problem solving or things like that”
Rodney Brooks questioned the fundamental symbolic AI principles (knowledge and reasoning) and proposed instead that intelligence emerges from the interaction of multiple simple behaviors, many genetically hardwired through evolution; he emphasized embodied intelligence and reactivity over abstract reasoning.
“Brooks questioned the fundamental principles on which AI had been working since the 1950s for 30 years and those principles were that uh intelligence uh can be solved through a process of symbolic reasoning...Brooks said actually I I just don't think that's how intelligence Works in human beings and he came up with an alternative Theory this kind of Behavioral Theory”
Agent-based AI (early 1990s) reconceived software from a passive recipient of user commands (like Microsoft Word) into an active agent cooperating with and acting on behalf of the user; this is the conceptual ancestor of modern voice assistants like Siri and Alexa.
“the software becomes an agent that's acting on your behalf that's cooperating with you working with you on the task that you set it um so it's not just the dumb recipient of instructions but it's actually now an active participant working with you”
Despite the 'bitter lesson,' there is still 'magic' in AI worth pursuing: understanding the principles governing how LLMs work, their capabilities and limitations, and the fundamental laws underlying these systems is now practical experimental science rather than pure philosophy—a watershed moment.
“but there is still magic there in AI I mean so uh the fact now that we have um machines like chat GPT that we can have a conversation with that we can turn the conversation to anything that we might care to imagine compared to where we were five years ago that is simply astonishing”
Large language models like GPT-3 and GPT-4 are trained on 'all the digital content in the world' and can discuss any topic—quantum mechanics, history, recipes—with impressive fluency, but this does not mean they are doing reasoning; they excel at tasks with abundant training data where low-stakes errors are tolerable, not at real-world embodied tasks like robotics.
“we've got large language models that you can have a chat about quantum mechanics the history of Christ Church College Liverpool Football Club uh you know the origins of the first world war the economic circumstances that led to the 2008 financial crisis or recipes for um uh for uh Arnold Bennett or whatever uh anything you can think of”
The distinction between strong AI (machines that really understand and experience intelligence like humans) and weak AI (machines that simulate understanding) is philosophically interesting but not practically relevant to most AI researchers, who focus on pragmatic capabilities rather than consciousness.
“I'm not terribly interested in strong AI except after a couple of glasses of wine in a in a in a chat with colleagues and I don't know very many AI researchers that really are interested in strong AI the goals of AI are much more pragmatic”
AI research in the Golden Age relied on divide-and-conquer: breaking intelligence into modular components (vision, reasoning, planning) and solving each separately, but failed to account for the fact that these components interact complexly and that simplified 'microworlds' abstracted away the real difficulty.
“they were looking at artificial versions of problems rather than real problems that is they were looking at some problem in the real world like a robotics problem and then coming up with a simplified simulation of that problem in a computer they were able to solve it in the simple simulated version but that simulated version didn't address any of the problems that were there in the real world problem um so classically in robotics people would do simulations of robots in warehouses”
The transition in computing from manual coding of algorithms to data-driven machine learning represents a paradigm shift: from 'what is the right algorithm?' to 'give us data, let machine learning sort it out'—sacrificing theoretical guarantees of correctness for practical utility.
“in Computing I genuinely believe that the world is Shifting now from a kind of an era where we were very interested in coding exact and optimal algorithms and thinking what is the right algorithm for solving this problem now it's give us the data we'll just throw it at machine learning and let machine learning sort it out do we care about exactly how it's doing it not necessarily it's just going to give us the answer but it's so many cases it turns out that the answer it's giving us is an incredibly useful one even though we sacrifice something and what we sacrifice is kind of guarantees of correctness and optimality”
Neural networks were abandoned in the 1970s and treated as homeopathic medicine through the 1990s because the computational scale required to train large networks seemed implausible; advances in GPU computing (2012) and data availability made neural networks suddenly viable at scales that made current deep learning possible.
“it is a remarkable uh change in fortunes for neural networks which I say 20 25 years ago was really regarded as kind of homeopathic medicine was in some sense not s taken very very seriously part partly because of the scale that would be required to build large neural networks and it didn't seem plausible 25 years ago that we would have computers that could process neural networks with 200 billion parameters or 500 billion parameters and yet that's that became possible because of the computer power that we have available now”
Large language models excel on tasks where massive amounts of training data exist and where errors have low consequences (e.g., bad recipes), but completely fail at tasks requiring embodied action in the real world (loading a dishwasher), creating an odd dichotomy where we have impressive conversational AI but no robots for basic household tasks.
“large language models succeed in remarkable ways and they are genuinely impressive achievements but they succeed on tasks where there are huge amounts of data available and in some sense where the consequences of what they do just doesn't really matter that much you know if you get a bad omelette recipe through chat GPT you get a bad omelet that's not the end of the world you know you build a robot that occupies the real world with human beings and it goes wrong you know it can create Havoc”
Researchers in 2000 observed a generational shift in AI: a new generation of ML researchers lacked background in philosophy and cognitive science, bringing instead probability, statistics, and economics; this reflected a fundamental change in what skills were valued in AI.
“a career in AI now demanded a background not in philosophy or cognitive science or logic but in probability statistics and economics”
The Golden Age of AI ended because researchers were solving artificial, simplified versions of problems (microworlds) rather than real-world problems; while the simplified versions were solvable, the solutions didn't transfer to real problems because they abstracted away the actual difficulties.
“people were looking at artificial versions of problems rather than real problems that is they were looking at some problem in the real world like a robotics problem and then coming up with a simplified simulation of that problem in a computer they were able to solve it in the simple simulated version but that simulated version didn't address any of the problems that were there in the real world problem um so classically in robotics people would do simulations of robots in warehouses and you'd look at a screen and you'd see a simulated robot carrying packages around and it looks very compelling you know great okay so show me the system in the real world but robots carrying a package round in a in a warehouse in the real world is nothing like the simulated version”
LLMs succeed remarkably on tasks with abundant training data where consequences are low (e.g., recipe generation); they struggle with tasks requiring real-world embodied action (e.g., a robot clearing a table and loading a dishwasher) because no LLM is good at robotic AI, and because real-world action carries real consequences.
“large language models succeed in remarkable ways and they are genuinely impressive achievements but they succeed on tasks where there are huge amounts of data available and in some sense where the consequences of what they do just doesn't really matter that much you know if you get a bad omelette recipe through chat GPT you get a bad omelet that's not the end of the world you know you build a robot that occupies the real world with human beings and it goes wrong you know it can create Havoc it can cause real harm”
The Turing Test is a pragmatic test for indistinguishability (can a human judge tell if they're talking to a machine?) rather than a test for genuine understanding or consciousness; Turing proposed it to move past philosophical debates that cannot be resolved and focus instead on observable behavior.
“touring gets frustrated because people dogmatically insist that computers will never be able to do X where X is creativity or emotion or whatever uh and he and crucially machines will never be able to understand something in the same way that a human being is so he invents the touring test”
Agent-based AI (emerging late 1980s-early 1990s, Waldridge's primary research focus) represents a synthesis: moving beyond the user-directed, reactive model of traditional software (where the user tells the system what to do) toward software agents that proactively cooperate with users on tasks and can interact with other agents.
“changing our relationship to computer software on Microsoft Word everything that happens because you make it happen you select something from a menu or click on an icon but there's only one agent in that interaction and it's you and you are just telling the machine very much like you know you're giving detailed low-level instructions to Microsoft Word somewhat like programming it right in the same kind of style and the idea that emerged uh in the end of the 1980s beginning of the 1990s and which I worked on was to change the relationship of software so that the software becomes an agent that's acting on your behalf that's cooperating with you working with you on the task that you set it”
Large language models are disembodied systems lacking real-world interaction or persistent existence; they have no awareness of the world, no ongoing experience between interactions, and lack fundamental features of embodied intelligence like sensing and manipulation.
