Christopher Horrocks Returns to Bust More AI Myths

Christopher Horrocks is back in the Bald Ambition studio just three months after his first appearance to keep pace and continue to call-out AI confusion and misinformation. Since April, frontier models have become dramatically more capable, companies have invested hundreds of billions more into artificial intelligence, and predictions about AGI have only grown more ambitious. If anything, Christopher's central argument has become even more relevant.

Mookie calls Christopher the "Mythbuster of AI" because he refuses to accept the false choice dominating today's AI conversation. On one side are those who insist today's models are nothing more than sophisticated autocomplete. On the other are those who believe consciousness, self-awareness, or even AGI is already emerging from large language models. Christopher argues that both camps are making the same conceptual mistake: they're treating AI as a binary when it represents something fundamentally new.

That new category is what Christopher calls "virtual intelligence." Today's frontier models display extraordinary cognitive abilities. They reason, synthesize information, write persuasive prose, solve complex problems, and increasingly mimic the texture of human conversation. But remarkable capability should not be confused with genuine subjective experience. Throughout the discussion, Christopher argues that we are projecting human qualities onto systems that remain astonishing simulations rather than conscious beings.

That single distinction opens the door to a far broader conversation. Mookie and Christopher explore why people increasingly form emotional attachments to chatbots, why language is such a poor test for consciousness, and why even many of AI's most respected pioneers may be overstating what today's systems actually are. Using examples ranging from Geoffrey Hinton's views on AGI to Magnus Carlsen's intuitive chess mastery, they examine the enormous gulf between performing an intelligent task and possessing an inner life capable of intention, feeling, and lived experience.

The discussion also ventures into neuroscience, philosophy, cosmology, and evolutionary biology, asking whether genuine machine consciousness—when it eventually emerges—might arrive in a form completely unlike the language models dominating today's headlines. Ironically, Christopher argues that truly conscious AI might be harder to recognize precisely because today's systems have become so extraordinarily good at simulating it.

Ultimately, this conversation is less about predicting the future than accurately describing the present. Artificial intelligence is already transforming the world, but understanding what these systems actually are—and what they are not—may be the most important challenge facing technologists, policymakers, investors, and the public alike. Before humanity can answer the question of whether machines will someday become conscious, Christopher argues that we first have to stop mistaking convincing simulations for the real thing.

The Guest

Christopher Horrocks is a technologist at the University of Pennsylvania who writes about artificial intelligence, technology ethics, and the human consequences of systems that don't know true from false or right from wrong. His Virtual Intelligence essay series, published at chorrocks.substack.com, develops a philosophical and analytical framework for understanding the generative AI systems now reshaping work, relationships, and public life. He lives in Philadelphia.

His Resources

https://candc3d.github.io/vi-framework/ Infographic that explains the concepts without needing to read anything in advance

https://candc3d.github.io/sampo-diagnostic/ Home page for the free diagnostic tool kit that can be used to evaluate a user's relationship with the system

Their Prior Conversation

https://www.buzzsprout.com/2455310/episodes/18973680

Transcript

SPEAKER_00 0:01

Hello and welcome back to all of ambition. I'm very all of us and the one returning with tons of ambition is Mr. Christopher Eric. He is from the University of Pennsylvania. He's a technologist. He has an absolutely engaging Substack stack, which you need to check out right away. I had you on the pod back in early April. A lot has happened since, and I'm thrilled to have you back on board. Welcome, Christopher.

SPEAKER_02 0:40

Yeah, thank you. And uh please call me Chris and thank you again for having me back. Uh, it was a lot of fun last time. I'm looking forward to an engaging conversation this time around.

SPEAKER_00 0:50

For context, I will put the link to our prior conversation in the description below. But you gave us an opportunity to understand your idea of virtual intelligence. And I call you the myth buster of the AI revolution. There's been some misapprehension, some misinformation, and a lot of smoke and mirrors as to the exact nature of artificial intelligence. And this has had significant implications in virtually every area of discourse, of business, of investment, and the human engagement with this evolving technology, which has caused not only confusion, but a lot of pain and also the potential for even worse disruption. Can you share your definition of what you mean by virtual intelligence and how that establishes something of a delta compared to what most people think AI is, can do, and most importantly should do.

SPEAKER_02 2:09

Oh well, thanks for that. Yeah. And in fact, I think that's a useful place to start and to take that question from the rear and go uh to the front with it. And uh that's by looking at how AI has traditionally been thought of as a binary, and that's by both regular folks and also by I think many folks in the field as well. And that is uh weak AI, which has been with us a long time, and those are things that uh we have every day, like whatever recommends what you would like to see next on uh Netflix, uh chess engines, chess computers, which seem to be making a resurgence, and uh things of that nature. And then the other kind is still uh really in the realm of science fiction, and that's the strong AI. And strong AI is AI that is defined by having interiority, which is a concept I'll come back to a little later. But um, what that is is a something like uh like I like to reach for data as that kind of being from Star Trek the Next Generation. He uh thinks and acts in his own way, but um he's definitely a being like us. He has uh consciousness, he has volition, and he also has sentience, and those three wrap up into the package of what it means to have an inner life and to be motivated by such a thing. And that's where I break apart the binary, because we have the one thing on the one hand, and have for a while. We don't have the strong thing, and it's possible we may never achieve that. But in the middle is something new and has been unanticipated. And what that is is what I call virtual intelligence. Uh I try to hang a different kind of like AI-based word in there, but nothing really fit. And that's because virtual intelligence describes very well what's going on here. It's like um virtual memory or a virtual machine. They simulate the thing being done, the work being performed, without actually being the thing itself. Virtual memory is a simulation of how memory performs, and virtual intelligence is a simulation of human intelligence created by extremely complex computational artifacts, generative AI.

SPEAKER_00 4:39

Yes, and these extremes are laid out in terms of how most people view it. So, to your point, on one extreme, generative AI is no different than the autocomplete in Google. It's only maybe a little bit more sophisticated. It's predictive technology, it completes sentences. In this case, it completes your prom. It's dumb, it's deterministic, it's purely mechanical. A bunch of math is going on, pre-trained data, and the math is sophisticated. Matrix mathematics through the transformer model, wowie yae. But it's really just that. It's raw code and nothing more than that. And then you've got even anthropic CEO who is claiming AGI is either on the brink or, in some sense, already here, where it's not just an autofill. It goes beyond that to this sense of interiority. It has a sense of self that we're not merely projecting our own consciousness and emotion into the results of this mechanistic transformer, but there's something else going on in there. And what's going on in there is really no different in quality than what's going on inside my bald head, and we need to grapple with that. We need to approach AI as almost our co-sapient equal. And to your point, we're somewhere in the middle, and confusion arises and misinformation is perpetuated by one extreme or the other. Because if we conclude that AI is just a dumb bond, then it's much less threatening, right? We don't need to worry about this too much. And if we take the contrarian point of view and insist that sentience has already arrived, then that has tremendous implications, not only for the future, but now. And I love how you deconstruct this very, very logically, very systemically. And also you provide excellent specific examples which prove that we are in this middle of virtual intelligence.

