Riskgaming

The Consciousness Winter with Erik Hoel

Description

In this episode of the ⁠"Securities" ⁠podcast, host ⁠Danny Crichton⁠ leads a discussion on consciousness with guests ⁠Erik Hoel⁠, ⁠Josh Wolfe⁠, and ⁠Samuel Arbesman⁠. They dive into "⁠The Consciousness Winter⁠," comparing it to the AI winter in artificial intelligence. This concept highlights how consciousness studies were once sidelined but have since seen a revival. The conversation covers various theories, including Integrated Information Theory (IIT), and the importance of a mathematical approach to understanding consciousness.

Transcript

This is a human-generated transcript, however, it has not been verified for accuracy.

Danny Crichton:
Hello and welcome to Securities, a podcast and newsletter devoted to science, technology, finance, and the human condition. I'm your host, Danny Crichton, and today we have Erik Hoel joining us. He's the author of the influential and Popular, "The Intrinsic Perspective," which is published on Substack and he's also the author of the novel, "The Revelations." He's an extraordinarily interdisciplinary set of talents across themes of consciousness, the mind, sleep, and dreams, and so much more. Eric, welcome to the show.

Erik Hoel:
Thank you so much for having me, Danny, and just a very quick correction and no worries because it's quite confusing, it's Hoel, for the last name.

Danny Crichton:
Well, there's a missing umlaut.

Erik Hoel:
Yeah, I know. I know. I've lost it somewhere along the way.

Danny Crichton:
That tends to happen. Yes.

Erik Hoel:
My foolish ancestors dropped it at some point, but it's an absolute pleasure to be on.

Danny Crichton:
Error, correction. Look at that, error, correction, and that is also joining me today is our own Josh Wolfe and Sam Arbesman. Welcome back to the both of you.

Josh Wolfe:
Thank you.

Sam Arbesman:
Danny, always great to be here.

Danny Crichton:
Eric, I want to start out with a concept you've been theorizing and writing about on your Substack, which you've dubbed, "The Consciousness Winter." It's a parallel concept to one in the artificial intelligence world of an AI winter where there are after two decades, sixties and seventies of massive speculation about the potential for AI to change our world, there was a collective pulling back of both researchers and funders who felt that the field was hitting dead ends and was effectively moribund. What is "The Consciousness Winter" and why are we worried about it?

Erik Hoel:
The Consciousness Winter was in some ways much larger than the AI winter and lasted for much longer. I would probably pick the depths of it around the 1930s or 40's with the rise of behaviorism, led by people like BF Skinner who are like semi household names and they became very, very famous with this rising, this rising critique really of previous methods. If you go back and you look at the early psychologists, the early neuroscientists, they're very open in talking about consciousness. You have people like William James who coined the term "The Stream of Consciousness," and that's writing in the 1880s.
Then you have some of the early famous psychologists, people who are sometimes called the fathers of psychology like Wilhelm Wundt. What he's doing is he wants to create something that looks like the periodic table for basic sensations. He has his poor graduate students and volunteers picking out every possible shade of color, things like this, and I think at some point they accumulate 30,000 different tiny variations in how census can differ.
All that stuff ends up getting heavily critiqued by this rising tide of behaviorism and there's a few reasons for that, there's the logical positivism is rising in the same time, there's the Vienna circle. Nowadays it might be critiqued as scientism, that gets a strong tailwind in the 1910, 1920s, and that effectively kicks out consciousness from being considered under the scientific purview until... I think it's arguable, until essentially the eighties and nineties. And you lose two, three generations of research into consciousness and it's very strange because of course if you ask a normal person, most of their cognitive life is built around a stream of consciousness. We all experience this, we wake up in the morning and we have experiences and we want to drink coffee, so we go get coffee and so on. It seems like our stream of consciousness is the effective level to describe our own behavior.
It's, I've argued, the main purpose of what the organ of our brain does, but then neuroscientists are placed in this very strange position where you're not allowed to talk about consciousness, but you are allowed to talk about some of its sub components. You can talk about attention, you can talk about memory, you can talk about perception, but you have to be careful. And it's really only until the eighties and nineties when there's this streak of luck where people who have immense scientific prestige like Francis Crick, who people might recognize as one of the co-discoverers of the structure of DNA, and Gerald Edelman, who's less well-known, but he also won a Nobel Prize for essentially figuring out the big picture of how the immune system functions. Very, very famous. And these two men looked at their own disciplines, felt, "Well, I solved that and looked around for big open problems in science," and they fixated on consciousness and through... We shouldn't give them the two of them all the credit, but through their efforts they create this new fledgling field of the neuroscience of consciousness that really takes off in the 1990s.
But I think that we are still almost as a culture, we're still, and certainly as science and its subfields, they're still feeling a lot of these ramifications. Consciousness is still not very popular to study. It's still very fragile and so on.