“but you know having conversation with chat GPT you go on holiday for two weeks and leave it hanging it's not wondering where you are it's not thinking where's wridge got to or it's not getting bored or anything like that at all it's not doing anything it is just a computer program that's paused in a loop human intelligence animal intelligence is fundamentally different to that we exist in a world we're aware of the world and that's what embodiment means it's not just having a body but it's actually being tightly coupled with the world we live in”
The Transformer architecture was designed for token prediction (predicting the next word in a sequence) and is coupled with enormous compute and data; despite not being designed for logical reasoning, planning, or robotics, Transformers have proven to be a powerful general-purpose tool.
“Transformer architecture and increasing scale a lot and a lot of of data some people seem to think that the architecture is already there we've solved it with the Transformer we have what we need to go to AGI all we need is more scale what do you think is wrong about that argument so firstly let me say what we've seen in the last few years in terms of Transformer architectures and that that which were released by Google a Google lab I believe in 2017 and what they are is an architecture for token prediction and were developed in order to enable large language models”
We have entered a watershed moment in AI history where philosophical questions about intelligence, consciousness, and understanding have become experimental science rather than remaining purely theoretical debates.
“we've gone this is really genuinely I think uh a watershed moment in AI history because we've gone from a period where a lot of questions in AI were purely philosophical questions they were literally reserved for philosophers until a few years ago uh and suddenly it's experimental science you know uh are large language models conscious well let's roll up our sleeves and do some experiment and find out no by the way they're not uh but you know these are now practical handson questions and to have gone from not having anything in the world that you could apply those questions to to this being actual practical Hands-On experimental science in just a few years is mind Bing”
The brain has functional structure beyond being a homogeneous neural network; our understanding of brain organization is incomplete, but it is much better than 30 years ago, and this improved understanding may guide future AI architectures.
“the brain is not just one big homogeneous neural network even though you know it contains vast neuro multiple neural networks but it has it has some functional structure and we we understand a lot more now than we did even 30 years about ago about the functional structure of the brain but a very incomplete understanding so that's going to be one way to go but I emphasiz again you know we are great apes that have emerged through a process of billions of years of evolution”
Despite the bitter lesson of brute-force scaling, there is still 'magic' in AI—the fact that we can have conversations with systems like ChatGPT about virtually any topic, compared to five years ago when no such system existed, is astonishing and worth exploring deeply.
“but there is still magic there in AI I mean so uh the fact now that we have um machines like chat GPT that we can have a conversation with that we can turn the conversation to anything that we might care to imagine compared to where we were five years ago that is simply astonishing and you know uh if I wish I was a PhD student now and having the opportunity to explore this kind of weird new landscape of AI and to try to figure out you know what are the what are the fundamental laws that govern these systems what are the principles try to uncover the science underneath this this technology um there is still some magic there you just have to look a bit harder to find it it might be a bit better if we needed the most advanced uh math or we need to event this fancy architecture to study human brains very closely for decades um but I think I love your word sobering because that's another way to frame Melancholy or disappointing”
Waldridge wishes he was a PhD student now to explore the weird landscape of AI and discover the fundamental principles governing these systems—suggesting that despite or because of its opacity, the field offers genuine scientific opportunity.
“you know I wish I was a PhD student now and having the opportunity to explore this kind of weird new landscape of AI and to try to figure out you know what are the what are the fundamental laws that govern these systems what are the principles try to uncover the science underneath this this technology”
'The Bitter Lesson' (article by Rich Sutton) argues that progress in AI has primarily come from willingness to throw more compute and data at problems rather than from clever algorithms or scientific insights; this is a sobering lesson suggesting that brute-force scaling, not theoretical breakthroughs, drives AI progress.
“there's there's an article by this called Rich by a guy called Rich Sutton called The Bitter lesson and rich is a very renowned uh machine learning researcher and he said look the truth is we've made progress primarily in AI by uh you know some core ideas but actually the the big steps in progress we've seen and when we've been willing to throw 10 times more comput 10 times more data and so that that is a sobering lesson”
The history of AI contains overlooked techniques and paradigms from 'paths not taken' (symbolic AI, logic programming, behavioral AI, agent-based systems) that were not fundamentally wrong but came too early or were abandoned due to paradigm shifts; these deserve rescuing for insights they might contribute to current and future AI research.
“the positive pitch I would say to give even to Serious technical researchers to the history of AI is that there are methods and ways of thinking about programming artificial intelligence in general that have been overlooked in our current Paradigm that perhaps might be rescued and is Perhaps Perhaps what we need to get us to the to the next Frontier”
The 'Golden Age' of AI (1956-1974) was characterized by genuine early successes—machines could play checkers, solve mathematical problems, perform rudimentary planning—leading researchers to believe that full general intelligence was only decades away, fueling massive optimism despite solving toy problems rather than real-world ones.
“by the end of the 1950s we've got machines that can show the rudiments of intelligence that can that can plan that can do mathematics which to be frank you know would be above the typical level of the people on the street...you've got machines that can do mathematics that can that can solve problems play games”
The low probability but high impact argument for x-risk is technically sound (if singularity happened it would be catastrophic) but psychologically it attracts people for reasons beyond rational cost-benefit calculation—it provides cosmic meaning and total explanatory frameworks similar to religious apocalypticism.
“I think it's the low probability but very very highrisk argument that I mean I think most people accept that this is not tremendously plausible but if it did happen it would be the worst thing ever and so very very very high risk and when you multiply that probability by the risk then it's the argument is that it's something that you should that you should start to think about”
The singularity narrative—that machines will become superintelligent, recursively self-improve, and spiral beyond human control—is deeply implausible because it requires that we give AI autonomous control without safety guardrails, which would be irrational policy, and current AI systems show no evidence of developing independent goal formation.
“it's deeply implausible...the kind of the Terminator thing that suddenly this will spiral out of control...if you look under the hood of of how these things work and how many patches are required to hold AI together um it just it just doesn't seem terribly plausible”
General neural network laws are implausible and inappropriate because neural networks are simply mathematics (linear algebra, statistics) that is difficult to distinguish from legitimate data analysis; regulation should focus on specific use cases (surveillance, healthcare, defense, finance, education) rather than the technology itself.
“I'm concerned about some some sort of naive attempt to create a neural network law you know thou shal not use neural networks or something like that um and that's what seems to me to be implausible because neural networks under the hood are just a bit of mathematics actually not terribly complex mathematics”
The CYC project attempted to encode all human knowledge into a massive logical knowledge base, with the goal of eventually making it so sophisticated it could write its own rules; it was widely ridiculed as a cautionary tale of AI overambition and is now used as a punchline (a 'microlennart' is a unit of measuring how bogus something is).
“the psych project has a somewhat mixed place in uh in the history of of AI...Doug Lennart was a really brilliant researcher...he became convinced that that the really big problem of AI the problem of building machines which are as fully capable of human beings is simply a problem of knowledge...he convinced some funders to support his work and at one point they had kind of warehouses full of people busy encoding all of human knowledge”
Philosophical questions about intelligence (Is it reasoning or pattern matching? Is consciousness necessary? Does understanding exist?) have shifted from pure metaphysics to experimental science; we can now 'roll up our sleeves' and test predictions against LLM behavior rather than debating conceptually.
“we've gone from a period where a lot of questions in AI were purely philosophical questions they were literally reserved for philosophers until a few years ago uh and suddenly it's experimental science you know uh are large language models conscious well let's roll up our sleeves and do some experiment and find out no by the way they're not”
Large language models are disembodied systems: they have no persistent existence in the world, no goals or desires beyond responding to the current input, and no coupling with real-world consequences—making them fundamentally limited compared to embodied intelligent agents.
“but let me say what we've seen in the last few years in terms of Transformer architectures... and there is a huge range of human activities that actually at the moment are well out of the reach of Ai and those activities are activities in the real world um doing robotic AI uh is just very very hard um large language models succeed in remarkable ways and they are genuinely impressive achievements but they succeed on tasks where there are huge amounts of data available and in some sense where the consequences of what they do just doesn't really matter that much”
AI did not have a good reputation two decades ago—neural networks were regarded as a dead field, homeopathic medicine; colleagues warned researchers that working in AI would damage their careers because the field was not going to produce results.