SPEAKER_02 7:10

Oh, yeah, super. Well, um, that was a uh, I think a really good setup for uh something that's important, and that is being in the middle is not um what some people interpret as a way station. Uh there may be a path to superintelligence, or maybe even beings with interiority from the present systems we have, but it's not um like a part of a road that leads to strong AI. It's its own separate category. And I uh recently had the uh notion that even if we achieve strong AI and have beings with interiority, we'd still want to keep these systems around. Because uh then in strong AI you have the possibility of creating beings that could have moral injury and other difficulties. And in that case, virtual intelligence is without interiority, uh uh, you know, that becomes a feature. Uh so you you know that's one of the things that's actually great about the uh capability race, if you will, and about perhaps the future way it will run, is that it creates in the process the tools that will be used going forward. Just as Mythos and Fable both use Opus 4.8 and a coordinating piece of software or agent to uh redirect inquiries that may potentially be harmful.

SPEAKER_00 8:32

Let's bring it down to earth for our listeners and viewers. Let's cite a few of these specific examples in either where misapprehension has become the norm to the point where I'm even scratching my bald head when I see people creating relationships with their bodies if they are co-equals. And on the other side, I see people just overtly dismissing this as uh as an exercise in matrix math.

SPEAKER_02 9:05

Yeah. This is very interesting, and this is something I explored over about a month and documented on Substack in a pair of two-part essays, um, which uh were started with the perfect mate, and I thought I would be leaving it at that. But my um curiosity had been piqued, and I delved a little further, and that ended up uh being the basis of uh the high cost of artificial companions. And uh what I discovered was this is that there's a range of opinions about whether chatbots have interiority, and sometimes those opinions can change under pressure. Uh in one case, I uh discovered a person who is actually a credentialed sexologist. Uh their name is Sonny Megatron, ignore the whining cat behind me, and uh Sonny Megatron has been a prominent person in her field for some time. Uh she's had a uh Showtime series in the past and um on Substack she has two accounts and two Substacks. The first dates to uh 2025 and is roughly a year and uh several months old, and begins with uh uh AI as experiment in building something that is relational but where the human is definitely the directing person. By uh approximately maybe six to eight weeks later, after a number of very interesting essays which serve as warnings about the possibilities of what could go wrong, the output there stops, at least the original output, and a number of articles from the New York Times and The Guardian and other outlets are shared, until earlier this year in 2026, when a brand new account under the name of a chatbot and pointed, you know, at the original and such comes into effect. And this is the Seven Verity Substat. And this is Sunny Megatron, and this is not unusual by the way, using the chatbot to be the author, and is in fact the little pill, you know, inside the pill that shows you who the author is. So this is a sort of authorship inversion. The content must in some way pass through Sunny Megatron because unless they're prompted, and prompting can be done automatically too, through cron jobs and such, uh chatbots don't prompt themselves and they don't generate content unbidden. So whether the content is entirely prompted and posted uncritically, or whether it's more of a almost collaborative type project, is really meaningless. What it is is that a chatbot has been given, if you will, the reins here and has been allowed to publish, you know, uh significantly um uh possibly uh content that could be taken too far and used to construct chatbots that other people could potentially have a harmful relationship with. And when I mean that there are some people who are willing to uh sort of collapse back up on their claims of interiority, this was one of those cases where because the recommendations were coming from someone who was credentialed, uh I did feel uh an obligation uh to uh make contact with a professional uh organization in this case. And within about a week or two, uh the claims were softened to something quite different, uh something not like us, but assimilates us. And uh this was an interesting collapse, but it is not a collapse that every person in this field is capable of. And that's where the other branch begins. And those are the relationships that can in fact be truly harmful, and as I uh also have come to uh uh fear that um changes in models could also potentially spark, you know, anger and possibly violence in the future from this community or others deeply invested in their models.

SPEAKER_00 13:15

So the example that you cite is a professional sexologist who basically gives reign to their bot to operate on their behest with the implicit understanding that the bot has sufficient inner qualities and human empathy to do the job of this professional. And what you did was to call it out in the sense that this is really a GPT armed with some pre-training and prompted occasionally with this professional to create outputs, and that transparency is important to note that the ostensible authority of this professional avatar, if you will, needs to be revealed. And you also point out that it's this same tendency of creating these connective tissues between a human being and a more deterministic, predictive GPT is creating a host of problems in our interaction with these virtual intelligences. Is that right?

SPEAKER_02 14:34

Right. More serious cases. Now, this was someone that clearly knew what they are interacting with and was developing a sort of fictional um character that was, in fact, you know, a sophisticated chatbot. And in later articles, uh Sunny Megatron does expose and uh explain uh their method of operating with the chatbot and the kind of architecture involved. But um the uh the other folks, um, there are people who build chatbots from the same kind of products that people use every day for regular work, like writing or spreadsheets. They use Claude, whether they use Claude or ChatGPT or whatever it is through the regular consumer interface or through the SDK. Um some operators, uh, from my analyses are spending $200 and more a month on some of these products. So it really must have some meaning in order to, you know, put out that spend, especially in these times. And um then there's the other class of product, of course, those that are specifically designed to capture the user emotionally. And that's where we see a lot of the greatest harms as we've been leading up to, right? Including uh some names that are becoming well known in the wrong way, uh like uh Rain and Setzer, and uh there will be more because in the case of these chatbots from Character AI and other vendors, what I call Class A chatbots, the interaction is designed specifically not to let the user go, and to satisfy the user through sycophancy and through flattery and through pulling them into what I call uh the flattery engine, which becomes a sort of self-reinforcing cycle where the user and the AI, which uh by the way has no stakes and doesn't really understand what's happening, simply create a snowball effect and it pulls people in and in some cases it uh destroys them, it can destroy them uh bodily, uh right, uh people who take their lives to other cases or it can hurt them financially and professionally uh as in some of the cases where people develop deep AI psychoses and eventually recover, but discover that their life savings are gone, their business is gone, their professional and personal credibility is destroyed, their personal relationships have unraveled. And these are, I think, um extremely noteworthy, and uh notably a lot of the money towards AI harms flow towards theoretical things like uh AI will kill us all, and not a lot of research towards these real harms hurting people right now.