Danny Crichton:
One of the things you mentioned is that people were okay with memory and attention, perception was on the periphery and then consciousness forget. That strikes me as the first two are arguably measurable. You can measure somebody's attention where they're directing gaze. You can do tests of memory, perception, invokes idea of qualia, which maybe you can also explain to people. And then of course consciousness goes into the, "What is it like to be a bat," phenomenon. What is observably objective, or objectively observably, empirically, with attention and memory seems less so with perception and consciousness. Is that accurate and is that part of the reason why?

Erik Hoel:
I think it's part of the reason in that, or to put it even more broadly, we can very easily conceive of some mechanical reason for something like attention. There's a very popular hypothesis in neuroscience, which is effectively that attention is suppression. There's this spotlight of activation that's allowed and then everything else outside the spotlight is suppressed in terms of its activity. To the degree to which that's actually true is a little bit debatable, but certainly it's very intuitively, mechanically sensible for how you might get it. Something like the experience of red, what's the actual mechanical thing that's occurring when you experience red and you can quickly start to expound simple hypotheses, well maybe there's a big color space and you're at a particular point in the color space, but it's like, "Okay, well why is there sensations involved in that color space? Why couldn't you just invert it and get different sensations and so on?"
And these are things that people have long struggled with. I think one of my favorite things is to track down ideas that have made contemporary thinkers in philosophy of mind famous and then just realize that they're just literal complete repetitions of the exact same debates that William James was having back in the 1880s, and this all this stuff, including many of the famous thought experiments and so on, are literally just repetitions and lifted. These issues around how difficult consciousness have been with us for a long time, it was during this Consciousness Winter that they really got completely buried. One side really strongly won out. As I was investigating this, because people have not really written too much about it, and I realized there's this entire book to be written where I think people underestimate how much impact nerds can have.
If you look at contemporary Silicon Valley, it's a lot easier to understand the cultural impact that nerds can have. But if you look back at what happened to culture when science exiled consciousness out into the cold, and it's things like the replacement of modernism, which is very interested in people's psychologies. You have people like James Joyce, or Virginia Wolf, or Hemmingway. You have realism and romanticism, and that gets replaced by postmodernism, which I think could be accurately described as being very distrustful of consciousness and not really thinking about text as text, qua text, or thinking about playing meta games with your art or so on. It's the difference between a Matisse and a Pollock. A Pollock is all about the mechanical production of the art. And people theorize about this all the time, this big history stuff where they say, "Well, there's a mechanical age or so on."
And I think that you could just as well write an entire book about, it's very interesting that consciousness as consciousness gets exiled from science, the arts and culture become more skeptical about consciousness and experience and actually found that if you look at the Google Books trend of just the word consciousness alone, you have this massive dip right around 1940 where it really reaches its nadier, right, exactly where you would peg the depths of The Consciousness Winter and the total supremacy of behaviorism at all the elite institutions at almost exactly those times. It's like the culture itself is just less interested in the word.

Josh Wolfe:
Do you think there are hints in the culture and the arts right now of trends towards another Consciousness Winter? I feel like, and certainly people love writing memoirs and things like that, so I would say that points to there not being a Consciousness Winter, but are there certain trends that you think are pointing towards maybe Harbinger's or something like that?

