• njikomes

Erik Hoel: Dreaming, Sleep, Machine Learning, Evolution of Fiction & Art | Transcript

Updated: Dec 3, 2021

Full episode transcript below. Beware of typos!

Nick Jikomes

Eric Hoel, thank you for joining me.


Erik Hoel 4:48

Thank you so much for having me.


Nick Jikomes 4:51

Can you give everyone a quick synopsis of who you are and what you do?


Erik Hoel 4:56

Yeah, sure. I'm a research professor at Tufts University. I'm a neuroscientist and an author. And I run a online blog called the intrinsic perspective. I also recently authored a fiction book called The revelations. And my interests kind of converge in talking about things like consciousness, emergence, and more recently dreams, which I think we'll get into here.


Nick Jikomes 5:24

Yeah, I, I read a paper that you wrote not too long ago called The over fitted brain. Dreams evolved to assist generalization where you basically outline your, your thinking for why we actually dream and what dreams are actually doing for us. And so I want to talk a lot about sleep and dreams and build a build a base for people in terms of what we know about the biology and the phenomenology of sleep, and dreaming how people have thought about, you know, what dreams are for if they're for anything, and then get into some of your ideas? So can you start off by just talking about some of the basic biology of sleep architecture, so you've got rem and non REM sleep in these different sleep stages? What does that architecture of sleep look like? And what do we know about why it's structured that way?


Erik Hoel 6:12

Yeah, well, I think that there's some very common things that are in textbooks. But those sort of clean distinctions, you know, between rem and non REM have become complicated of late. So you know, the traditional story is that the brain has these very distinct phases of sleep. And then it's like this whole brain phenomenon, the brain is either in what is called non REM, or non rapid eye movement, sleep, or in REM, which is rapid eye movement sleep. And the traditional association is that rems are involved in dreams. And non REM, in sleep doesn't involve any dreams. And for an intuition about REM sleep, and why we associated with dreams, right? Anyone who has a dog, and sees their dogs sleeping, you know, their eyes will be flickering back and forth. And there'll be, you know, moving your body around, you say, you know, it's dreaming. And this sort of global holistic view, wherein you have these separate sleep stages, everything kind of always goes in the correct order. And the brain is either in a global state or not, has been, you know, complexified, as late as things often are in neuroscience, you know, where it turns out that actually parts of the brain, it may be possible that they they're locally sleeping, so the predominant feature of non REM are these big slow waves of activity that traverse the cortex. So it's kind of like imagining a big wave of neurons spiking, going across the cortex as this kind of the traditional view, we do pick those up. But you know, they can happen kind of locally in local areas. And so I think, you know, we, we should think about dreaming, and sleep as something that is quite widespread, and something that's quite kind of wrapped up. Um, it's, it's actually true that, for example, if you do sudden wake ups of people, where you wake them up, when they're asleep, even if they're in deep non REM, you'll get Dream Dream reports, not all the time, but you'll get them. And so, you know, we should be aware that the things that you know, an electroencephalogram that reads brainwaves can pick up are generally pretty surface level phenomenon. And there's all sorts of, you know, wonky stuff involving local field potentials, and so on. So, you know, I, I think that it's important to, it's important to have kind of like a conceptual basis about the types of sleep and the stages that the brain goes through. But at the same time, I think it's also important to note that we don't quite know where the clean separation is between exactly when the brain is brain is dreaming, and when it's not.


Nick Jikomes 8:48

So so there's not necessarily as clean a separation between rem and non REM sleep as people once thought, but in general, just give us the broad strokes. So you mentioned that if you wake someone up from non REM sleep, they can report dreams just as they do with REM sleep, how often does that happen? Is there a frequency difference there? And in addition to that, are the dreams are they do they have a different kind of content when people report them coming out of non REM versus REM sleep?


Erik Hoel 9:16

Well, that's the first step. So first of all, you know, there's no like, actual final statistic for any of this stuff. You know, I think scientists or scientists get used to this very thing with like, 20%, right? It's like, well, what study is that from this Sunday? It's 20% Some studies 30%. Right? It depends on the time of night. So you know, you're, you're the sort of dream that you have changes throughout the night. Particularly, as you as you get kind of later on you you generally get more rem and your dreams get more kind of like elaborate. Early on in the night. Dreams are a little bit more kind of like mind wandering. And in fact, if you if you kind of think about that, that moment between waking sleep, you know, as you're losing control Just generally it's a it's an it's an episode of mind wandering, right. So there's a deep sense in which dreams are an extension of mind wandering. But, you know, to, to what degree that there's, you know, it's probably very dependent per person, and so on. But I do think that there is good evidence that that dreams are quite important. And I don't think that nature really kind of makes mistakes, and has these, these epi phenomena just hanging around, they seem, they seem, you know, the entire structure the brain seems bent on, you know, allowing for dreams. And that, to me speaks that probably they have a ever have an evolved important evolved function, like a high level function. And that's where this overfitting brain hypothesis came into play. Because there's a number of interesting viable hypotheses about what that overall function of dreams are, but I really hadn't heard this one, like, like well articulated, or well represented, you know, kind of, kind of aspects of it, it fits with with kind of other theories, right. So aspects of it have shown up in some other some other people's works on dreaming, but I really hadn't heard the hypothesis kind of put together and argued for in a way that I thought was kind of as convincing as it as it nearly could be.