“go back two decades and AI actually didn't have a good reputation at all I mean in science AI was viewed as kind of homeopathic medicine neural networks were regarded as a dead field a dead end and I can remember colleagues saying you know why are you working in AI you know this is uh this this is not a field that's going to be good for your career”
The singularity narrative has come to dominate AI safety discourse and funding at the expense of concrete near-term risks like synthetic media, election interference, and surveillance systems that are already deployed today.
“that narrative whenever it comes up in sort of serious debate about where AI is going and what the risks are you know there are real risks associated with AI it tends to suck all the oxygen out of the room... it tends to dominate the conversation and distract us from things that we should really be talking about”
Regulating AI technology by banning neural networks is implausible and ineffective because neural networks are mathematically simple (linear algebra, statistics), making them indistinguishable from ordinary mathematics; regulation should target specific problematic uses (surveillance, military applications) rather than the underlying technology.
“what I'm concerned about is some some sort of naive attempt to create a neural network law you know thou shal not use neural networks or something like that um and that's what seems to me to be implausible because neural networks under the hood are just a bit of mathematics actually not terribly complex mathematics there's a lot of it but it's not terribly complex and so regulating that well where do you draw the line”
The Cyc project attempted to manually encode all of human civilization's knowledge into a logical knowledge base with the goal of building a system with human-level intelligence; it was based on Doug Lenat's conviction that intelligence is purely a knowledge problem with no shortcut, and that a machine given all human knowledge would eventually become capable of writing its own rules.
“the psych project has a somewhat mixed place in uh in the history of of AI so the vision this was the vision of Doug Leonard Leonard was a really brilliant researcher who really dazzled people in the early 70s with uh with his work um and uh he became convinced that that the really big problem of AI the problem of building machines which are as fully capable of human beings is simply a problem of knowledge and he said there's no shortcut to this we're just going to have to give the machine all this knowledge so uh he convinced some funders to support his work and at one point they had kind of warehouses full of people busy encoding all of human knowledge in these forms of rules”
The Golden Age of AI (1956–1974) began with extraordinary optimism because early computers could perform tasks requiring intelligence—mathematics, planning, game-playing—in ways humans found impressive; this led researchers to adopt a 'divide and conquer' strategy of decomposing intelligence into separate faculties and building systems for each.
“by the end of the 1950s we've got machines that can show the rudiments of intelligence that can that can plan that can do mathematics which to be frank you know would be above the typical level of the people on the street you know here in Oxford you know you have chances of getting somebody who could uh who could tell you what goal back's conjecture was or something like that um would would be limited so you've got machines that can do mathematics that can that can solve problems play games and so there is this real excitement that you know actually we're going to be very quickly making progress towards something like full general intelligence and it's called the Golden Age because you know we went from having nothing to having machines that could do those things”
We are currently at a paradigm shift moment in computing: the field is shifting from an era focused on coding exact and optimal algorithms to solve specific problems toward a data-driven era where we provide vast data and compute to machine learning systems and accept that we sacrifice guarantees of correctness and optimality in exchange for practical power.
“in Computing I genuinely believe that the world is Shifting now from a kind of an era where we were very interested in coding exact and optimal algorithms and thinking what is the right algorithm for solving this problem now it's give us the data we'll just throw it at machine learning and let machine learning sort it out do we care about exactly how it's doing it not necessarily it's just going to give us the answer but it's so many cases it turns out that the answer it's giving us is an incredibly useful one even though we sacrifice something and what we sacrifice is kind of guarantees of correctness and optimality”
AI had a poor reputation two decades ago (around 2003): it was viewed as 'homeopathic medicine' and neural networks were regarded as a dead field; colleagues would discourage younger researchers from working in AI as a career risk—a situation that has transformed completely.
“I can remember colleagues saying you know why are you working in AI you know this is uh this this is not a field that's going to be good for your career it's just extraordinary how much that's changed”
Logic programming (a paradigm using languages like Prolog) offered an elegant approach to knowledge-based AI: instead of procedural rules, developers would express knowledge as logical facts and predicates, then rely on built-in logical reasoners to derive conclusions—a seductive idea that proved inefficient and unsuitable for real-world problems like robotics.
“the idea of logic programming takes symbolic Ai and knowledge-based AI one step further and it says that okay if we want to build machines that have knowledge the way that we give them that knowledge is by expressing that in the form of logic we give them these these logical these logical descriptions of the world and this is Aristotle essentially right if if Socrates is a man all men are mortal Socrates is Mortal this is Aristotelian logic 101 exactly so we give it but we give it all of all of the knowledge about a particular problem whether it's diagnosing blood diseases or solving planning problems and so on we express that in a logical form and then inbuilt logical reasoners will sort out the details for us they will they will do the logical reasoning”
The Cyc project's main historical role is as an extreme example of AI hype that publicly failed to deliver on ambitious promises; it has become a unit of measurement for bogosity in computing folklore ('a microLenat = a unit of how bogus something is'), though its knowledge graphs eventually contributed to search engine technology.
“psych's main role in AI history is an extreme example of AI hype which very publicly failed to live up to the Grand predictions that were made for it the founder of Psych Doug lenn's role in AI has been mythologized in a piece of computing folklore a mik micr lenit so the joke goes is the scientific unit for measuring how bogus something is why a microl lenit because nothing could be as bogus as a whole L”
Machine learning and neural networks grew historically as a separate field starting in the 1940s alongside symbolic AI, with distinct milestones: connectionism in the 1980s (the idea of simulating brain structure with computation), backpropagation (enabling multi-layer networks), deep learning (more layers and scale in the 2000s), and Transformers/foundation models in the 2020s.
“what I want to move on to the last part of our conversation which is I focused most of our time talking about the history on the symbolic AI side because that I I feel like it's almost a forgotten history at this point because when we think AI we think ML and not the symbolic explicit programming side um I just want to trace out and round out this history for for our viewers because what was fascinating to me was that AI people didn't use to associate AI with ml in fact machine learning it seemed from your book grew as a separate field starting in the 40s right this idea of can we recreate the brain structure with uh electric neurons with computation and then the big Milestones Connection connectionism in the 1980s this is when we figured out back propagation basically a way to add more layers to to actually simulate to train these networks deep learning even more layers and more scale in 2000s and eventually Transformers Foundation models in the 2020s”
In the second wave of AI (1980s), the dominant paradigm was rule-based expert systems based on the belief that intelligence is primarily a problem of knowledge; developers extracted knowledge from human experts and encoded it as discrete rules (e.g., 'if temperature > X and blood test = negative then Lassa fever with probability 0.7').
“the big idea in the second wave AI is that intelligence is primarily a problem of knowledge and so if you want to build a machine that can do something for you translate from French to English or to play chess or whatever then the key problem is to figure out what knowledge the human beings use when they do that task and give that knowledge to a machine uh that was the big idea knowledge knowledge is the key to intelligence um and the AI is primarily a problem of giving machines the right knowledge”
Roomba vacuum robots exemplify Brooks's behavioral AI ideas: they are embodied systems that navigate space through reactive behaviors (move forward, detect obstacle, turn random amount) without top-down search or explicit world modeling, demonstrating that simple behaviors can accomplish useful real-world tasks.
“by by the mid 90s I think but what Brooks did is he was able to build successful robotic systems famously the Rumba robots uh are built using a version of right his ideas and I think uh that is the best example of all the ideas that he's talking about right so so the the the vacuum robots essentially uh that go around in your house they're embodied they're Rob robots they're not just a software but importantly if you look at the programming behind the robots it's not like a top- down search and go through the entire space it's kind of like go straight if there's an obstacle to take a random number turn this amount of degrees and map out the space it's it's very reactive right”
The real risks from AI in the next one to two decades are not existential but tractable: AI-generated content will dominate the internet making it impossible to distinguish real from fake, societies will fragment around algorithmic echo chambers, and autocratic states will use AI-generated fake news to drown out truth.
“we are heading into a world where basically within a decade two decades think at the most um pretty much everything we read uh and see on social media and the internet is going to be AI generated and we're not going to know what's real and what isn't real in that world”
Multi-agent systems (Waldridge's main research focus) extend the agent concept: multiple AI agents could cooperate, negotiate, and delegate tasks to each other (e.g., my Siri could directly contact your Siri to arrange a meeting rather than me calling you).