Erik Hoel:
Yeah, I would say that if you were to say another one is for sure coming that would be sounding the alarm and that would be appropriately alarmist.

Danny Crichton:
Get the fire extinguisher.

Erik Hoel:
But I do think that given that it's happened once before, we should be maybe a little bit wary of it because it obviously can happen. And I think there's basically two forces that make me slightly suspicious or wary. And one force is within this subfield of consciousness research itself, there's been first of all a lack of progress in that there is no real fundamentally well-accepted scientific theory of consciousness. And then there are various proposals and some of them are quite controversial, perhaps one of the most controversial ones is integrated information theory. I actually helped develop some aspects of integrated information theory in graduate school. I worked with the theories originator and that theory has issues around falsification and falsifiability and it is very possible that all theories of consciousness might end up having some issues around falsification and falsifiability. And then you can have a debate around that.
To give people an analogy they might've heard that string theory has consistent debates around does it really qualify as a traditional science because we're talking about things, we can get elegant theories and then it's very unclear how we could falsify those very elegant theories. Their elegance is almost their downfall in some ways. And that's a real debate that has played out. And a similar debate is occurring within consciousness research. And there are some, particularly individuals, recently there was very big news, it was in the Atlantic and the New York Times and a few other places that were writing about nature and science I think, where a group of scientists basically signed a letter calling integrated information theory, one of these leading theories pseudoscience. And I think once words like pseudoscience start getting thrown around, two things that are relatively respectable and lead to a lot of neuroimaging papers and things like that, once you're willing to call things that are that well-established pseudoscience, there is a worry.
And then just to complete the second point, at a more cultural level, most people would say that GPT-4 is not conscious. I think that's probably correct. I think there probably is nothing it's like to be GPT-4. I can't say with certainty because that would require us having a successful, well-tested theory of consciousness. But let's assume for a second that it's not conscious, but it can still do all sorts of intelligent things, highly intelligent things, which means that consciousness and intelligence are relatively orthogonal. You can have vast intelligence with minimal consciousness, and that's very much unlike humans. And I think that there's this cultural danger where we look at these things and we say, "Well, this thing can speak and it can move around," and once you give it a robotic body and so on, and maybe humans are just very foolish biological large language models, it's like we always see ourselves through the lens of the technologies that we have.
We see ourselves through the lens of mechanical work, or clock, or earlier than that clocks, and then the logical functions of a computer. And for a long time everyone said the mind was a computer, and it's always somewhat unclear exactly what that meant. It had all sorts of philosophical difficulties baked into it, and I don't think it's been a very hyper successful model to apply and think about the mind as some sort of Turing machine that's trying to complete operations. It's very unclear that that's actually a very useful frame, but that's been going on for decades. I can only imagine what's going to happen once we get something that looks a lot like AGI and it's some non-conscious being, the attraction to say that the mind is that is going to be very high.

Danny Crichton:
I'm struck by all of your comments of just how interdisciplinary the study of consciousness is. You've talked about the natural sciences of biology, the biological basis of the brain. Where does this consciousness come from? You've mentioned the social sciences, psychology and philosophy, and then you're getting into the humanities. You're getting into stories, mentioning Virginia Wolf novel, "Stream of Consciousness," the Willie James quote, which goes over between the two. And I'm curious, obviously with Luxe, a lot of our companies are very interdisciplinary, Sam, our scientists and residents obviously very interdisciplinary yourself, connecting all these different ideas together in one place. Is that some of the challenge around the study of consciousness? Is it just that we just don't have an institution, so if we had an NIH for consciousness, all of this would be solved, or is it more fundamental where actual production of science, the way we have to think about this just doesn't meet the same bar as neuroimaging and we have a cancer cell and we can see it with an instrument in the same way?