Nick Jikomes 11:24

Yeah, before we get to this overfit brand hypothesis itself. I want to talk more about, you know, why why you would make the strong case that dreams serve an evolved function, that they're adaptive in some way. Because, you know, throughout history, people have also argued, as you mentioned, that dreams are just epiphenomenon. They don't, they're not really doing anything themselves. I'm hoping you could discuss what the evidence is that dreams actually evolved for an adaptive purpose. And I'd like you to discuss that by by maybe talking about both the homeostatic regulation of sleep and what that means, as well as its phylogenetic conservation in, in the animal kingdom.


Erik Hoel 12:03

Yeah, well, I mean, two things. So one is that scientists love calling things up the phenomenon that they don't understand. Right. So this is like the go to move, right. It's kind of like the particularly particularly in biology, it's very similar to what occurs in medicine when when people say maybe it's psychosomatic, right, like maybe maybe all these, you know, you know, hundreds of 1000s of of Lyme disease sufferers, you know, maybe it's just maybe it's just psychosomatic right, maybe they're just crazy, because we can't really find anything. And I think, you know, so this kind of shows up in the, that's a worst case scenario, right. But that kind of shows up in medicine. And it shows up in the sciences, when we have a phenomenon that we can't really explain in biology or psychology. Another great example of this is consciousness. And then we say, this is a, this is an epi phenomenon, right? And I think that what's why, you know, in particular, for dreams, you know, as you as you mentioned, dreams are kind of, well, pretty well conserved seems, you know, they pretty early animals, that they do a lot of learning. Dream, a great deal. Again, we are making a bit of a judgment there, you can't do sudden wake up of animals and ask them about what they're, they're dreaming about. So again, we're relying on this sort of association between rem and non REM, that is not really 100%. Right? You can like not get dream reports when when people are in REM, and you can get your important people or non REM right. So, you know, there's always like strange cases in biology where people will be like, well, there's some bat that you know, dreams like 20 hours of the day or something like that, if you judge off of its non REM, right it's like yeah, but maybe right and maybe that's a big problem or maybe we just can't do sudden wake up experiments on on bats and you know, their their brain architectures are subtly different from ours and so on. You know, in general, it seems as if organisms that need to do complex learning dream so mammals dream and what is the what is the defining feature of a mammal? mammal, mammillary glands, mothers, like mothers are the defining feature of mammals, they're what make mammals mammals and mothers teach their young, right? That's the the fundamental and most important function of of the mother do things beyond mammals dream, almost certainly. But it's a get again, it gets a bit harder to figure out because now you're making claims by the the homology or you're making claims by like, this neural area kind of looks like this neural area and human and so on. So, but I think I think we can be confident that in general, it seems most complex organisms that that need to learn a lot, do dream, you know, it would be interesting to get kind of final, you know, proof if like cephalopods dream or something like that, but I think we should kind of take take dreaming very seriously as as evolution seems to do. And therefore we should treat it as not an epi phenomenon, and particularly, we should pay attention To the the way that dream, the experience of dreaming. The phenomenology, what philosophers and psychologists call phenomenology, which is the structure of your conscious experience. So like phenomenology is what changes if you're on acid or something, right? The phenomenon or when you're dreaming versus sleep, more relevantly, the phenomenology of dreams is very particular. And it's very different from awake in a couple distinct ways. So I think a theory of dreams should need to explain why dreams are so dreamlike. Right. And I think that that's kind of the most important thing.


Nick Jikomes 15:31

Yeah, I mean, not only so you mentioned, you know, we can't really wake an animal up and ask them if they were dreaming. So we have to make these inferences about the association with dreaming and REM sleep. And the fact that, you know, majority of time that you wake a human up from REM, they typically say that they're dreaming. But even that's not completely reliable, right? Because I imagine there's cases where, you know, you wake someone up, and maybe they don't report they're dreaming, but they simply don't remember that they were dreaming.