“if I want to arrange a meeting with you why would I call you why would my Siri call you why doesn't my Siri just talk directly to your Siri that is the idea of what's called multi-agent systems”
Rodney Brooks challenged the foundational assumptions of symbolic AI (that intelligence requires explicit knowledge representation and reasoning) by proposing that intelligence emerges from layers of reactive behaviors evolved or learned to interact directly with the environment, not from top-down symbolic reasoning.
“Brooks questioned the fundamental principles on which AI had been working since the 1950s for 30 years and those principles were that uh intelligence uh can be solved through a process of symbolic reasoning that we give the machine the knowledge it needs to solve a problem and that those are the key components of intelligence Brooks said actually I I just don't think that's how intelligence Works in human beings and he came up with an alternative Theory this kind of Behavioral Theory and roughly speaking what he said is we are a mass of conflicting behaviors”
Agent-based AI, developed starting in the late 1980s/early 1990s, reconceives software as active agents that pursue goals on behalf of users and cooperate with other agents, rather than passive tools that respond only to direct user commands—a paradigm distinct from both symbolic AI and behavioral AI.
“the idea that emerged uh in the end of the 1980s beginning of the 1990s and which I worked on was to change the relationship of software so that the software becomes an agent that's acting on your behalf that's cooperating with you working with you on the task that you set it um so it's not just the dumb recipient of instructions but it's actually now an active participant working with you”
Rich Sutton's 'The Bitter Lesson' articulates that AI progress comes primarily from willingness to throw more compute and more data at problems, not from scientific breakthroughs or clever architecture—this is a sobering lesson about the actual drivers of AI advancement.
“there's an article by this called Rich by a guy called Rich Sutton called The Bitter lesson and rich is a very renowned uh machine learning researcher and he said look the truth is we've made progress primarily in AI by uh you know some core ideas but actually the the big steps in progress we've seen and when we've been willing to throw 10 times more comput 10 times more data”
The Transformer architecture was designed specifically for token prediction (next-word prediction in language models) by Google researchers in 2017; its remarkable capabilities emerged when coupled with massive data and compute, but this does not mean Transformers are the key ingredient for other AI tasks like logical reasoning or robotics.
“the Transformer architectures and that that which were released by Google a Google lab I believe in 2017 and what they are is an architecture for token prediction and were developed in order to enable large language models so that you could give a prompt and they could predict essentially what should what should come next”
Science is not orderly progress from ignorance to truth—it is messy, full of false turns, and involves ideological crusades that resemble religious movements, as demonstrated by the history of AI's repeated boom-bust cycles and paradigm shifts.
“I think one point in the book I say you know if you if you think that science is about orderly progress from ignorance to truth absolutely is not it's messy false turns um almost kind of like ideological Crusades I mean and it really is ideology religious this rounds in a full circle the apocalyptic uh mentality of the of the ex risk people”
LLM planning is better characterized as pattern matching than genuine planning: when facing a problem (e.g., planning a trip), the model recognizes it as similar to thousands of trip-planning guides in its training data and uses those patterns to generate a plausible response.
“when it can when it's looking at planning a trip you've seen thousands of trip planning guides and trip agendas and so on and it's doing patent matching to pick up on that uh and help you plan the trip but is it actually planning from P first principles how to organize those various actions to plan the trip right so at the moment I say at the moment uh the weight of evidence is that it's not capable of doing uh logical reasoning or problem solving those kinds of things not in a deep way”
Roombas, the autonomous vacuum robots, are the most successful real-world embodiment of Brooks' behavioral AI ideas, using simple reactive rules (go straight, turn random angle on obstacle) rather than top-down planning or symbolic reasoning.
“famously the Rumba robots uh are built using a version of right his ideas and I think uh that is the best example of all the ideas that he's talking about right so so the the the vacuum robots essentially uh that go around in your house they're embodied they're Rob robots they're not just a software but importantly if you look at the programming behind the robots it's not like a top- down search and go through the entire space it's kind of like go straight if there's an obstacle to take a random number turn this amount of degrees and map out the space it's it's very reactive right”
The near-term risk of AI is not recursive superintelligence but rather pervasive AI-generated misinformation: within a decade or two, most internet content will be AI-generated, creating an epistemic crisis where people can't distinguish truth from falsehood and lose trust in all information.
“we are heading into a world where basically within a decade two decades think at the most um pretty much everything we read uh and see on social media and the internet is going to be AI generated and we're not going to know what's real and what isn't real in that world and there are many many risks associated with that”
Moral responsibility for AI systems must rest with the humans who build and deploy them, not with the machines; the risk is that humans will abdicate moral accountability by claiming the machine made the choice, particularly in military contexts where autonomous weapons could commit atrocities without human culpability.
“what I want is not moral AI I think it's moral human beings and it's the people that build and deploy the AI where the responsibility and the ethical considerations have to sit and they are the ones that we need to hold to account for the actions of the machines that they deploy”
Neural networks are a 'hack'—an engineering solution that works well at scale but is not based on deep cognitive science or philosophical theories of intelligence; they were inspired by brain structure but are not faithful recreations of how brains actually work.
“without wishing to denigrate these systems at all there is a very real sense in which they are a hack they are an engineering hack that's put together they are not following some deep model of mind or some deep philosophical theory about how human intelligence is or or some deep cognitive science theory of human intelligence um they are a technological hack”
Consciousness in artificial systems is not inherently morally relevant except insofar as it might make them subjects of moral concern; the stronger question (whether we owe moral consideration to machines) should not be confused with the weaker question (whether machines can be moral agents).
“for me the only reason the consciousness of a uh a Computing machine uh has or does not have um the only real concern for me is is whether we have to treat them as moral agents right if you think that a a machine might be suffering it doesn't want you to turn it off we might have to give some weight to that but that seems to be like the only possible reason why someone would be interested in in strong versus weak right”
The history of computing suggests that the future of AI will necessarily involve multiple AI systems interacting with each other, not isolated systems; this follows from patterns in how computing has evolved toward networked, distributed architectures.
“I think it is inevitable one way or another I don't I think absolutely the history of computing tells us that this surely must all the lessons that we we learn from the history of computing point to the future of AI being not just one big isolated system but multiple AI systems interacting with one another because that's how Computing the history of computing has gone”
Science is not orderly progress from ignorance to truth but rather messy, involving false turns, ideological crusades, and what amount to religious commitments (paralleling the apocalyptic mentality of x-risk advocates).
“if you think that science is about orderly progress from ignorance to truth absolutely is not it's messy false turns um almost kind of like ideological Crusades I mean and it really is ideology religious this rounds in a full circle the apocalyptic uh mentality of the of the ex risk people”
We are at a paradigm shift moment where the world is transitioning from pre-GPT to post-GPT era—this is comparable to historical paradigm shifts in science like Kuhn described, where fundamental ways of thinking change.
“I think that's right I think we are to use Coon's phrase we are at a paradigm shift moment there's preg GPT and post GPT it's been boiling up for a decade or more really”
The singularity narrative—that machines will become superintelligent and recursively improve themselves—is deeply implausible because it requires us to voluntarily give AI control over critical systems without guard rails, which would be irrational policy.
“it's deeply implausible and I became frustrated with that narrative for all sorts of reasons one of which is that that narrative whenever it comes up in sort of serious debate about where AI is going and what the risks are you know there are real risks associated with AI it tends to suck all the oxygen out of the room”
Arithmetic is arguably now solved as a capability for large language models (they can reliably do arithmetic calculations), but the question of whether they're really 'doing arithmetic' (original calculation) versus recognizing patterns from training data (arithmetic problem patterns) remains ambiguous.
“can it can it really do arithmetic or can it do something that looks like arithmetic I mean there is a big question mark around whether um what large language models are doing is doing those things or whether they're doing something that looks like patent recognition so arithmetic I'll concede you probably now is is a solved problem”
The Singularity narrative is deeply implausible because it requires machines to become superintelligent, recursively self-improve in uncontrolled ways, and escape human oversight—but for AI to harm us it must first be empowered with control and guardrails, which would be irrational for humans to provide.