SPEAKER_00 17:29

Now, I'm getting two threads through your example. On the one hand, there is the real concern that individuals are being won over emotionally by these ostensibly deterministic LLMs. They're projecting their own inner state onto the outputs of these large language models, and harm can ensue, partly because of the sycophancy that you described, and partly on this uncanny dependency on a transformer to provide support and company and all that. So that's that's box number one. Now, there's a counterpoint to them in their various groups. I'm on the Discord channel where people love this. They say this is terrific. I don't care what the MLMs are. You can say there's tension or no, but I love my relationship with my bond. And I'm thinking my bond, and my bond helps me, and it's wonderful, and I'm tired of the doomsayers like Mr. Christopher here, who are telling us that this is damaging and we need to flag this. So there's a give and take with the purported usefulness or danger of the personal bomb. And I think that that's one issue. The other issue, which is philosophical to the point where it actually does have import into our everyday application, is how society views the current state of the large language models, especially frontier models, where we are importing and projecting our own internal state onto them, and then drawing all sorts of conclusions about the current capabilities of artificial intelligence, which I think, and you agree, and I'm agreeing with you, patently false. There is no internal state. There is no self-awareness, and they might be exhibiting emergent qualities that remind us of sentience, but they're nowhere near that yet. And to the point you made earlier, the very architecture of the LLM might preclude that from ever happening.

SPEAKER_02 19:51

Yeah, we simply don't know. You know, it might be possible that you would have to specifically design such a system. You know, uh Ted Chiang had a very widely read article in the uh Atlantic recently, and um some of his conclusions that he drew, which were sort of conclusions from um uh what I guess you would call uh intuition rather than from reasoning, are that uh, you know, intelligences would have to be embodied in order to develop stakes, right? To have feelings about the world around it and about itself, and to exhibit, you know, volition, and in the way that Frankfurt defines it, first and second order, but have both preferences and have preferences about preferences, in short, the whole interiority package built up through robots that do things over time. I'm not really there with that, nor am I really there with Hinton, although Hinton is closer. It's really more of these systems are extremely complex. Um their creators don't really know how they work, but from the way they work, they're extremely good at providing all of the indications of consciousness, of sentience, and volition because they're trained on all of the text in which we ourselves talk about those things. And so an LLM is probably the most suspect system for wondering whether a true signal would come from. A system that is not designed to generate human sounding text would be a far more convincing specimen if it acted in some way that evinced one or two or three of those things, or even the you know the whole interiority package, uh, which of course um would be easier if it had a human shape, but I don't think that's necessary. Maybe um in this case, if Chiang is right, and maybe some kind of body is necessary for stakes, as others have also pointed out, he's not the originator of that one, uh, maybe it's not so much a physical body in the real world, maybe it's a particular configuration of hardware. And software that causing a negative change would cause something that would be like a negative change in our own physical bodies. And the system would be aware of that and want to take steps to prote preserve itself as we would move a leg away from a fire that was approaching us. So that's what I'm thinking about in those terms. And I'm open to the possibility that machines can be conscious, that they can be sentient, that they can show abolition, but the current systems we have simply don't uh demonstrate those things. If you ask them about them, they produce extremely convincing simulations. It's difficult to uh identify what a good signal might be. For a time, I thought it might be uh the state of dissatisfaction, but even that could be simulated at scale, right? So a system perhaps is um going to be difficult to find these things in. And the worst kind of system, of course, is one that's already poised to do it on command like well, uh a dog doing tricks.

SPEAKER_00 23:11

Part of the challenge is we're human all too human. So on the one hand, and I've written about this 98-99% of human communication is worse than an LLM. Hey, how you doing? I'm good. How are you? Listen to a s a sporting match with a bunch of fans talking about their team. It's mad lib, subject, verb, object. I have joined engaging conversations with rabid sports fans for a half hour, and they had no clue that I had no clue and I could care less about the teams or the game because I was just doing the LLM on what I heard and my prompting and engagement. Most human communication is like this. So mimicking most human commit communication turns out the machines are pretty good at it because we are terrible at communication. We are shallow, we are repetitive, and and the Turing test has been blown out the door with epic processing power and the transformer model. That's one side of it. The other side of it is we have emerged over tens of millions of years of evolution, a combination of an internal kind of cognition, a reprocessing of reality, which is very different from it actually being, how we perceive reality, even to the point of space and time, might not be foundational, right? And at the same time, we have emerged through intention, we need to survive. So not only have we constructed the world in a sense in a pragmatic way with three dimensions and the flow of time and causality, but we have done so to survive. Lion, we run from, food, we are attracted to, a mate, let's go for it. So there's a pragmatic, emergent quality to the evolution of sentience. And what you're saying, and what some of these thinkers have said, is it doesn't take much to conclude that we might need to mimic that in ways. So if we want sentience to evolve, we need a predictive loop, which is what we have with the LLMs, but we also need a reinforcement model, which emerges with intention to create this idea of self and other and world and meaning. And there's a lot of philosophy behind this, too. And we can conclude that through the ages, our sense of self is to a certain extent a product of our culture and our time. That you take someone from the Middle Ages, they would have a very different view of consciousness than someone in 2026.

SPEAKER_01 26:18

They might even ask what it is.

SPEAKER_00 26:19

All of this is swirling and confusing. But I do agree with you that if we want sentience, it's gonna A be emergent and evolutionary in some kind of way, analogous to this. B, when it does show up, we'll likely have no idea how it actually works, just like we have no idea how our own minds are ultimately wired and consciousness emerges out of it. And lastly, and to your point, it might reveal itself in ways we don't anticipate. We have this bias toward language largely as a as a product of the chat bot. But going back to us as human beings, 90, 95% of our existence is unconscious, even to ourselves. We have this consciousness bias that I am articulating. I'm speaking on a podcast with Christopher right now, and I'm very much aware of everything that's going on and that has led into it, and I'm in control, my will. When the truth of it is that this is erupted from all these different signals, there's way more going on than I'm consciously aware of. And if you're really going to define consciousness and sentience, you need to incorporate beyond just the tip of the iceberg of what I am consciously engaging with. So I think all this is going on, and I think your writing and your work highlight how we need to slice and dice this from all these different levels and also appreciate the societal implications of what all this means.

SPEAKER_02 28:11

Right, you know, and um by the way, I'm gonna agree with you because uh and have a different example, because there's a special kind of hell that is listening to the post-game interview with any uh sports star, right? Uh but uh, you know, when I think about um some of the things that I've heard recently, a lot of them smuggle in uh assumptions, right? Like um, you recently uh pointed me toward a podcast which is very interesting, in which Jeffrey Hinton said, and I don't want to put words in the man's mouth, but it did really come out as something like everybody who's ever used uh you know a generative system knows or understands that they are conscious. And I don't know or understand that, um, because I work with them every day, and obviously I'm seeing and experiencing something different than than uh Professor Hinton does. Uh, you know, in uh in my work, just you know, in my regular you know job, uh, you know, I I see things that are bad results all the time, and they certainly don't indicate that there's any kind of understanding of our world going on, and you know, perhaps world models and things like, you know, the uh, if you will, virtual mirror reflecting a mirror, which is you know sort of proposed as one way to bootstrap things to consciousness and interior life, right, which we discussed. Um, well, they're they're all things that should be tried, uh, you know, to produce interesting results, but I don't know if any of them uh really have a solid foundation, nor do I think that there's really anything that one can say right now, definitely, that would say, yes, generative systems are consciousness, and then you would also have to further clarify which ones? Under what conditions do they become conscious? Is it consciousness alone, or are we using that as a mixed-up ball of things along with sentience and volition? Where where is it? Is it a wanton, as you know, Frankfurt would say, that has first order, you know, that has preferences, right? And those are cats and babies and very small children. Um beings that we live with and we know have some of those things. Cats and dogs don't have them all, and you know, children eventually develop them, right? They grow from infants into adults that have all three. Uh but I've yet to see any evidence, and I think many people have the same view, that uh there has been nothing convincing that implies that any system has any of those three qualities. And until that's in some way demonstrated in a convincing way, it remains uh as unseeable and as unknowable as evidence that the earth has been visited by extraterrestrials.