Erik Hoel 15:54

Yeah, it could be it could be. You know, I remember talking to someone who studied the hemispheric sleep in dolphins, because it seems as if that what's going on with these, these organisms that need to keep moving, like how they do sleep is that they do this sort of time sharing thing where one hemisphere is asleep, and the other one is awake. Maybe they don't do it all the time. But they do it some of the time, right? And the way you can tell right is that one hemisphere kind of looks like it's awake. When you when you measure its brainwaves, and then the other hemisphere looks like it's in deep sleep with like, big, big waves. And, you know, I asked like, Well, what about when the other hemisphere is dreaming? Like, how do you know when it's doing that, right? Like that looks pretty similar to wake. Now if you do like a really detailed analysis of humans and you do a lot of kind of statistics, maybe you can kind of figure out that there's actually like a difference just off of the off the brainwaves, but basically, a dreaming brain looks quite similar to an awake brain. So you know, if there's if there's local sleep or hemispheric sleep, what about local dreams, right or hemispheric dreams? Right? So we it's it's very difficult to tell what's going on, you know, in this huge constellations of neurons that make up the cortex, when all we have is this like brief verbal output. And, you know, this causes huge amounts of problems throughout neuroscience. And I think you're absolutely right, that it could very well be the case, that we're dreaming almost constantly when we're asleep. And that, it's actually just that during non REM, or during these cases, you all your kind of reportage systems are come online last, and all the long term story stuff comes on last, and so you're almost never reporting anything. So you know, it, it would not shock me beyond belief to that in 100 years, the neuroscientific understanding is that there is like a steady thread of consciousness that maintains itself the entire time, even when you're out with anesthesia, or with or within a deep sleep and action, but it's just very minor. It's very thin. So, again, you know, I and the reason why I kind of talk this way and things this way is that, you know, I don't, I really don't want to mislead people about the current state of research, which which I view as as primitive like, like very, very primitive in that we're, we're in incredibly early days and trying to figure out how the brain as a whole works. And and even something like dreams, which we spend multiple hours in every night, you would think it would be well understood. It's still kind of up there, and then we up in the air and that we don't, we don't really know the function of it.


Nick Jikomes 18:25

And what does it mean, for stages of sleep, sleep to be homeostatic, Lee regulated? So as an example, if you somehow deprive someone of REM sleep, what happens when you stop depriving them? Yeah.


Erik Hoel 18:37

So you know, when, when you have? And again, you know, this is so, you know, again, when this is done in these studies, you know, it's we're talking about studies with, like, 20 people in them, right. And what you're talking about is this, this REM rebound effect, which is that if you deprive someone of a REM, they generally have more REM the next night. And I do think that that's that's quite interesting, and is evidence that Rabee dreams specifically have a homeostatic component where you kind of need to dream a certain amount. Right? And so you might say, Okay, well, that's case plays close there. But what about these cases of you know, where people, there seems to be some drugs that suppress dreams? Sometimes people take them for long times, sometimes people are completely fine doesn't need to have it have a detrimental effect? Right. So we don't, you know, there have been some interesting experiments where people try to like to deprive an animal of dreams entirely to see what happens. They did some experiments like this, and then in the 80s, particularly involving, I think rats, and you know, it's like, well, how do you do that? Well, you have to have the rat. Basically wake up whenever there's, whenever they move from non REM to REM, right? So you have you're tracking what's going on in their brain, you're waking them up every time they get into REM and trying to selectively deprive them ablate The Dreaming aspect of sleep, but how You do that? Well, you have to wake them up, which is very stressful, right? I mean, you're basically torturing these rats, right? I mean, these These wraps are in, you know, some, some, it's not a fun time to be a rat in that experiment, right. So now they're they're mentally stressed out, they're on like this moving platform that, you know, kicks into gear and keeps them moving. And so then, you know, some of these rats started dying. And then it's an interesting question, right? Which is that is it the lack of dreams? That's killing them? Is it kind of like the the stressful experience of the of the experiment itself? That's, that's killing them. Um, you know, these, these things are quite difficult to untangle. So I think that probably from from the compendium of evidence, it does look like, like, creaming and sleep, certainly sleep in general, is homeostatic regulated, like there's no question at sleep in general is having a statically regulated? But, you know, for dreams in particular, I think probably, but my guess is, is that it just doesn't start, it doesn't matter as much as you get older. And this is something that I think people, you know, within neuroscience need to face, which is that there may be functions of things that it's almost impossible to tell over a single night, given almost any experiment you run. So an example being like, is there really a cognitive deficit from not dreaming for single night? Well, if the impact of dreams is cumulative over your entire life, then there might not be at all right, which would explain why I mean, even for sleep, right, like, when we think of sleep, there's something incredibly important, there are plenty of elderly people who get four hours, three to four hours of sleep every night. And many of them continue to function at a relatively high cognitive level, particularly compared to their population. Right, you just sleep less as you get older. So, you know, the question then is, okay, so what, what does that tell us about, you know, the, the fundamental properties of sleep and what it does, and what it might tell us is that, listen, we're dealing with something that's, that's cumulative over life. So, you know, sleep, and learning is about crafting. Something that is successful at, at various tasks, right. So that's what a neural network is, it's something that gets good at various things and gets fitted to various things. So, you know, eventually, when you have a very kind of good human brain that's well fitted, well generalized, and and and it's kind of appropriately learned what it's supposed to learn, you probably don't need to sleep or dream that much, to be honest, that that's, that's what I think. And then you think how difficult that is then to show up in like a single night's experiment by sleep deprived, and keep on hoping that they have some effect on their cognitive performance and so on.