“it's deeply implausible and I became frustrated with that narrative for all sorts of reasons one of which is that that narrative whenever it comes up in sort of serious debate about where AI is going and what the risks are you know there are real risks associated with AI it tends to suck all the oxygen out of the room”
Brooks built robots using a layered behavior architecture starting with obstacle avoidance, then adding exploration, then trash-finding; the approach worked in real robotics (exemplified by Roomba robots) but hit limits when trying to organize many behaviors and reason about their interactions.
“he built an architecture a framework for doing this where you would start with the most fundamental behaviors...the most basic behaviors imaginable and in robotics famously the most fundamental behavior that you learn on day one of any robotics course is obstacle avoidance...you start out by building your very first layer is obstacle avoidance and then imagine a robot that's going to go around this room picking up trash the next level of in Behavior might be exploring right”
The 'microworld' problem in early AI: researchers built simplified simulated environments (e.g., simulated robot warehouses) where their systems worked well, but failed to transfer solutions to real-world problems because they had abstracted away all the difficult complexity.
“people were looking at artificial versions of problems rather than real problems...they were looking at some problem in the real world like a robotics problem and then coming up with a simplified simulation of that problem in a computer they were able to solve it in the simple simulated version but that simulated version didn't address any of the problems that were there in the real world problem”
Current AI research focus on mapping the capabilities and limitations of large language models—what they reliably can and cannot do—is one of the most important areas of science right now; models behave in 'weird ways' where small prompt changes yield dramatically different outputs, making this characterization difficult.
“enormous numbers of people in the AI Community are grappling with is is trying to get to grips with the capabilities of large language models to really map out what these models can reliably do and what they can't reliably do”
AI history teaches that it is very easy to get over-excited and read too much into breakthroughs; people have repeatedly claimed to have found the 'magic ingredient' (search, deep learning alone, now Transformers) that will solve AI, but the field has repeatedly discovered that additional unknown ingredients are necessary.
“it's very easy to get OV excited and to read too much into what you're seeing in Ai and people have done that on multiple occasions in the past”
Many in the AI research community were surprised by how capable large language models turned out to be—researchers who claim they accurately predicted LLM capabilities are being misleading, as the actual performance exceeded reasonable prior expectations.
“honestly AI researchers that tell you that they were not surprised by how good it was I think is is misleading you a little bit they are genuinely remarkable they took me by surprise I didn't expect how good they were going to be”
The divide between symbolic AI (modeling the mind through explicit rules) and neural networks (modeling the brain through learned weights) represents a fundamental choice in AI methodology; current success of neural networks doesn't prove the symbolic approach was wrong, only that neural scaling happened first.
“historically AI has adopted one of two main approaches to this problem put crudely the first possibility involves trying to model the mind the alternative is to model the brain to model the mind is what we've been talking about as symbolic AI to give it explicit instructions of what to do to to model the processes that we rationally consciously go through in our heads to model the brain that's machine learning that's the neural Nets”
Rodney Brooks proposed behavioral AI in the late 1980s as a reaction against symbolic AI, arguing that intelligence is not primarily symbolic reasoning or knowledge but rather emerges from the interaction of many conflicting, embodied behaviors (some hardwired, some learned), organized in layers from simple obstacle avoidance upward.
“Brooks questioned the fundamental principles on which AI had been working since the 1950s for 30 years and those principles were that uh intelligence uh can be solved through a process of symbolic reasoning that we give the machine the knowledge it needs to solve a problem and that those are the key components of intelligence Brooks said actually I I just don't think that's how intelligence Works in human beings and he came up with an alternative Theory this kind of Behavioral Theory and roughly speaking what he said is we are a mass of conflicting behaviors that some of which are genetically hardwired into us through through evolutionary processes some of which we learn throughout our lives but we're just a mass of these behaviors and somehow uh human uh human intelligence arises from the interaction of those behaviors”
Emergence is a real phenomenon in intelligence—we don't understand how the electrochemical processes in brains give rise to experience, consciousness, and behavior—but current neural networks are nevertheless engineering hacks rather than faithful models of how the brain actually works.
“however the point you make about emergence I think is an entirely valid one we don't understand how intelligence emerges in human beings how does all that gooey stuff in our heads all those electrochemical processes and so on give rise to you and me we don't understand that in a deep way at all and that's what's so exciting about the present time that let's roll up our sleeves and find out how it's actually doing this but without wishing to denigrate these systems at all there is a very real sense in which they are a hack they are an engineering hack that's put together they are not following some deep model of mind or some deep philosophical theory about how human intelligence is or or some deep cognitive science theory of human intelligence um they are a technological hack um and although neural networks artificial neural networks were inspired by the structures we see in human and animal brains they are not an attempt to Faithfully recreate that”
Humans are not simply neural networks or next-token predictors; human intelligence evolved for millions of years to inhabit Earth, interact with other humans, and understand the physics and dynamics of the physical world—capacities that Transformer architectures do not embody.
“I don't think humans are a Transformer architecture I don't think that's what we're doing I think there's a lot lot more that's going on we are animals that have evolved to inhabit planet Earth and to interact with other human beings and to understand the fundamentals of human nature I think you have to understand those two things Transformer architectures are not that not by a long long long way”
LLMs are not reasoning or solving problems from first principles; when you change the terminology in a problem statement to words the model has never seen in training, its performance collapses, suggesting it performs pattern matching on familiar phrasings rather than abstracting underlying problem structure.
“there's a huge body of work looking at whether these things can actually solve problems that are not just variations of something they've already seen in their training data and the question of is it really originally solving a problem versus just doing patent recognition at the moment that's one of the big questions and the jury is very much out on that”
The Transformer architecture was designed for next-word prediction and does not obviously generalize to logical reasoning, robotic AI, or embodied interaction; there is no reason to believe that increasing scale alone will overcome these architectural limitations.
“so is that the issue that you're that you're just is that the deeper issue you're gesturing at so that's what Transformers were designed for next word prediction and the surprising thing was how useful and impressive that turned out to be”
When LLMs appear to plan trips successfully, they are pattern-matching against thousands of trip guides and itineraries in their training data, not organizing actions from first principles—the evidence is that when you use novel terminology for the same problem structure, they cannot solve it.
“when it's looking at planning a trip you've seen thousands of trip planning guides and trip agendas and so on and it's doing patent matching to pick up on that uh and help you plan the trip but is it actually planning from P first principles how to organize those various actions to plan the trip right so at the moment I say at the moment uh the weight of evidence is that it's not capable of doing uh logical reasoning or problem solving those kinds of things not in a deep way”
Alan Turing solved the Entscheidungsproblem (decision problem) by inventing the Turing machine as a mathematical abstraction, and this work on automated computation later led him to realize machines could actually be built; thus computers were invented as an incidental byproduct of solving a pure mathematics problem, not as an original goal.
“in the 1930s he's doing his PhD um in uh in Cambridge and there's one of the big mathematical problems of the age the en shidong problem it's called the translates as the decision problem...churing set himself the task of attacking the idun's problem and solved it very very quickly but to solve it he invented a kind of mathematical machine”
The Golden Age approach of 'divide and conquer'—splitting intelligence into separate faculties and building search algorithms for each—hit a fundamental ceiling: most real-world problems are NP-complete or harder, meaning no efficient algorithm exists, and exhaustive search becomes computationally impossible at realistic scales.
“for the traveling salesman problem it belongs to a class of computational problem that's called NP complete now what that means roughly speaking is that we don't have any efficient way to do it there is no more efficient way than looking through all of the candidate Solutions”
The second wave of AI (expert systems, 1980s) was based on the principle that intelligence is primarily a problem of knowledge; the key was to extract domain expertise from human experts and encode it as rules, then use those rules to make decisions (exemplified by the MYCIN system for diagnosing blood diseases).
“the big idea in the second wave AI is that intelligence is primarily a problem of knowledge and so if you want to build a machine that can do something for you translate from French to English or to play chess or whatever then the key problem is to figure out what knowledge the human beings use when they do that task and give that knowledge to a machine”
Logic programming (exemplified by the WARPLAN system in Prolog) was theoretically elegant—the idea was to express all knowledge as logical statements and let automated deduction handle inference; WARPLAN solved complex planning problems in 15 lines of code versus thousands of lines required in imperative languages.