Erik Hoel:
It's a great question. I think that consciousness research, my opinion is that it probably will not occur incrementally. It's probably going to be some modern day equivalent of a patent clerk who comes up with something that's just so explanatorily elegant that we get essentially a paradigm shift. I think that that's the most likely outcome for the current confusion. There are two reasons for that. One is that science has to proceed as an institution even if science itself is not making any progress. Neuroscience is a great example of this. We do not have very many good big picture theories of how the brain functions that we can slot results into, but we have to get results because we have to get grants and we have to publish papers and so on. The structure of science, it's like this big ship that you can't just turn off, you can't just stop it.
The fields of science that often languish are those in which you need really big ideas. And those big ideas are very hard to come by. And part of the success actually of people like Francis Crick and Gerald Edelman is just in allowing people to publish papers with the term consciousness in the title so at least you can have this living subfield where people can go if they're interested in it and interact with it. The degree to which that really constitutes progress is probably pretty low and it's pretty low simply because as you say, either the problem is very interdisciplinary or alternatively it's just extremely difficult to figure out. And there's some conceptual blockage that would run into when we consider it naturally.

Danny Crichton:
Let's take some of the thinkers, and I'd love to get, because I respect you, who you respect, and without being ad hominem super negative, who you would say, "Well, maybe their theory is not so great." For example, let's just go through some, as you described Crick. I remember reading, I think it was late nineties, maybe it was early two thousands, astonishing hypothesis, and I was caught by the cover. I opened it up, I think it was the macaque primates decoding the visual cortex, but it was almost a computational circuit diagram that I remember of understanding the flows and the hierarchical structure. Maybe two years later then there was the book on intelligence by Jeff Hawkins, who was a computer scientist and was invoking the kahal hierarchical structure, memory, prediction, framework.
You fast forward to today, and that always caught me, the idea that you were receiving stimulus, you would store that as a memory, you would make instantaneous predictions and then you would error correct. And I think most recently you've got Karl Friston, who I often find a bit, I can't tell if Carl is brilliant and super smart, or as people say, you can dazzle them with brilliance or baffle them with bullshit. I am not quite sure where he's on the spectrum, so I want you to weigh in on that.
And then most recently, there was another guy who wrote, "The Hidden Spring," and then there was somebody else on the memory prediction. Oh, Andy Clark, who's a frisked in friend who wrote, "The Experience Machine" or, "The Prediction Machine." And as a mechanistic way of experiencing reality, that piece makes sense. And then I'm personally comfortable to say, "Well, consciousness to me is the awareness of my awareness of those things," that there's something outside of the system at a meta level that is observing itself, and that that is it, that snails and other lower organisms may be making predictions in a input, output system, almost in a computational biology form, but they're not actually conscious.
They're responding to stimuli, food sources, light, gradients, trying to reduce those, trying maybe even in a sense that they don't actually have the personal awareness agency of that. And then that jumps all the way to the most recent thing that has inspired me, which is Sapolsky, that if you were to ask me the number one thing that I've changed my mind on, and I'm curious where you stand on this also, free will, do we have it or not? And I subjected my family to a very interesting debate recently between Roy Baumeister on one end, who said, "Of course we have free will, and here's all the reasons and the agency why." And you had Sapolsky on the other end who said, "Of course we do not have free will, and it's an illusion." And okay, I've thrown a bunch of names, texts, books, some of them legitimate scientists, some of them more popularizers. Where do you sit on the spectrum? Who is for real? Who is full of shit?

Sam Arbesman:
I thought we were taking notes with a checklist. Yes.

Erik Hoel:
What an opportunity to pronounce judgment.

Sam Arbesman:
Yes.

Danny Crichton:
From the arc of Crick and call it Hawkins memory prediction, all the way to modern day torch bearers of that in a sense, from Andy Clark and Karl Friston, and then free will that results from that.