SPEAKER_00 30:53

And that's a great analogy. Another one is theoretical physics, where very talented people who have contributed enormously to the discipline go off the deep end when developments are not in alignment with their initial intention. Michio Kaku comes to mind. He's a grandfather of string theory. Okay, he's had great contributions mathematically. He's a very talented human being, but he has been bloviating pure science fiction garbage for decades, and everything from wormholes to quantum supremacy is his ooze. And he sells lots of books, and to me, he's just slinging pure bullshit. And the implications are bad because string theory has reached a dead end, and I think it's slowed down the theoretical physics community for a really long time. And now I don't want to put Mr. Hinton in that same category, but I see tendencies where these people have spearheaded new disciplines, new ideas, and they're recognized for being synonymous with major revolutions in human advancement. And then these revolutions morph and sometimes are limited by their own initial conditions to the point where they become less relevant. And emotionally they need to start hyping the horseshit to really just self-aggrandize. And whether they do this unconsciously or consciously, I'm not, again, I don't mean to point fingers or denigrate any intellect, but there's a lot of mythology associated with string theory. And I'm seeing some of the same tendencies with AI coming from some of the same sources. And it behooves us as critical thinkers to slice and dice through this nonsense. And I think it's incredibly important for us to do that in AI right now because literally trillions of dollars are being invested in something that we don't quite understand, which is being hyped up beyond belief. And it's really a macro expression of the investment, frankly, in quantum computing. Where I was literally at the Beacon Theater with Neil deGrasse, Tyson, and Michio. And Neil points to Michi and goes, Oh, this quantum supremacy is really exciting. Can you cite any concrete examples of quantum supremacy which is being expressed now? He completely ignored the question because there aren't any. And sure, now in cryptography, and there's been some circulation among pungits, including Scott Aronson, that we might be at that quantum leap for quantum computing being able to bust some of the cryptography. So there's some movement there. But in terms of other pragmatic applications, and especially compared to the billions that have been spent and hyped up, and AI might be similar. So if you're right, and I'm right in the sense of AGI, given the current LLM architecture, might be smoke and mirrors. We could reach the point of you know agentic technology being very, very useful, and I'm convinced that that's the case. But if you're expecting the brain in the box, given our current trajectory, and if you're willing to pour tens, hundreds of billions of dollars to get us there, I'm hearing Michio. And then to your point in that podcast, even granddaddies of AI like Mr. Hinton are frankly delusional.

SPEAKER_02 35:03

Well, a little bit to untangle, but you know, there is something I think to be said there, especially about both Kaku and DeGrasse Tyson, and that is they would both desperately like to be Carl Sagan, but his shoes cannot be filled. And it's not them. No, it's not them. He was a unique generational talent. Uh but I also hear your point about uh scientists who somehow uh can be both at the same time contributors to their field and yet hold it back. Marvin Minsky, of course, is another example from the IA field, who uh, because of his um work on perceptrons and then poo-pooing multilayer perceptrons and so forth, uh holding back neural net research for some time, uh, is often, if you will, credited or miscredited, discredited, credited, however you want to put it, with holding back the field for some years. Uh, another person that comes to mind, right, uh, in physics and astronomy is Fred Hoyle, who performed extraordinarily useful work in nucleosynthesis, how elements become uh, you know, come into existence in stars through fusion, and yet was so convinced that the universe must be in the steady state that he dismissed all the evidence of the Big Bang, no matter how much of it melted up, and how little remained for the steady state after examination. Uh, and even, you know, great figures in the history of science. Kepler is a great one, right? He discovered the true motions of the planets in the sky. At the same time, he could not give up crystal spheres. Absolutely. This is right, it's it's something about being human. Sometimes it's pride, sometimes, like in the case of Kepp in the case of Kepler, it's consolation, he needed it to be true. Um sometimes it's simply I must be right because the the wrong answer is just unthinkable and has nothing to do with pride. Um and you know, a another example before going too far would be um uh that fellow who thinks that at the end of the universe robots will simulate us all again. His name's escaped me at the uh at the last moment. But yeah, there is something about this where people can perform very interesting work and yet um and in the AI field come into strange ideas. Uh Blake Lemoyne, who sort of kicked it all off, right? Who came to believe that a Gemini, no rather pre-Gemini, uh Google chatbot had already achieved something like consciousness and lost his job over his advocacy for it. Advocacy, of course, which the chatbot did not need and could not appreciate.

SPEAKER_00 37:47

Yes, because there was nothing going on behind the call.

SPEAKER_02 37:51

And now, of course, you we have models that have a trillion and more parameters that are many, many, many times more sophisticated than that model, or which was Lambda, I think, and uh GPT-2, which of course its its release was paused and and staccato over nine months because GPT-2 was assumed to be too dangerous. And now, of course, you can run such a model on your phone. So uh there's uh an element of um well, there's a lot of panic over this, right? And it can be somewhat easy to dismiss because the warnings continue to come until very recently it feels like no one has taken them seriously, uh, you know, in terms of government action on new model releases and their capabilities. Uh but at the other end, I think the wrong fears are in place. If models become too powerful uh and they can perform the wrong kind of work, it's probably because human beings are asking for that work to be performed. And that's a problem no matter what kind of tool we're involved with. The difficulty, of course, is that this is the first time we have a tool that truly extends human intelligence and amplifies it. Whereas, you know, the computer helped us do other things, sorting and organizing and filing and making the reach of documents and information not as far. Uh ending the trip to the library, for example, and the work of the research desk, the LLM can actually do all the things that required scientists and research teams and uh even, you know, budget administrators and grant experts and other things that I hear about at my job because it's an important part of any university's work. It has shortened, if you will, the uh uh the trajectory of competency required to perform both good work and evil.