“the war plan planning system written by David Warren in 1974 which could solve planning problems including the blocks world...and far beyond that required just a 100 lines of prologue code”
The Golden Age of AI (1956–1974) succeeded in creating machines that could play chess, solve planning problems, and perform mathematics within a decade, moving from near-zero capability to crude but genuine intelligence—but this success bred overconfidence that progress would be swift to general intelligence.
“by the end of the 1950s we've got machines that can show the rudiments of intelligence that can that can plan that can do mathematics which to be frank you know would be above the typical level of the people on the street you know here in Oxford you know you have chances of getting somebody who could uh who could tell you what goal back's conjecture was or something like that um would would be limited so you've got machines that can do mathematics that can that can solve problems play games”
Problems in AI such as traveling salesman, blocks world, and most reasoning tasks belong to the NP-complete complexity class, meaning no known efficient algorithm exists—you must examine all exponentially many candidate solutions, creating a combinatorial explosion barrier that symbolic search approaches cannot overcome.
“so for example if there are something like 70 cities there would be more possible candidate Solutions than there are atoms in the universe you will never have a computer that could exhaustively look through all of those candidate Solutions”
NP-completeness hit AI researchers with a theoretical ceiling: many problems in search, reasoning, and computer vision are NP-complete or worse, meaning there is no known efficient algorithm to solve them—one must exhaustively search all candidate solutions, which is computationally infeasible for problems with many variables.
“a lot of approaches to AI in uh in the early days involved something called search and search just means if you're given a particular problem just look through all possible candidate Solutions so uh we mentioned the idea of the traveling salesman problem the the traveling salesman problem you're given a particular map and that the salesman so to speak has to visit a whole bunch of cities on this map and return to return to base can the salesman do that on a certain budget of fuel um that's the traveling salesman problem and so one way to approach that is just to look through all the possible candidate Solutions the problem is that the number of candidate Solutions in that case just grows astronomically um so for example if there are something like 70 cities there would be more possible candidate Solutions than there are atoms in the universe”
Perhaps humans are not primarily logical or rational creatures engaging in first-principles reasoning, but rather pattern-matching systems like neural networks; if this is true, the success of neural networks over symbolic AI suggests we have been overestimating human rationality.
“maybe what humans are doing is not first principles thinking maybe we're we're just pattern pattern matching maybe it's all pattern matching down there maybe I mean I I believe I don't really believe this one but maybe we are just doing next word production when we're having a conversation and there is an entirely serious uh school of thought that thinks actually perhaps we need to rethink what the the what humans are doing and that actually that we have overblown expectations about what beliefs about what we're what we're doing”
The Transformer architecture (released by Google in 2017) was designed for token prediction—given a prompt, predict the next word—and represents a paradigm shift from rule-based to data-driven AI, from exact/optimal algorithms to statistical pattern matching on massive datasets.
“there's a Transformer architecture 2020 there's gpt3 those are the kind of moments but we are in a paradigm shift right now and in Computing I genuinely believe that the world is Shifting now from a kind of an era where we were very interested in coding exact and optimal algorithms...now it's give us the data we'll just throw it at machine learning and let machine learning sort it out”
Expert systems in the 1980s treated intelligence as primarily a knowledge problem: if you capture expert human knowledge in rules (if-then statements) and encode them in a machine, it can perform expert-level tasks in narrow domains like medical diagnosis.
“the big idea in the second wave AI is that intelligence is primarily a problem of knowledge and so if you want to build a machine that can do something for you translate from French to English or to play chess or whatever then the key problem is to figure out what knowledge the human beings use when they do that task and give that knowledge to a machine”
The shift from exact/optimal algorithms to data-driven statistical approximation in AI is a 'sobering' or even 'depressing' lesson: the chief source of progress is not scientific insight but brute-force increases in compute and data (the 'Bitter Lesson' in AI research).
“oh that is a that is a that's a very depressing lesson I mean the fact that you know you would think we the chief source of advances in AI is scientific developments actually no it's just more compute more data”
The psychology driving existential risk narratives resembles Christian apocalyptic patterns—people are attracted not primarily by rational probability calculations but by a desire to grasp something 'total and eschatological' that orients the entire world, much as millennial Christianity and climate risk narratives do.
“I study religious history and when I talk to people in the exis world the psychology kind of reminds me of uh the the Christian apocalyptic...there's these people throughout Christian history that are like Now's the Time you know this happened most recently probably when we were uh going through the Millennium right 1999 and it's this psychological drive that wants to grab at something total and eschatological in a way to orient the entire world”
The interviewer suggests that neural networks might reveal that human reasoning is not first-principles thinking but rather pattern matching like LLMs, and that consciousness and explicit reasoning might be less central to human intelligence than we assume—inverting the critique that LLMs lack reasoning by questioning whether humans truly reason in the way philosophers assume.
“maybe by by just IM imitating that even though they're not designed for for logical reasoning because we've imitated that the structure of the human brain that it's this emergent phenomenon know maybe what humans are doing is not first principles thinking maybe we're we're just pattern matching maybe it's all pattern matching down there maybe I mean I I believe I don't really believe this one but maybe we are just doing next word production when we're having a conversation”
It is unclear whether neural networks as currently designed are the right substrate for logical reasoning and abstract problem-solving; these capabilities might require different architectures or additional mechanisms not present in Transformers.
“I don't think that's what it was designed for but iiz that doesn't mean it's not useful and I'm as dazzled as anybody when I use when I use this technology”
LLM behavior is weird and unpredictable: changes in prompting that seem innocuous can produce completely different outputs, making it difficult to extract stable principles or rules from LLM behavior.
“it's also really frustrating because these models frankly behave in slightly weird ways you think you've got some principle or some rule one day and then you just change your prompt slightly in ways that seem innocuous to you and you get a completely different answer the next day and it's uh okay so what went on there what how why did it change”
Humans are not simply neural networks; we are great apes evolved through billions of years of evolution to inhabit Earth at sea level, learning the physics of our world and coordinating with other humans; embodiment and evolutionary history are fundamental to understanding human intelligence, not reducible to artificial neural networks.
“I don't think humans are a Transformer architecture...we are animals that have evolved to inhabit planet Earth and to interact with other human beings and to understand the fundamentals of human nature I think you have to understand those two things”
The debate around whether large language models can actually do arithmetic shifted recently—GPT-4 can perform arithmetic at a higher success rate—but the fundamental question remains: is the model solving arithmetic from first principles or recognizing patterns of arithmetic from its training data?
“can it really do arithmetic or can it do something that looks like arithmetic I mean there is a big question mark around whether um what large language models are doing is doing those things or whether they're doing something that looks like patent recognition so arithmetic I'll concede you probably now is is a solved problem”
Humans are not simply large Transformer neural networks; human intelligence evolved over billions of years to inhabit planet Earth at ground level, interact with other humans, learn physics, and solve embodied problems—all dimensions that Transformers do not engage.
“I don't think humans are a Transformer architecture I don't think that's what we're doing I think there's a lot lot more that's going on we are animals that have evolved to inhabit planet Earth and to interact with other human beings and to understand the fundamentals of human nature I think you have to understand those two things”
Biomimicry—studying brain structure and neural organization to inspire new AI architectures—is one promising direction for future progress, though we still have incomplete understanding of how brains are functionally organized.
“given how much success we've had about imitating a specific structure of brain right how how neurons are are linked together in in computation should we be looking more into biomimicry and should we be studying the brain more and see if there's other structures we can replicate is that the path forward to finding out the architecture is to take us to to I think that's one that's one way forward and I think we will surely get some insights”
The Turing Test proposes that if a machine can be indistinguishable from a human through conversation, the philosophical question of whether it 'really understands' becomes moot; the test collapses metaphysical questions into empirical phenomenological ones.
“maturing test says if you cannot reliably tell the difference that is if this machine can effectively pass itself off as a human being then stop arguing about it there's no point in arguing about it because you cannot distinguish between what the machine is doing or what a human does by any reasonable test”
The first AI winter (early 1970s) occurred because research stalled due to combinatorial explosion, limited computational resources, and the gap between microworld demonstrations and real-world capabilities, leading to loss of funding and reputation.