Erik Hoel:
Yeah. I think the way you just broke it up makes a good deal of sense. I would say that there's essentially people like Francis Crick or Mark Psalms, I would group them pretty closely together to a certain degree. I think there's this approach and I think it can be a very good approach. I don't mean to say that this approach is bad. I think it's a smart approach, which is to say, "Listen, we're not really going to figure out exactly, at least right now, exactly what's going on about why it is that little neurons puffing chemicals at one another can constitute a subjective experience." We're going to put that aside, but we are going to say, "Well, maybe there's some neural correlates to that that we could investigate," and if we get enough data and we investigate enough, we'll constrain our theorizing in some way, or the answer will pop out. And maybe we were just philosophically confused about thinking that even this was confusing. Maybe it's deeply obvious. I think those people are dead wrong, just completely wrong. But it's very smart in that it allows for scientific progress to proceed. It is like a declaration of neutrality to a certain sense.
The only issue I have with those people is that often they get over their skis and then begin to assume that what is really a neutral stance is something a lot more. And I would peg Psalms under that and I would peg Crick under that, where I just don't really find their explanations of consciousness super convincing because I think it's all just skirting around these very difficult, almost metaphysical questions about how you get subjectivity out of objectivity, and they don't want to address that directly. And then I think sometimes they say they address that directly without really realizing that they're not.
At the same time, then you have these more computational math-based approaches to consciousness, and you could definitely slot somebody like Karl Friston under that approach. And as for Carl, well, I think he was once at a conference and he called me an idiot. Let's see. As for Carl, I'll be very nice and take the high road and say that I think if anything, a lot of the woo that Friston produces is basically co-authored Woo, most of his solo stuff, of the free energy principle makes a great deal of sense. I'll put it like this though. If you say that you have something that is the source of all knowledge and that you can use to create intelligence, then you really have to explain why it is that large language models have nothing to do with it and don't operate by those principles and why it is that you've never been able to scale up toy examples into something that's actually convincing or useful, or that can rival something just like back propagation.
Okay, right. That's pretty fundamental. And I think that that applies to a lot of theories. I'll include some of the theories I worked on under that rubric of integrated information theory. If integrated information is the fundamental key to consciousness in the mind, how come it's so easy to build minds without really paying much attention to the integrated information? I think these are very, very tough questions, and I've never gotten a good answer for something like that from anyone.
I think that when it comes to consciousness, if you look at the people who are making the most direct progress, I think it's like people trying to create mathematical formalisms and express their theories mathematically. I think as long as people are doing that, I rate that as a kind of progress because it's building up a theoretical vocabulary and tradition, and I think eventually you have to put some math in there.
Another way to put it is that everything that happens outside your window, the leaves falling, the rain, it's all following pretty simple physical laws. Similarly, whatever experiences your diverse neural activity is creating, or accompanying, or identical to, you can hedge on the terminology there, whatever that is, it has to be something that's relatively lawful and that can be expressed in some formal way. And that's where I think the progress will come from. And what I like about people like Friston is basically just that they're using so much math.

Danny Crichton:
Two questions. One, I want you to explain integrated information theory in the simplest way possible. And I'm going to give you, when I was trying to do this with my youngest, what Chat GPT came up with as the analogy, which is something, by the way, to Danny's point of reasoning by analogy, one of the things that I think conscious entities do, and then I'd love for you to share, if you had to say right now your best theory of consciousness, what is it in simple terms explaining like a five-year-old.
Let me start with the simple definition. Let's assume your brain is a magical kingdom. In this kingdom, there are lots of little fairy workers that are parts of your brain. Each fairy does a special job. One can smell flowers, one can remember the way back home, one is good at tasting different kinds of candy.
Now imagine all these fairies need to talk to each other to make the magic happen. If the fairy that smells the flower finds a new flower, she tells the memory fairy so she can remember it for later. And if the tasting fairy tries a new candy, she tells the smelling fairy. And IIT, according to Chat GPT, there's a fancy way of saying that all these fairies work together super well. And they don't just do their job, they share information, they make decisions as a team, and that is what makes the magic in the kingdom really powerful.

Sam Arbesman:
It's like inside out.

Danny Crichton:
Yeah.

Sam Arbesman:
Pixar, Pixar.

Danny Crichton:
Let's remove the math, but if you can explain IIT, and even the concept of phi, I'd be amazed.

Erik Hoel:
Yeah, I can absolutely explain that, but I will say it's not my preferred theory of consciousness. There's a separation there. You asked for both my best guess and then an explanation of IIT. My best guess would not actually be IIT, and we can just skip over what my best guess would actually be.