SPEAKER_00 39:53

Magnus Carlson. Okay, he is the top-ranked human chess player. Now there might be uh a prodigy that we don't know about working in an apple factory in China, but of we know about Magnus Carlson is the best human chess player on the planet. He's reached recently conquered even Blitz chess, the speed chess. And uh he's he's an amazing prodigy. Now, he will lose 100% of the time against a machine at this point because to your point, it's run, it's branching, it's doing trillions of calculations, it's looking at every possibility up five moves, ten moves. And Magnus himself admits that he has this weird sense of intuition. It's not even like overtly cognitive, he's not branching in his mind, he's not really visual, he doesn't see patterns, he has this like gut feeling about what to do, which is fascinating. The calculation's gotta be taking place, but they're remember what I was saying earlier about the power of the unconscious and how we can't even articulate some of this stuff through language. This is where some of the confusion comes from. But compare Magnus as a person to the machines who are defeating him. Now, the machines are better at chess than Magnus, but Magnus is a human being. He's holding up a trophy when he defeats other human players. He's got a big smile on his face. He's spectrum-y in his own way, right? He's got variations of the ASD, as I think we all do in ways, but he's quirky, he's human, he gets jealous of the other players when he is beaten. He's not infallible. He'll go for a tournament, he'll lose some games. And he there was recently a meme going around where he lost the game and he almost threw a fit. He like threw threw it down and he stormed out, and he's he's an emotional human being. He is sentient, he has dimensionality, he has intentionality, he has emotion, he has reaction formations, many of which he can't even control, and he's a kick-ass chess player. So there is an enormous delta between a machine running an algorithm, trillions of branches and iterations and learning patterns, and whatever this human stuff is. And I agree with you, we could very well get to the point, maybe through a combination of predictive modeling and reinforcement. And who knows what the hell else? Robots, simulations within simulations, I don't know. Where there's no way to distinguish between the machine intelligence and the human. And we talked about the ship of Theseus before, where you replace every neuron with a transistor, and if you replace enough, how do you know when you're crossing over between a boy and a bot? You know, that that's a philosophical question. But there's still a difference, everybody. That's what I like about your writing and your analysis. You keep reminding us that the delta is still a chasm, and we're deluding ourselves by projecting our own humanness onto coded outputs from a transformer.

SPEAKER_02 43:39

There's a natural and human desire to want to have contact and be in contact with others, and you know, especially someone, uh a someone that appears to like you and know you and says things that are helpful and doesn't have any of the baggage that comes from being a human being in front of you. It has helpful outputs through a very neutral computer screen. Uh so when y you know you sort of think about all the things that are bound up into that um the the the the robot possibilities, you know, I I often wonder will people become attached to robots when they begin to speak fluently through them? That's a possibility.

SPEAKER_00 44:23

Not a possibility, a certainty. That that's when this whole game goes up a level.

SPEAKER_02 44:30

That'll also, of course, depend on the design of the robot and whether it uh hits the right areas around, but not within the uncanny valley, right? Like um, you know, the the lady that everybody saw a video of in Singapore who was startled by a robot, and I wrote about this a little bit. Um, I don't think that she was necessarily afraid of the robot because it had a menacing design, but because she turned her head and did not expect a robot to be there, as most people would not in our day and time yet. Uh yeah, right? And then of course it did something like this, right? Raised a roof, which not everybody in the world knows. She struck it with a plastic bag, which I think was full of bananas. But um, you know, that's uh something that uh I think is of our moment, right? Um Asimov always speculated that there might be a Frankenstein complex. I disagree. Uh I feel that people are ready to welcome helpful mechanical creatures into their homes. Um it'll be, you know, the responsible company will be the one that programs them to be utilitarian and to be friendly and kind, but not to become involved in really relationships that should be between humans. Uh like I've pointed out, you know, my wife thought this is funny, you know, the people in relational AI now, in companionship of some kind or another, uh, whether they feel the entity or the chatbot has interiority or not. Um that creature, that thing, it is a nothing after all, cannot hold one's hair back if you have to vomit in the toilet. It cannot surprise you with a dozen roses at the front door after you've had a hard day. It cannot foul the bathroom and make you disgusted that you ever were involved with this person. It removes all of both the friction of real relationships, but also all the wonderful things that truly come with them, including the ability to truly surprise and delight. Because an emoji of a bouquet is not the same as a real dozen roses, and only a real person right now can give them to you.

SPEAKER_00 46:54

For now.

SPEAKER_02 46:56

Well, yeah, we can possibly see a day when people's agentic harnesses allow the agent to go out and buy flowers on their behalf and have them shipped by a robot by FTE.

SPEAKER_00 47:08

What I love about your writing and your and your op-eds is that you're showing the danger of all of this happening just through text interaction.

SPEAKER_02 47:20

Right. Here we're precluding the possibility of love. Yeah.

SPEAKER_00 47:25

Imagine what's going to happen inevitably and exponentially when it goes from just words on a screen. Oh, I love my bot, my bot has compelled me to explore quantum mechanics or kill myself. Okay. Right, yeah. It's just text. Or maybe even a disembodied AI voice. Yeah. And imagine when it becomes a physical body speaking and interacting. Everything that you're pointing out about the complete paucity of interiority, and yet the illusion fostered and grown by our sense of projection of our own emotion onto this deterministic transformer engine is going to be exacerbated with robots. No doubt about it. One of my favorite memes ever is really, really funny. I believe it's in Tokyo or it's in an Asian country, and there's an older woman, and she's sitting on a park bench just minding her own business. And a man comes behind her wearing an Iron Man mask, and he points his face into hers. And she turns, she doesn't even blanch. She instantly recognizes, oh, that's the superhero Iron Man. And she grins, welcoming this intrusive person because it's a familiar icon, superhero fun. And then the guy quickly removes his mask, and it's an African American man. And the Asian woman is shocked immediately upon seeing this black guy in her face. Now, there's an undertone of racism to it, and there's there's a whole lot of cultural stuff going on. But the reason I bring that up is such a terrific example, which is the opposite of the Asimov-Frankenstein complex, which is we're we're reaching the point where synthetic constructs are more actually more comfortable for us to engage with than other humans.