Nick Jikomes 22:32

Yeah, so before we talk about dreaming more, let's actually step back a little bit and talk about sleep, sleep itself. So you know, all of all animals that sleep, it may or may not be true that all of them dream, there's probably a different amount of dreaming that happens in different kinds of animals, humans being an example that that probably have relatively high levels of dreaming that happen. But what do we know? Or what are some of the major ideas out there about the biological or physiological function of sleep itself at like the cellular level? What's like, what's actually happening in our brains every night? Yeah, it's


Erik Hoel 23:09

probably a form of, it's probably a form mostly a form of housekeeping. So I think that that that's been a long standing hypothesis that housekeeping can come in various kinds of shapes and sizes. So some people recently, there's been some very interesting work that shows that the brain is basically being flushed with cerebral spinal fluid, like, almost like, the brain goes through a wash cycle every night, right, where essentially, you're being the interstitial space is being flushed with the cerebrospinal fluid. And it's just like clearing out the the garbage because, of course, you know, neurons all run on these, you know, onic pumps, they're, they're constantly cut, they're these little like, Maxwellian demons, you know, who are constantly taking in, you know, things and then spitting things out, right? So they generate garbage. And it could very well just be the case that you just need a time to pick up the trash. Now, again, and the various forms in which that can happen, right, you could have flushing, you could have kind of other mechanisms. But that's a long standing. And I think, probably probably mostly correct hypothesis about the very overall purpose of sleep. Of course, once you have an animal, I mean, most animals are diurnal, right, so that they have a particular time in which they are geared towards, right, they're selected to operate within this particular time. But it's funny, right? Because your your, your split into, you know, basically to eat meat periods, and you have to pick your your pick either day or night, right. If you're nocturnal, then you're gonna be out at night. And then or you could be out during the day. And depending on which one you pick, of course, you're going to have a lot of downtime. So either housekeeping takes a while and there's downtime and maybe dreams evolved during that period. Or alternatively, it could just be the case that you really don't have very much to do as an organism when it's not your time and you have to find some way to fill the hours right? And then maybe during devolve, because it gave you some advantages when you did that. But, um, I think that, you know, in terms of the overall purpose of suite is something that we understand a little bit better. But still, again, surprisingly, not nearly as much as, as we should, particularly because, you know, it's the state in which we lose consciousness, right? So we lose the primary property of the brain. But if you look at what's going on inside the brain at that time, you might think, for example, that all the neurons are quiescent, that they're just totally silent. But they're not right, like neurons during non REM can fire more than neurons when they're awake. So, you know, I think that, you know, this sort of research it, it'll be very easy to be kind of fooled as to kind of how far along we are. But there are some very strange facts about sleep, for example, the firing neurons, ones that really isn't kind of well accounted for by contemporary theories, like why slow waves, like find it, find it find a good hypothesis about the purpose of sleep, that really, really explains why it's the slow wave, like why a big wave of activity across the cortex? Why is that necessary for like, metabolic meaning?


Nick Jikomes 26:10

What about this synaptic homeostasis hypothesis that I've read about? What does that actually say? And how much evidence is there for that?