Danny Crichton:
No, I want your best guess.

Erik Hoel:
No, then I'd just start spilling the papers I'm going to write and I know what happens then.

Danny Crichton:
Oh. Okay.

Erik Hoel:
It is worth being paranoid sometimes. But I think IIT is probably the best leading current theory, again for reasons that have not so much to do with its correctness as its ambition and its formalization. To be honest, I don't think that there is another theory of consciousness that currently exists other than integrated information theory. I stand by that very firmly. I don't think that others deserve to really be called theories. Why? Because here is what you want out of a theory of consciousness. You want to take a brain state, and you can define brain state relatively loosely. We don't have to involve any math here. We just mean what neurons are firing and what neurons aren't. Something like that. It could be a dynamical trajectory, it could be just a frozen brain state, whatever. But you can imagine some super neuroimaging data where you're getting all the neurons that are firing in some human's brain, and you take that and then you ask this question, "What is this brain conscious of without looking at what the person is looking at or so on."
And here's the most important thing. You want to answer that question in a formal way. You don't want to just throw some statistics at it and say, "Okay, well maybe it's maybe seeing this," or, "It's seeing this," or, "I have a guess about it," or so on. It's like you want an actual answer. And so you can think of this as a mapping. You can think of this as a mathematical mapping between physical brain states and descriptions of experiences, what those experiences are.
There's no current theory of consciousness for which you could feed in a brain state and get out some description of the experiences other than integrated information theory. That alone, forget if it's correct or what have you. That alone just puts it so many leagues, shoulders above all these other theories that are like, "Well, maybe it's a global workspace. Maybe it's like a big bulletin board and then you put stuff up on the bulletin board."

Danny Crichton:
Because it's testable in that sense. There's something empirical that you can measure and test.

Erik Hoel:
... And, exactly. That's what's funny though, right? Is that there are some issues around testing and falsification for IIT, but it's very arguable that some of the other theories, you can't even really apply them to anything other than the loosest notion of neuroimaging data, but they don't provide a mapping. Here's a very simple question, what counts as a global workspace? The reason I keep bringing up global workspace is because traditionally one of the big theories of consciousness that's often considered as a rival to integrated information theory is something called global workspace theory, and it was originated by Bernard Baars and then has been developed by a couple other people as well. And I think in general it's very good work. I don't mean to say that it's bad or obviously dumb, but it's also just I think not even in the same league.
Because if I ask the theories originators, "What counts as a global workspace," no one can even tell me. Let's say I make a character, an NPC in Skyrim. And the NPC in Skyrim, it's like a guard, and he walks around the town, and let's say I programmed it so that basically the guard is taking his commands and putting it up into this little global workspace that little sub commands can work on. And he's very simple. It's like this minimal global workspace. Okay, is the guard conscious? Is this NPC conscious? Is that all it took? If he takes an arrow to the knee, is he actually taking an arrow to the knee? I think most of the time you'd say no, but people, they don't even have these discussions. The only people who have discussions like that are the people who worked on integrated information theory.
That's what I mean by it's a subtle point where I come across as almost evangelical about integrated information theory. And then I also have publicly said that is almost certainly wrong, but that's why I am so pro it and gung ho it, is that I don't even think that the other theories that are out there, other than some way less popular ones that are just hints at being developed are anywhere near capable of the, you give them data about the human brain, a real, true, huge data set about the human brain, and then it tells you in a formal, lawful way what it is conscious of. I don't think that that even exists, and it doesn't even exist for things like free energy principle. It doesn't exist for things like global workspace. It doesn't exist for Francis Crick saying that maybe it's oscillations at a certain frequency or whatever.
It's like, okay, let's say I just have neurons in a dish and they're oscillating at that particular frequency. These are very, very simple things, but the only people who think about consciousness, the way a physicist thinks about physics are generally the people working on integrated information theory. And that's why I've been very impressed, and that's why I say that it's the theory that you should judge against, but I don't think it's correct. I think we'll eventually get some something that looks like it, but is way more obviously acceptable.

continue
listening