“by the early 1970s it becomes clear really that progress is stalled and there are lots of reasons why progress stalled one of the reasons that progress stalled is people were looking at artificial versions of problems rather than real problems”
The Cyc project attempted to manually encode all of human civilization's knowledge into logical rules, based on the belief that intelligence is purely a knowledge problem and that explicit logical deduction would scale indefinitely—but it failed and never delivered at the anticipated scale.
“the vision this was the vision of Doug Leonard Leonard was a really brilliant researcher who really dazzled people in the early 70s with uh with his work um and uh he became convinced that that the really big problem of AI the problem of building machines which are as fully capable of human beings is simply a problem of knowledge and he said there's no shortcut to this we're just going to have to give the machine all this knowledge”
Consciousness is a private, subjective experience of the world from a personal perspective; nobody understands how consciousness arises from physical processes, but we do broadly agree that subjective experience and qualia are defining features of consciousness.
“one of the fundamental components of human beings is that we have experiences we experience the world um that's you know nobody really understands what Consciousness is but roughly speaking people agree agree that that that ability to experience things from a personal perspective and that your personal perspective is private and unique to you and I can imagine what you're experiencing but it really is private and unique to you”
The stoic philosophical tradition held that humans are primarily rational creatures driven by implicit propositions about what is good, while Freud argued our unconscious drives are largely unknowable to us—the success of opaque neural networks over explicit systems suggests perhaps Freud's view is more accurate than the stoic rationalist view.
“I'm preparing a lecture on the stoics right now and the stoics famously think that humans are extremely rational creatures that even unbeknownst to us when I desire something I'm making an implicit proposition that that thing is good behind most human behaviors there's an explicit or sorry there's an implicit uh true or false proposition someone on the opposite extreme is probably someone like Freud where our unconscious is not known and perhaps even greatly unknowable to us”
The Stoic view of humans as fundamentally rational beings (every action backed by implicit propositions about the good) contrasts sharply with Freud's view of the unconscious as primary and unknowable; the success of neural network black boxes over explicit symbolic systems suggests human intelligence may be more Freudian than Stoic—mostly unknowable to ourselves.
“I'm preparing a lecture on the stoics right now and the stoics famously think that humans are extremely rational creatures that even unbeknownst to us when I desire something I'm making an implicit proposition that that thing is good...someone on the opposite extreme is probably someone like Freud where our unconscious is not known and perhaps even greatly unknowable to us”
The agent-based paradigm is agnostic between symbolic and connectionist (neural) approaches—agents can be implemented via either explicit rules or learned representations; the paradigm abstracts over the question of whether intelligence comes from modeling the mind or modeling the brain.
“what I found fascinating about the agent Paradigm is that it almost it's agnostic to and it cuts across the symbolic modeling the the mind and modeling the brain because how it is proactive or how it is reactive or how it interacts with other agents that's you abstracted away from the type of questions you're thinking about”
The existential risk (X-risk) narrative has psychological roots in religious apocalyptic thinking—a human drive to find something total and eschatological to orient the world—not primarily a rational calculation, similar to millennial Christian fears or climate apocalypticism.
“I study religious history and when I talk to people in the exis world the psychology kind of reminds me of uh the the Christian apocalyptic that that there's these people throughout Christian history that are like Now's the Time you know this happened most recently probably when we were uh going through the Millennium right 1999 and it's this psychological drive that wants to grab at something total and eschatological”
The Turing test, while beautiful in its simplicity as a test for indistinguishability, is philosophically and practically limited because it collapses metaphysical questions (does it really understand?) into phenomenological ones (can we tell the difference?), which may not be the right equivalence.
“there's two ways to interpret the Turning test philosophically one way is to reduce metaphysics to phenomenology and this is to say look the metaphysical question of does it understand something is it really thinking is totally collapsible to the phenomenological the empirical question can we distinguish the outputs but or it could be making an epistemic point which is to say the metaphysical question doesn't really matter let's just focus on the empirical question”
Attempts to equip AI with ethical reasoning and moral capabilities are dangerous because they allow humans to abdicate moral responsibility—enabling military and civilian actors to blame machines for decisions rather than holding themselves accountable.
“there is a uh there is a body of work which is all about trying to equip AI with kind of ethical and moral reasoning um and I understand why people want to do that so that we have machines that make choices that we would want them to make what worries me about that is that it allows people to try to abdicate their moral and ethical responsibilities wasn't my fault it was the machine's fault”
Previous AI paradigms overlooked techniques and ways of thinking that may be rescued and applied today; like philosophy losing moral intuitions through paradigm shifts, AI risks discarding valuable insights from symbolic AI in the shift to machine learning.
“in philosophy there's an idea that a lot of moral intuitions good moral intuitions are lost through Paradigm shifts so we gain things in this whole Christian worldview but we moved away from the Roman world and the Pagan world and there's things to be rescued from that world that have been forgotten and my my training early training was in stem stem usually doesn't study historical stuff right you usually just study the latest physical theories but there is a view of even stem Innovation Thomas [ __ ] being the biggest proponent as being these Paradigm shifts that there are things that are important that are lost in previous paradigms”
There is a striking analogy between Cyc's approach (humans manually interpreting and coding all knowledge) and modern large language model training (exposure to all digital data without human interpretation); the key difference is that LLMs find order in raw data without explicit human encoding, through a process we don't fully understand.
“the difference is that in the site case human beings were interpreting all of that and writing coding down The rules in the computer language with large language models none of that goes on it is just presented to the model and in some sense and I'm waving my hand madly at this point in some sense it it finds order in that and how it does that actually we don't really understand as we were talking about earlier”
Attempting to regulate AI through general laws against neural networks is implausible because neural networks are mathematically indistinguishable from basic statistics and linear algebra; instead, regulation should focus on specific use cases (surveillance, healthcare, defense, finance, education) and the consequences of technology rather than the technology itself.
“what I'm concerned about is some some sort of naive attempt to create a neural network law you know thou shal not use neural networks or something like that um and that's what seems to me to be implausible because neural networks under the hood are just a bit of mathematics actually not terribly complex mathematics there's a lot of it but it's not terribly complex”
The responsibility for ethical AI lies with the humans who build and deploy AI systems, not with the AI systems themselves; attributing moral agency to machines allows humans to abdicate their ethical responsibility, which is particularly dangerous in military contexts where blame can be shifted to the machine for targeting decisions.
“what I want is not moral AI I think it's moral human beings and it's the people that build and deploy the AI where the responsibility and the ethical considerations have to sit and they are the ones that we need to hold to account for the actions of the machines that they deploy”
The agent paradigm's agnosticism regarding whether AI systems model the mind (symbolic) or model the brain (neural) is a strength: agent architecture abstracts away from these details and focuses on goal-directedness, reactivity, and social coordination, which can be implemented through multiple approaches.
“what I found fascinating about the agent Paradigm is that it almost it's agnostic to and it cuts across the symbolic modeling the the mind and modeling the brain because how it is proactive or how it is reactive or how it interacts with other agents that's you abstracted away from the type of questions you're thinking about yeah so what does multi-agent systems have to offer to the current Paradigm of foundational models now”
One current research direction in multi-agent systems is architectural decomposition of LLMs: rather than one monolithic model, train multiple specialized LLMs (one for math, one for intuition, one for creativity) and route tasks to appropriate models—which is itself a form of multi-agent work within a single LLM system.
“there's another way in which multi-agent systems I imagine are are currently being deployed even llms not multiple llms talking to each other but how you split work within one llm right so the the rough intuition is you know maybe uh it's better to train actually three hidden LMS one's good with math one's good with intuition one's good with creativity or language and then you have a a a a job sort of processing unit that gives uh the different llms different tasks to to process that that also is a type of multi-agent work”
The question of how to leverage large language models within agent-based systems remains an open research frontier: should LLMs only handle natural language and conversational aspects, or can they be used for actual problem-solving and planning? Current skepticism suggests limits on LLM problem-solving capabilities.