Erik Hoel 26:18

Yes, the synaptic homeostasis hypothesis was actually originated by my by my mentor or two, actually, because Charlie, Charlie is another one. So I went, I got my Ph. D. in neuroscience working with Giulio Tononi, who's out of team for city of Wisconsin, in Madison, who's probably well known for two main scientific theories versus integrated information theory, which is a probably the first well formalized scientific theory of consciousness. And then the second one is synaptic homeostasis hypothesis, which is the idea that during that during non REM, one of the big or maybe the main housekeeping tasks that the brain is emphasizing and doing is a net reduction of synapses. So, you know, as the, as you're learning, your synapses are potentiating. And the idea is that, in general, there seems to be more potential creation, right than depression. So in general, your synapses are kind of getting larger, and as and if you just kept that that, right, what happens? Well, you know, eventually you'll, you'll get like an epileptic fit, because your your your synapses are, too, potentially, they're too strong. And all sorts of things will be going wrong. So you get cut out of whack. And then the idea is that there's kind of this net downscaling that occurs, where listen, if we all just downscale, then we'll keep kind of the same pattern of interact of interactivity, right, we'll keep the same dynamics, but we'll all kind of be net downscaled. And that kind of takes care of this, basically, this problem of learning. And it's a hypothesis about about slow waves. So it's one of those rare things where it's like, no, probably they think the slow wave kind of creates is a signal for the neuron to perform this downscaling task, you know, something like, roughly like that, right? So that is a nice hypothesis, because it's directly relating something that we observe and is characteristic to, to hopefully this this synaptic downscaling that they have there. Now, you know, the evidence directly for is I think, is I think, good and that we know that synapse is to change overnight. I think that that's well established, whether they all kind of collectively do this collective downscaling that's much more in the open, there's been a number of studies that have kind of contradicted this showing that you can have upscaling downscaling, you can have all sorts of like learning effects that you would normally expect. And also, I, I've always had one particular issue with it, which is that if you tried to implement that in an artificial this is like my test for everything in neuroscience, by the way now. Yeah, I grew up before the deep learning revolution. And I really think that neuroscience has not, and neuroscientists have not adequately paid attention to what's going on. Now. They're like, ignorant of it, but they haven't really kind of grokked that. If we can't explain something in an artificial neural network, there's just no way we're going to be able to explain it in a brain. And a simple example of this is that we know that in artificial neural networks, which are formed of these little networks of artificial neurons, these units are nonlinear. And that means that if you did a net linear reduction of the synapses, you totally changed the dynamics in the patterns of what that of what that artificial neuron was neural network was doing. I see and I haven't seen that ever seen that articulated as an objection to shy I think I brought it up you know, once or twice and I forget what what what Julio and Terry's answers were but you know, it's it's That is I think, problematic. And I think that we should really mean towards hypotheses that kind of makes sense and artificial neural networks.


Nick Jikomes 30:07

Interesting. Before we get there into the artificial neural network stuff, let me just make sure I'm following so. So there's almost certainly housekeeping functions being performed. While we sleep, meaning at the cellular level, basic, basic stuff has to happen, the brain has to get cleaned out in some sense. And it makes it makes some sense that that would happen during sleep while the animal is offline and not behaving, which would be at night, if you're diurnal, or during the day, if you're nocturnal. And there's also this idea of synaptic homeostasis, which, you know, maybe has some mixed evidence for it. But the idea is simply that, you know, if I'm a human who's awake during the day, I go about my day, I'm taking in all kinds of information, I'm learning things, I'm seeing things I'm doing things my synapses on that are going to get potentially aided, meaning like, literally, they're going to get bigger, more proteins are going to be shuttled there, I'm going to spend a bunch of metabolic energy to make all the synapses bigger and stronger. And if you just do that, across all of the billions and billions of synapses in the brain, that's just way more energy than you can afford, because most of the information throughout each day is extraneous. And so the idea of this hypothesis is that then at night, maybe particularly during REM sleep, you sort of scale back some or many or all of those synapses, because you simply can't afford to maintain that much connection. Yeah,


Erik Hoel 31:27

precisely. So you summarize it really well. Yeah, exactly. And I think it's a it's an elegant hypothesis, you know, whether or not it kind of survives, like, all the, the empirical testing remains to be seen. And it's also a bit specific to non REM, I mean, at least the way it's usually presented, it's specific to non REM. So it's kind of, you know, it may be that the brain is basically doing a huge number of tasks, right. And sleep is like the Swiss Army knife, where it's deploying, you know, 20, different kinds of functions, all of which have some, some relevance and theorists are kind of picking up on on little bits and pieces of it, it may also be the case that the brain just doesn't doesn't do anything like that. I mean, it may just be the case that actually, you know, potential ation isn't really a problem there sort of Things only get so potentiated, naturally, and you actually get as much like long term depression, as you do get potentially ation in the long run, and everything balances out. So, you know, again, it remains to be seen, but you know, I, the second, this whole, like, this whole test that I have now is, is just like take the hypothesis, see if it makes any sense and artificial neural network, if it makes like no sense, then I'm very skeptical. You know, so within neuro science, right, so you know, some things do kind of make sense, like grandmother neurons, population coding, stuff, like that makes sense. And then some other stuff just like doesn't make any sense.


Nick Jikomes 32:48

So before we get to this over for the brain hypothesis, it'll make sense to talk about deep neural networks before we get there. Because some of the ideas connect here in interesting ways. Why don't we talk a little bit more about the phenomenology of phenomenology, phenomenology of dreams. And, in particular, you mentioned in your paper, three characteristics that dreams have and then you kind of connect this to artificial neural networks? You say that one, they're sparse, they're hallucinatory and last, that they have narrative content. So what do each of those things mean? And can you just connect that to the basic phenomenology, phenomenology of dreams that people will be familiar with?