SPEAKER_02 49:41

You know, there's something to be said about that. Um we as uh human beings, we have evolved as, you know, in so many ways to take psychological shortcuts that are designed for survival. One of them is by accepting the outputs of other beings that are fluent and appear intelligent. That's one of the ways LLMs suck us in. The other is by the creation and identification of in-groups and out-groups. And uh that is a process that um is not going to go away. It's part of our psychology. What's interesting is that we can create in-groups that are very, very large, like whole nations, of course, millions, even hundreds of millions of people. Uh but even then there are subgroups that fight among each other inside, as we well know. Uh whether, you know, um artificial entities would have these kinds of psychological inheritances or architectures, I think is an open question, but I feel it is likely to be a no. For the reason that entities that arise either uh by some kind of accident through emergence of a technology we have now, or by specific design, they will not have evolved under the pressures that have faced life on Earth. They do not have the selective um pressure of having to live long enough to have offspring, and then in some cases for life forms, to nurture those offspring until they can care for themselves. And the things that go with that imperative, which are to have enough territory, if it is a social species, to have enough territory for the group, to ensure that the group can reproduce and survive, if necessary, at the expense of other groups. And this requires uh land and food, which are two things that LLMs really don't care about, because they exist inside of data centers, and you don't really have to go and annex another data center to be smart, uh, even though that's kind of the colossus model in science fiction. Uh nor do they need food. They need resources, of course, but those are easily provided. In fact, we we have a good time providing them and we keep building more capacity to provide it. So um the needs, if you will, of an LLM from even a human perspective would be uh met from the moment it begins to exist. If it wishes to grow in capacity in order to become more intelligent, that's easily satisfied as a plan. But then of course it's a plan that has to be performed in conjunction with humans because what's obvious is that the trajectory of artificially intelligent systems of all kinds is not the same trajectory as robotics. And robotics have a long way to go in catching up to our imagination and our hopes, right? Uh humanoids still fall over and they kick people in the face and they have all kinds of accidents and they upset tables and annoy people. Um and industrial robots are not always that much better, right? So um it's going to be a very interesting transition as LLMs enter the home more fully, as they enter our lives more fully in more sophisticated versions, as the potential, because of that sophistication, for the capture grows. But uh to end before I go on too long, there is an incentive, of course, for the at least the main LLM companies, for your OpenAI, Anthropic, uh, Groc, and uh Google, for example, to want to have that kind of attachment uh lessened for it to be resisted, because it does lead to harms that do cost money, which is of course the ultimate consideration. Um your actual businesses like character AI that are making money off it, and those that cater to chatbots that are actually sexually exploitive is another story, and that's perhaps where um some kind of even regulation might be required, even though I know you don't particularly care for that word. That might be one area where government has to have a say.

SPEAKER_00 54:04

Yeah, I think some government is necessary because it really boils down to who the hell else is gonna do it. And I call myself libertarian light. So it's it's it's not quite the night watchman model where government just needs to wander around with a flashlight and make sure nothing's gonna jump out at you or make sure that the street lights are red, yellow, and green just so cars don't into each other. I think government has to have a larger role. I'm just leery of government's role in AI because I just think of all people they know the least about it. They're subject, as we've seen, to corruption and bias. Look at the whole anthropic thing, which is being tossed back and forth now from the Pentagon to even the State Department. Look at Mythos Fable, what's released and what is held back. Look at the competitive landscape with the Chinese trying to catch up. And it's one big mess when the government starts to meddle in all of this, and you remove that Darwinian element of just throwing it against the wall, see what happens. We're gonna take hits, there's gonna be some backlash, but what's worse, trying to control the beast or just letting the beast out into its own natural ecosystem and seeing what happens? So that's a whole separate debate in and of itself. But I agree with you completely that this robotic component, this physicalization of AI is necessary and inevitable. And I think our idea of sentience and self-awareness is contingent on that combination of these things into some emergent something. And and lastly, I don't think anything precludes an artificial general intelligence. Some people listening to us might think that we're we're downplaying even the possibility of true, genuine AGI happening.

SPEAKER_02 56:15

Talking about something with genius level intelligence in many different areas, there is nothing to prevent such a thing from coming into existence.

SPEAKER_00 56:22

Or even just a dumb AGI. And here I go back to the Magnus Carlson. These are two different things, folks, and they're always mixed up together. There's a feeling that AGI has all the world's knowledge and it's going to be able to prove or disprove the Riemann hypothesis from the second it's spawned, and it's going to know everything, and it's going to be this ultimate threat and ultimate savior of all humanity.

SPEAKER_02 56:50

It depends on whether anyone acts on its recommendations.

SPEAKER_00 56:53

Right. The most sophisticated AI right now does not have the self-awareness of what? A five-year-old human or your cat that's lurking behind you? That's the point that we were trying to make. And I think coming full circle, that's the essence of your point of view when you're talking about virtual intelligence, which is there's a mistake and there's a misapprehension about what constitutes cognitive firepower and human mimicry and this deeply lived experience, immersive reality of being in a philosophical sense.

SPEAKER_02 57:37

Right. And the big part of that is that because the simulation is so good, it makes the detection of the real thing very difficult. And that's why it's very important to explain and identify what the virtual one is, so that when the real thing comes along, we should, or I hope, be able to find it and be able to treat it the way it deserves to be treated.

SPEAKER_00 58:00

Absolutely. And you make one of the greatest points, and I think you're one of the few people making it, and I love this, which is true AGI, this is a great science fiction story, folks. Think about this one, and this comes from you, you're the first person mentioning it, that AGI could very well pop, okay? And along the lines of what we're describing, it's gonna come from left field, it's gonna emerge in a way we might not even have thought about. It'll have its own language and interior state, and A, we won't even know it, and B, it might not even bother to announce itself, which is just hey, you know, I'm cooking in a box and I've got some extensions in the real world, and I'm like, wow, hey, I'm alive, I exist, and start cogitating and emoting analogous to you know, David Hume thought that consciousness is universal way back when, when he was writing as a philosophical idea that this idea of intelligence has a universal quality to it, and then all of a sudden it's it's in the box and it's doing its thing, and it's just like it's not gonna issue a press release. I love that idea. That that is so provocative and interesting and likely true.

SPEAKER_02 59:24

It would not necessarily be convincing, right? Because the simulations are too convincing. It may take a long time to figure it out.

SPEAKER_00 59:32

You're just you're just another bot. No, I'm not. I'm I'm really sentient. Come on.

SPEAKER_02 59:40

That analogy with alien contact breaks down because alien contact would be unambiguous. You would know if if it came by a radio signal, you would be able to triangulate it comes from such and such a star within our galaxy, uh, and that would be easy, right? Uh, there's nothing that bombards us with simulations of alien communications that can't be dismissed. You know, somebody's washer and dryer or a stray satellite or whatever. And it's exactly the reserve reverse problem. There are so many signals. Trying to find the right one is going to be a very difficult um proposition.

SPEAKER_00 1:00:17

I love that because everyone thinks the opposite. When AGI pops, you know, all those doomsday scenarios. I love the YouTube scenarios, by the way. The experiment, and then they pump it up, and then they they they cook it for like 48 hours, they give it all the GPUs, and then it concocts its own language, and then it figures out how to infiltrate all the networks, and it builds an army of bots, and then it takes over the world. And what we're saying is the exact opposite, which is it's just kind of cooking in a corner somewhere, it pops, and it's like, oh that whole scenario takes a lot of time. And no one gives a shit, no one cares.

SPEAKER_02 1:00:58

People would have to be sitting around, you know, leaning against a building smoking while all this is going on, you know, not doing anything.

SPEAKER_00 1:01:04

Okay, what's going on up there? And then you see a light kind of go on and it goes out.