“we have large language models and they are not sort of full general intelligence but they are nevertheless very capable how do we actually deploy those in our agents do they just handle the natural language part the conversational part or could we actually leverage them to do problem solving or things like that now we've already talked about the idea you know can large language models solve problems and I'm a bit of a skeptic at the moment about the extent to which they can do that but how exactly do we leverage this technology in the best way possible uh is is right at The Cutting Edge of research right now”
There is melancholy in discovering what AI techniques have and haven't worked: success came not through philosophy, cognitive science, and logic (the elegant symbolic approaches) but through mundane mathematics (statistics, probability, linear algebra) at scale.
“there is I think something Melancholy about what AI techniques have worked and what haven't...a career in AI now demanded a background not in philosophy or cognitive science or logic but in probability statistics and economics”
The difference between embodied and disembodied AI is not merely whether a system has a physical body, but whether it is tightly coupled with a real-world environment, sensing consequences of its actions and adapting based on real-world feedback.
“you're having conversation with chat GPT you go on holiday for two weeks and leave it hanging it's not wondering where you are... it's not doing anything it is just a computer program that's paused in a loop human intelligence animal intelligence is fundamentally different to that we exist in a world we're aware of the world and that's what embodiment means it's not just having a body but it's actually being tightly coupled with the world we live in”
Large language models are an 'engineering hack'—a technological solution that works empirically but is not grounded in deep models of mind, cognitive science, or philosophy; they succeed through brute-force scaling, not through faithfully recreating cognition.
“there is a very real sense in which they are a hack they are an engineering hack that's put together they are not following some deep model of mind or some deep philosophical theory about how human intelligence is or or some deep cognitive science theory of human intelligence um they are a technological hack”
Despite correctly predicting that computer networks would become ubiquitous and central to the future, Waldridge completely failed to anticipate the World Wide Web, Amazon, or Google—major missed opportunities for transformational work based on not fully understanding what the networked future would actually look like.
“by the way having realized that networks were the future I completely failed to anticipate the worldwide web or Amazon or any of that I look I look at the missed opportunities in my life um for for doing transformational work I totally got that networks were going to be the future but I still didn't understand exactly what that future was going to look like”
During AI winters, being an AI researcher had the advantage of relative solitude—few people worked in the field, so a researcher could explore without intense competition; conversely, when the field boomed, the character of research changed because talented researchers converged on the same problems, creating competitive pressure rather than the freedom to explore.
“for most of the time that I've been studying AI it was a relatively quiet existence and the nice thing about that was I just got on with my thing there was very few people working in the same area and as a researcher actually that's quite a nice thing uh as a researcher having you know a big space to yourself is actually really quite sort of refreshing you can just explore the territory so when the field becomes became popular we found huge numbers of people flooding into it now the nice thing about that is huge numbers of very talented people but as a researcher what you're finding is you're no longer the only person that's looking at your problem you're surrounded by extremely capable people all working on exactly the same problem”
The history of computing shows that systems have evolved from isolated machines toward increasingly interconnected systems; this historical pattern suggests that AI will inevitably move toward multi-agent architectures where multiple AI systems interact with each other, though the exact form remains unknown.
“I think it is inevitable one way or another I don't I think absolutely the history of computing tells us that this surely must all the lessons that we we learn from the history of computing point to the future of AI being not just one big isolated system but multiple AI systems interacting with one another because that's how Computing the history of computing has gone so I absolutely believe that the problem is I don't know exactly what that's going to look like and that's what I'm trying to figure out now that's what my current research is trying to figure out”
Despite correctly predicting that networks would be ubiquitous, Waldridge completely failed to anticipate the actual form that future took—the World Wide Web, Amazon, and Google—indicating that even when you correctly identify a transformational technology, predicting its actual instantiation is nearly impossible.
“I look I look at the missed opportunities in my life um for for doing transformational work I totally got that networks were going to be the future but I still didn't understand exactly what that future was going to look like”
The existential risk narrative appeals to deeply human psychological patterns analogous to Christian apocalyptic thinking—a primal drive to latch onto eschatological narratives that orient the entire world, similar to patterns in climate risk discourse.
“I study religious history and when I talk to people in the exis world the psychology kind of reminds me of uh the the Christian apocalyptic that that there's these people throughout Christian history that are like Now's the Time you know this happened most recently probably when we were uh going through the Millennium right 1999 and it's this psychological drive that wants to grab at something total and eschatological in a way to orient the entire world so so people I guess what I'm trying to highlighting is maybe you can see some of the psychology and climate risk as well”
Despite the sobering lesson that progress comes from brute force rather than elegance, there is still magic in AI: having machines we can have conversations with that respond intelligently to any topic is simply astonishing compared to five years ago, and discovering the fundamental laws governing LLM behavior is genuinely exciting research.
“but there is still magic there in AI I mean so uh the fact now that we have um machines like chat GPT that we can have a conversation with that we can turn the conversation to anything that we might care to imagine compared to where we were five years ago that is simply astonishing”
The AI Winter (early 1970s onward) occurred because hype had created unrealistic expectations, funding dried up, researchers were portrayed as charlatans, and the public became suspicious of AI claims; this was not unique—AI has experienced multiple boom-bust cycles.
“by the mid '70s because of the hype as well as the the series of technical problems that the AI field ran into it went into its first but not certain certainly not only winter and what that means is just funding dried up interest dried up people were sometimes portrayed as charlatans”
Waldridge became interested in multi-agent systems because he realized in the 1980s that computer networks (ARPANET, Janet) would become ubiquitous, and combining networked systems with AI suggested AI systems would eventually need to communicate with each other.
“as an undergraduate in the 1980s uh I was fascinated with AI uh but I also became fascinated with computer networks and you have to remember at the time computer networks were not common you know the the the the predecessor of the internet the arpanet um developed by the advanced research projects agency in the US essentially military research funding agency had you know a very incomplete International network with just a few nodes connected in the UK but I got the opportunity to work on the UK's extension of that called Janet The Joint academic Network and I had a kind of moment of Revelation at which point I realized this is going to be the future networks they're just going to be everywhere”
The debate around AI-generated misinformation and election interference has not yet materialized at feared scale (as of 2023), but this should not lead to complacency; the risk remains very real and should be monitored carefully.
“going into elections in the US the UK I was really worried that what we were going to be see was social media drowning in AI generated fake news we didn't see that as it happens at least not on the scale that I feared it might occur um but nevertheless I wouldn't take my eye off that as a risk”
Waldridge works on multi-agent systems and is trying to understand how large language models can be deployed within agent architectures—specifically whether LLMs should handle only natural language or also problem-solving roles; this is at the cutting edge of AI research.
“what does multi-agent systems have to offer to the current Paradigm of foundational models now oh wow well this is we're really at The Cutting Edge now we're I mean this is a big research question is about we have large language models and they are not sort of full general intelligence but they are nevertheless very capable how do we actually deploy those in our agents do they just handle the natural language part the conversational part or could we actually leverage them to do problem solving or things like that”
Waldridge did not initially expect GPT-3 and current LLMs to be as capable as they turned out to be; AI researchers who claim they were unsurprised are being misleading—the phenomenological power of LLMs genuinely exceeded expert expectations.
“honestly AI researchers that tell you that they were not surprised by how good it was I think is is misleading you a little bit they are genuinely remarkable they took me by surprise I didn't expect how good they were going to be”
Working in a quiet research area during an AI winter is more pleasant for individual researchers than working in a crowded boom period, because there is intellectual space to explore without constant competition.
“for most of the time that I've been studying AI it was a relatively quiet existence and the nice thing about that was I just got on with my thing there was very few people working in the same area and as a researcher actually that's quite a nice thing uh as a researcher having you know a big space to yourself is actually really quite sort of refreshing you can just explore the territory”
If one were a PhD student now, exploring the landscape of LLM capabilities and trying to uncover the fundamental principles governing these systems would be intellectually rewarding—there is genuine scientific discovery to be had.
“you know uh if I wish I was a PhD student now and having the opportunity to explore this kind of weird new landscape of AI and to try to figure out you know what are the what are the fundamental laws that govern these systems what are the principles try to uncover the science underneath this this technology um there is still some magic there”