Erik Hoel 33:25

Yeah, sure. Of course. And, and I think it's really important to discuss phenomenology as a neuroscientist, I think it's really important to pay attention to subjective experience. And again, make sure that there is kind of fit fit with them. So sparseness is, is it may at first seem seem slightly weird, but let me explain what I mean, which is that ultimately, the dream world is lower resolution than the waking world. A really easy way to see this is just that how many dreams have you had, where you're doing something minute? An example would be that most people very classically, don't dream about their phones. Like in a dream, you know, your dream because your phone is broken, right? Because it is basically your brain. It's like the matrix, right? Like coming up with this like glitch of like, okay, I can't, I literally don't have the resolution in your conscious experience to represent like, although icon, iconography, and although letters and everything like that, it's the same reason you don't really read when you're dreaming as well, even if you spend a huge amount of your day on your phone, right? Like this is something that you really should be dreaming about. But you just don't. Another simple example is is just that, you know, in general, dreaming is, as we said, and this is, again, something that's kind of proven, relatively well proven experimentally dreaming is related to mind wandering, and is related to kind of imagination, right? And when you when you are thinking about something or imagining it, you're you're generally considering it in like an abstracted framework where it's really kind of lacking the details, right? And so, you know, your imagination is sparser than your waking life. There's basically no living human There's some people kind of claim this, but I'm kind of doubtful where we're like they can imagine something. And it's just as vivid and real as real life. Right? If you imagine a dinosaur or window, it's it's not like a scene from Jurassic Park. It's, you can imagine it, you can see it, but it's not really as vivid as right. So dreams are similar to this, right? They're like these sparse events, where they have detail, but they don't have the luxurious, open detail of waking life, probably just because the brain can't can't really support that level of detail from the bottom up. It can't really generate that. Oh,


Nick Jikomes 35:32

yeah. I mean, it makes sense. And once you mentioned it, I do have to say, I don't think I've ever I cannot remember a single dream I've ever had where I was looking at my phone or reading any text.


Erik Hoel 35:43

Hmm. Generally, some, some people are like, I had this one dream once.


Nick Jikomes 35:49

Yeah.


Erik Hoel 35:51

But you know, in general, I think it's I think it's true that it's pretty low resolution, I think probably even even like your attentional window is probably shrunk. You know that there's, it's very difficult to figure out exactly what these rules are. But But in general, you're just not getting this bottom up sensory stimuli. The second property of dreams is that their hallucinatory, right and this is a little bit easier to talk about, because most, pretty much everyone has had crazy listeners where your dreams, they're very category breaking. So sometimes it'll be in your house, but then later, it's kind of spaceship, maybe it's kind of both somehow, right. And you're kind of mixing up the plots. You know, strange events are common dreams, like there's a reason we call you know, Twin Peaks dream like David Lynch's works dreamlike right, dreams are mansion, which is super weird. It's like, why would biology evolve something lynchin. Like that's such a strange, seemingly epi phenomenal property, right, that I think is actually not epi phenomenal. I think it's very functional, which I'll get to in a moment. But I think that these two are like very broad properties about phenomenology of dreams, and they probably reflect to its function. So there's something about experiencing the world in this sparse hallucinatory format, that may be very helpful to learning are helpful to the overall cognitive health. I don't mean health and medical sense, like cognitive health is in that the brain is good at certain things, the overall cognitive health of, of the animal. And then in humans, humans are have these narrative dreams. And I think, you know, that probably this is, in a sense, the least important property, because I think it's very particular to humans. I'm not, you know, I do think that all critically, mammals are, you know, they're their agents that are situated in their world in their conscious experience. So they do, in a way, conceptualize the world as primitive as that's a primitive stories, humans conceptualize the world as stories and events, like, it's always like this, this, you know, we're always in a scene, there's always events that are occurring. It's not kind of that the howling gale, that it actually is in physics, you know, we we have this very kind of constrained view of the world. And that just gets filtered in. And I think that that's probably, that's probably worth keeping in mind for long term consequences. But probably for other animals, it's not nearly as intense, and maybe they just have scenes, there's some evidence that that, you know, developing children don't really have narrative dreams, they have more like scenes that occur like still scenes, things like that. So animals in general, they might just kind of have these sparse hallucinatory scenes or quick events that aren't nearly as kind of like, imagined in a narrative as humans. Because


Nick Jikomes 38:34

dreams have these these properties, they're sparse, they're sort of low resolution, and they're not very detailed compared to what you see, when you're walking around looking at the world, they're hallucinatory, the the things that are happening in dreams are very strange, and they don't go together in a way that makes a lot of logical sense much of the time, and their narrative in their structure, at least for humans, you know, they sort of have a story like quality to them with you as a character, even the hero at the center of them. So with these things in mind, let's talk about deep learning now. So to start with, for people have no background in deep learning, what is deep learning? And what is the deep neural network?


Erik Hoel 39:08

Yeah, so, um, deep deep learning is basically just this trick that people figured out, which is that if you if you train, or if you make artificial neural networks into many layers, so rather than just having one layer of neurons that are learning the task, now you have multiple layers that feed forward from each other. And you can stack these layers quite deep. And what that allows is some sort of like modularity or hierarchy of abstraction, like probably roughly, this is what it allows, and is what's driven in such amazing progress in self driving cars, you know, or these are filtering the spam in your in your in your email box. So this this kind of extrapolation to these deeper neural networks, and it should be made clear that listen, that was always a good idea because if you look at the brain, you do see a lot of hierarchies. You do See a lot of kind of layered structure. And so probably that that's something fundamental and innate.