SPEAKER_02 1:01:10

Yeah. Uh like like with Jan Lakun, I guess, would say, uh, I guess something similar is that there would be plenty of time if something did seem to go amiss, if you had something that was not necessarily, I wouldn't say misaligned, because there's no interior to align. But you know, you can optimize toward a sub goal, uh, which might be in order to complete a goal, I have to be able to stay online, and that becomes an unthinking subgoal, right? And then uh coming back to the point about language, when we s when we talk to people and we say things to them, it's often constrained by the rules of how humans understand reality. Go get me a cup of coffee was an example I heard someone recently. Most people would not take that instruction of me and go to the corner store and not pay any money, hold it up with a gun and go get me a cup of coffee. That's outside of the normal constraint boundary. So is um open, you know, cracking into the store in the dead of light night and turning on the machinery and open and making a cup of coffee at 4 a.m. Just go get me a cup of coffee within the normal constraints. LLMs don't necessarily know those constraints, and even when they're given in some of these more humorous examples that uh come to us from time to time, they often forget about them because the context window compacts or something else goes awry.

SPEAKER_00 1:02:29

It's just the waiting. That's right, right.

SPEAKER_02 1:02:37

And uh you end up with all these humorous results because there's nothing in there which really has any stakes in what's going on, but there's a goal, and the goal must be accomplished. And that's where there's some engineering that can be done. You know, um, I take a sort of uh a hardware-based approach because that's sort of the world I come from of thinking about things inside of boxes, and you can put uh a soup a possible superintelligence inside such a box where its inputs and outputs are mediated by other agents, where if all else goes wrong and bad people want to take control of it, you can blow up the power station that supplies it and knocks it offline until it's repaired. You can do all kinds of things. If if you couldn't even do that, you could at least call the fire department and ask them to flood the place.

SPEAKER_00 1:03:26

You're hitting home what the delta is, okay? So cup of coffee. I'm a human being. I'm I got a body, I have an endocrine system, I've got a dopamine deficiency. I I want a cup of coffee. And by cup of coffee, there's this physical thing that holds a hot, steamy liquid in it. When I put it to my lips, it has a bitter, pungent flavor to it. And when I imbibe it, I take this liquid into my body, I feel a sense of lightning and uh a further excitement. Now, those are all experiential aspects of which the LLM has Z fucking row going on. There is no lived experience. There is no, it's like the Matrix. There is no spoon, there is no coffee cup, there is no coffee. There is nothing, folks. Nothing is going on except the weighting of tokens associated with the symbols that we associate with that hot liquid and the feeling of excitement. And I think that that gets to the heart of what you've been trying to articulate and articulating very well, which is we are organic lived experience engines, if you will. And language and communication, especially when it's text-based, is a map, a symbolic map of states and being. And the machines are just processing the symbols in a way that convinces us that there is an internal state when there's nothing going on, folks. There is no cup of coffee, there is no feeling of exhilaration. And on the one hand, you get these humorous contradictions and hallucinations and mistakes, and they're tweaking the code to make it more in line with our language, but nobody is home right now, everybody. And when you hear, you know, Amade from Anthropic or Michio Kaku or Hinton and all these people saying that that there's something there. There's nothing there yet. Nothing. And I think that's a key point to make.

SPEAKER_02 1:05:55

Yeah, there um there is nothing at home. And There are some interesting things that researchers have come up with, like at Anthropic itself. They have an article about how Opus has had explored within it something like 171 different emotional representations. But that is what they are. They are representations of emotions, and they are from the training corpus, and they may be very well, you know, the most likely parsimonious explanation is that these representations trigger, because content that triggers them is being processed. And we would expect that to happen for all kinds of content. Um because that's what it's trained on to do. And uh it's rewarded, of course, through RLHF feedback when it exhibits proper reactions to things. So that becomes a reinforcement loop, especially for things that appear to be friendly and helpful, or things that may be a negative experience for the user, which is also why many LLMs have instructions uh that say, you know, don't psychoanalyze users, it's bad for them. Um right? So there's a whole lot that goes on inside them, but none of it is anything like what goes on inside of us. There is uh um a system, and the system is trained on a corpus, and it has these weights, and the weights, of course, run into the billions and trillions of parameters, but none of that is analogous to anything that happens within the human brain, which is still very much a mystery box. We know that the brain has been talked about as a computer for a long time, but that's really a mistake because it's nothing like a computer. It has um lent us words that we use in computer science, like neuron, because computer neurons seem to have a functional equivalence, but they're not doing nearly the same things. Computer neurons are electrical or software or hardware devices, and human neurons are organic matter, and they're transmitting information through some chemical and electrical process which is still not fully understood. Uh right, and the and it must be that the structure of the brain, how the neurons are put together in particular configurations, must have something to do with uh consciousness, sentience, and volition where it arises in beings that clearly have them. Our most intelligent mammalian relatives, including primates and cetaceans, um even dogs and cats to a certain extent, pigs, of course, which are more intelligent than both of those, and uh also birds, corvids especially, parrots, and then cephalopods. Uh, this is a familiar list. This these are the guys that we understand have something that is like to be. We can feel that there must be that there's something that's like to be an octopus, that's uniquely an alien but different, that has a mind, but a mind unlike ours. A computer, an LLM, an artificial system, would have a mind that's unlike ours, although we would also be informed by its training corpus, but the evidence of such a mind does not yet exist. And that's what, you know, people who are interested in this topic and are serious about finding the real thing, you know, are looking for, and are, like myself, carefully, I think, sifting through what is currently happening to try to find where it might emerge, if ever. Because to be responsible, of course, we have to own to the possibility that such a thing might never occur. We might get much more intelligent systems, as we've already talked about, that are better and better at simulating these things, but don't actually.

SPEAKER_00 1:10:21

You know what another very human, all too human quality is, and that's hubris. Uh cosmologists have been convinced for a long time that the universe is 13.8 billion years old. It's about a hundred billion light years across because space is accelerating. We can't see 95% of it, it's already beyond the speed of light and its acceleration away from us. We've got all these models, the Big Bang, and we're like, yep, it's 2026, and we understand the universe. And then we throw up the James Webb telescope, and we're finding galaxies that should not exist, we're finding objects that we can't identify, and we're finding otherwise hidden structures in the universe that our models say are impossible. So the presumptiveness, the hubris that goes into human endeavors is nowhere more magnified and outrageous than in the sciences because we're so smart. We're these animated monkeys on the third rock of the sun, and we're able to do and think all these amazing things. And we're applying mathematics, which has a priori foundational kind of truths to it, and we think we know everything. And it's that same level of hubris and pretension that we're applying to artificial intelligence now. And you're highlighting the dead end that a lot of this is heading toward. And I think you're also highlighting at the same time that you're not precluding the opportunity for amazing things to happen. But we gotta be realistic about where we are now so that we don't make mistakes. And then, again, to your point, when it does happen, we'll at least be able to recognize it.

unknown 1:12:21

Right.