Nick Jikomes 40:05

So when you're when you're talking about deep learning and you say neural network, what you mean is pieces of computer code, sort of separate pieces of computer code that are linked to each other in a way that sort of mimics the way that different neurons in the brain are linked to each other.


Erik Hoel 40:20

Yeah, exactly. Like the computer is just the operating environment, right? It's just like the world in which this thing exists. And the thing that exists is made of code. And what is made of code is basically an artificial representation of a brain. There's various simplifications that have gone into this. And of course, it's not a mammalian brain, it's like a some super abstracted notion of a brain. But it's still very similar in the sense of you have these discrete units, which are neurons, they're connected to and talking to one another. And the signal kind of feeds those through it. And you, you train these, you, you train these artificial neural networks, and the way they train them, by the way, is probably very different from the way that that our brains learn. That's something that is probably very fundamentally different between the two. But overall, I think that neuroscientists should look to deep learning and what works in deep learning. And one of the things that they discovered, as they were doing this, which is that when you're training these neural networks, so you know, these things are trained on, like some set of data. So if it were, if it were a neural network that drove self driving car, it would be like images of the road, something like that. And what they found was that, you know, what's weird is that a self driving car will actually get better at self driving, if you do this one strange trick to the data that's learning if you augment it, and you can augment it in weird ways, an example would be that, imagine I'm training the car to drive. And but now I'm training it to drive and I've placed black bars across all the input screens that it's getting. So imagine, like a driving scene, but now there's like a completely black bar, just put right across it. And, and you might say, wait a minute. So now the data is worse. Right? How is the car now getting better at driving tasks that if I didn't randomly put in weird black bars into the data, that doesn't make any sense. But it does, if you if you think about it, in detail is paradox, but it does. And an example would be like, imagine the car is trying to learn where stop sign is. And what happens is it becomes very sensitive to exactly how stop signs look. And the moment something is obscuring the stop sign, it no longer knows that there's stuff in there. So imagine, like a branch is obscuring it. But now it's been trained under these conditions where it expects random black bars, like over like halfway through the stop sign, and says, Listen, even if I just do a partial stop sign, I know to stop, right. And I didn't get that from the original dataset. So what we've done is augmented the data by making it sparser. This is a very common technique called dropout, it works pretty similar to what I've described. And that seems to improve generalization by feeding this neural network sparser data. So, and similarly, the same thing is done using by basically making these artificial neural networks hallucinate, right, you do something called domain randomization, right. And an example of that might be I think, there was a case of at MIT, where they're training artificial neural networks to solve a Rubik's cube using a robot hand. And what they did was that they, they're basically taking the incoming data stream, and they started manipulating it. So that was pure to the artificial neural network, although this wasn't happening in the real world, that the Rubik's cube was getting bigger and smaller. Right. So imagine you're trying to solve a Rubik's cube. And now like, it's like growing and shrinking in size, right? It's like this crazy, difficult, difficult thing to do. But even though that could never happen in the real world, training, artificial neural network on that actually improves the performance of it. And the reason why is that it helps them generalize. And so basically, I think that dreams share these properties that had these properties, because their times have to train offline training of the brain using broadly very roughly similar techniques as as is found in deep learning for that augmentation.


Nick Jikomes 44:07

I see so so to take the the artificial neural network example, let's say you're training this computer code to, to work as a self driving car. And it seeing clean images of the road pristine images of the road, you see all the lines, you see every stop sign perfectly clearly. What you're basically saying is if you train a network on a very clean and perfect, quote, unquote, data set like that, it becomes overfit to the data such that it doesn't perform well. It's learning there doesn't generalize well to the real world. And the reason for that is, as you mentioned, the real world isn't always filled with instances where you have a clean, crisp view of that stop sign. Sometimes there's someone blocking it because they're walking across the sidewalk, sometimes there's a tree branch in front of it. And so the resolution to the paradox that you stated was, you know, these these networks weirdly get better at generalizing when you give them corrupted or incomplete data. And the reason for that is the real world itself is not, it's not filled with perfect data.