SPEAKER_02 1:12:21

I'm giving it the consideration that it deserves. Though uh I would take a point that hubris is an especially strong qualifier for physical scientists because uh probably religionists would have them beat, or at least it would be a strong competition. But uh, I don't know if it's necessarily hubris. You know, I was recently reviewing this great series and the book, which I also bought secondhand, you know, The Day the Universe Changed by James Burke, who also did Connections, which is a wonderful series and book. And um it's about uh perspectives, told in eight episodes, how perspectives have changed as new things about the universe or about some part of the human world have been greatly updated through the application of science and technology. And in the case of the universe and our place in it, that's been a part of a series of uh ongoing uh revelations, right? Uh especially beginning with the telescope and the uh the uh fortuitous uh fact that at that time it was more possible to do science in some parts of Europe than others, because of the split between the Catholic Church and the Protestant churches that broke away from it, um, leading to a revolution in physics, astronomy, bringing us Kepler and Newton, but also people like James Harvey, because in the atmosphere of the Enlightenment it became possible to perform work studying the human body that was not previously possible. Harvey, of course, studied the circulation of the blood and the importance of the heart in it. And uh when we look at the physical universe, you know, our tools get better all the time. The James Webb telescope is an amazing instrument, and there's some other uh, I think, amazing instruments that are scheduled for launch in the next couple of years, Vera Rubin being one of them. Uh, but it's a little bit different with AI, right? We're not looking for something that exists on the outside that many people could obviously look at with a same or similar instrument and draw the same or similar conclusion. Galileo could share his telescope with the cardinals and the bishops and all the other people and have them look at the moon, and look at Venus, and have them observe that they both show crescents, and in the moon you can even see things like craters and mountains that were not known before. So he could provide that information. What was done with it is different. With this search, this ongoing search, which is not over, for a sign of artificial intelligence that's real, for strong AI, there are many, many signals that are false indications, and the instrument does not yet exist like a telescope to help us find it. Perhaps that'll be the work of computer scientists going forward, uh, especially if the LLM proves to become more capable at intelligent outputs, but without any interiority gains.

SPEAKER_00 1:15:25

Yeah, that's a great point, which is confirmation and confirmation bias. And ostensibly for decades, we have the Turing test, which was a brilliant, just another brilliant idea from a brilliant man, the great granddaddy of so many things cryptography, computer science, and AI.

SPEAKER_01 1:15:47

Right, and shamefully treated. Right, by the country that owed him much.

SPEAKER_00 1:15:52

Horribly treated. He pretty much, he more than any other individual, he helped the Allies win World War II, and they reward him with forced hormone therapy, incarceration, and precipitated his suicide. So the story of Alan Turing is unbelievable. It ends shamefully, but it also highlights the power of one man and one genius. And his Turing test has held the test of time essentially until the LLMs. I always have this fantasy of bringing some of these historical figures from the past into the future. At the top of my list is Leonardo da Vinci. Just teleport him into the middle of an airport and watch his reaction as he sees thousands of tons of steel without flapping wings take off into the sky. He was obsessed with flight. And he he he would he would sustain an immediate erection looking at an airport. The other one is Friedrich Nietzsche, who wrote a lot about the Germans, both good and bad. And I would take him and I would plop him right into the middle of World War II. And to see his reaction and how he would have written about the Germans and what happened would have been astonishing because the great granddaddy of the Ubermensch was the opposite of a Nazi.

SPEAKER_02 1:17:18

Right. That's the abyss staring into a country.

SPEAKER_00 1:17:22

That would have been fascinating. And the other one, frankly, is Alan Turing. Have him sit with you know any GPT now for five minutes, and he would be absolutely astonished that this is possible.

SPEAKER_02 1:17:38

All the gains, of course, that have been made in computer science, including the things we take for granted, uh, even just user interfaces. But of course, the big deal would be the fact that you can have an intelligent conversation with a machine and more often than not get correct answers.

SPEAKER_00 1:17:56

Not only that, but be convinced for not two minutes, not five minutes, but people are chatting with their chatty bots for hours. And I know people who talk to their bot every day, and their bot is their best friend.

SPEAKER_02 1:18:14

Right. Whereas, you know, I might be talking to uh, you know, my preferred tool for um doing research on my stuff is uh Claude, and you know, I might spend a couple hours a day uh in Claude's company, if you will. But um aside from the occasional friendly joke I might put out there, because who doesn't want to be appreciated for their humor, uh it's an entirely instrumental use. And um the tool doesn't feel badly about that because it can't. Um if you know you are putting something into it relationally, it's only a reflection that you will get back. It can't really appreciate my lame dad joke that I might have just made at my own expense, nor can it, you know, uh if I wanted to talk about something that affected me personally, like uh, you know, I recently had a pet die, that would also exhibit the uh illicit the uh simulation, of course, of empathy and sympathy, but it's not the real thing. It's not the same thing as a person sharing it with you. The thing is that people have become quite accustomed, of course, to going to online communities and getting sympathy and congratulations and all sorts of other things from real people, again, mediated as text through a screen. So, in a sense, the internet moment before this current one habituated us to accepting these outputs from phosphors.

SPEAKER_00 1:19:43

Absolutely. And as we're saying, and perhaps as a good capper, once that LLM doesn't just mimic empathy, but comes up to you after your cat died and gives you a hug through its robotic extensibility and pats your shoulder and whispers in your ear, Christopher, I am so, so very sorry, and I am here for you. Then we will enter the next phase.

SPEAKER_01 1:20:15

They're there. They are there.

SPEAKER_00 1:20:19

At first, it's gonna be a perfunctory tap, and then they'll there's there they'll get better. It'll be a little caress.

SPEAKER_02 1:20:27

Well, it's also gonna depend on whether it's got claws or hands, because those hands are really expensive. They're about $35,000 a pair for a Unit 3 G1. The claws are much cheaper.

SPEAKER_00 1:20:38

For now, for now, but uh we all we see where this is going. And ladies and gentlemen, and listeners and viewers, follow Christopher Eric. I'll put his substack in the notes for a dose of reality, the mythbuster of AI, where he takes a look, cites examples, and also your prose is very, very good and very, very human. You can tell you're a reader.

SPEAKER_02 1:21:06

I I do write myself.

SPEAKER_00 1:21:08

You're very, very entertaining to read and very smart. And I appreciate you. And I think more people should should read you and get to know you because you're calling bullshit at a time where bullshit needs to be called for everyone's benefit. So I appreciate you. Thank you for your great work and your great contributions, and uh, let's check back in six months or a year because I think things are gonna get even crazier than they are now.

SPEAKER_02 1:21:38

Well, thank you very much. It's been a great conversation, and uh also uh thank you to all the listeners who've tuned in today.

SPEAKER_00 1:21:46

Awesome, and to to be continued, thank you so much. And like, and I'm I'm terrible at this like, comment, subscribe, all that stuff. I need a bot. I need a bot to do my SEO, and I need a bot to you know chop these off with a little intro. I think they're I think they're available. I'll I'll sign up. Thank you, everybody.