Erik Hoel 45:09

Exactly, you never have, and no organism will ever have a good an actually good sample of the environment. We're always under sampling our own environments. And the way that I think about it, and this is kind of the this is the pitch behind that overfitting brain hypothesis, which is that, listen, evolution has given us this amazing, particularly mammoths has amazing ability to learn very quickly and very well. But the problem with learning quickly and learning well is that you learn to well, you become too reliant on, you know, the fact that the stop sign is always in bright sunlight right in front of you, or you become too reliant on, you know, the fact that you can see all the letters, right, so you're always reading stop, but the moment those letters are obscured or some letters are missing, you mess up. And so I think that there's, there's both overfitting in like this very technical sense of machine learning and deep learning. And I'm using it in the same way. But there's even a broader sense of just getting too good at one task at the exclusion of others. And I think that that's probably something that all organisms face, because frankly, life can be repetitive and boring, right for humans, but also for for all other mammals, right, like you're a wolf, you get up, it's the same den, right? You just get very, very used to certain things, and you learn them very, very well. And that means that you're unprepared when you need to quickly learn or develop a new novel tasks, because you've become so fitted to your environment. And so dreams are like these weird breaks, where it's like, Okay, now we're going to do something completely different, almost like putting, you know, it's almost like keeping like, like an astronaut muscles working by injecting little, little needles into them that shock the muscles, right? There are these machines that can build muscle, just kind of shocking you, and you're not actually doing anything, but they're just in the background, kind of shocking you. I think dreams might be similar to that for cognitive performance. There's just this like background exercise, that that the brain does to kind of maintain the generality that prevents us from actually learning too well, because it's because it's such an advantage to be learning. So, so quickly. Thanks. So we're just so prone to this.


Nick Jikomes 47:22

Yeah. So you had this great line in your paper where you said, it may seem paradoxical. But a dream of flying may actually help you keep your balance running, which I thought kind of captured the essence of this hypothesis quite well. So so let's just state it very explicitly. Can you summarize people? What is your over fitted brain hypothesis? And what characteristics are what observations about sleep and dreaming that have been out there for a while? Does this explain in your view?


Erik Hoel 47:49

Yeah, so the overfunded brain hypothesis is that particularly mammalian brains and brains of good learners are always in danger of over learning of basically over committing to their vocal daily environment, which is always going to be some under sampling of what's really going on. So you're never going to have kind of the generality that you need to. And then nature, kind of realizing this at some point. I know I'm speaking anthropomorphic Lee here, but realizing this, at some point, begins to kind of create augmented corrupted versions of daily life, that are out of distribution, that they're different from what you're sampling. And this creates a a force that combats the natural tendency to become over fitted, and hyper specialized in good learners like humans and other mammals. And that it does this in a way that is at least somewhat reminiscent, like if you look at the Venn diagrams of common techniques that people in deep learning use, and you kind of put them together and you look at their overlap there overlap kind of looks like the phenomenology of dreams. So it's not so much that that that the brain that, you know, the brain is actually implementing directly, you know, the techniques from deep learning, there's probably it might be implementing some version of that. And, you know, I think that this explains some, some interesting facts about dreams. I mean, first of all, it certainly explains dream phenomenology. So we should keep in mind that that is in a sense, one of the main things that needs to be explained is just why we have the phenomenon why the phenomenology of dreaming is what it is, and it fits here in that just the phenomenology itself, just experiencing the dreams would have a function under this hypothesis. So in that sense, I think it does something that no other hypothesis of dreaming does, which is directly explained why the phenomenology of dreams is our why they is why it is and why, why that would have a beneficial functional physiological effect. So that I think is a big one. I think it fits overall and that we've always known that dreams are important for learning, but You know exactly in what way has been slightly unclear, right? So we know that sleep is important for learning. But exactly, I'm always slightly unclear. And this kind of shows why it's a rather complex situation. There's also traditional associations of dreaming to things like creativity, certainly, if you're waking up, and you're in a state of like, higher receptivity, because you've just basically been worked over, like you've just been augmented, essentially, you know, in that sort of playground, you're going to be much more creative, and so on. So I think and then, you know, there's various, you know, particular empirical studies that people have done, where they've looked at stuff like generalization versus memorization. And, you know, it seems like maybe pure sleep doesn't help with pure memorization so much as it kind of helps with cognitive performance in general. That's another big thing. And then finally, I'll just add, one last thing is that I think this hypothesis explained one of the few well known experimental effects, but dreams, which is that you can trigger dreams about a particular subject by either making people think about it as they go to bed, or overtraining them on it during the day. So a simple example of very famous case was having people who had never done it before, do like a virtual ski simulator, and they learned this ski simulator, and there's doing the simulation over and over and over again. And then you'll find that at night, they end up dreaming about this simulation, right? And then under this hypothesis, that's very kind of well explained. Because even though they're not reviewing their memories of what happened, they're reviewing this weird, dreamlike, corrupted version of the game. And then why that might improve performance also also makes sense. So I think that there's both some very particular empirical results. But I think overall, it just seems to fit well, you know, I'm open to kind of contradictory evidence. But I think it's just important to also just get new hypotheses out there that are kind of interesting and put stuff together.


Nick Jikomes 51:48

I'm wondering, you know, what kind of predictions this hypothesis might make, in terms of the types of learning deficits that you would see if you could somehow specifically or, you know, if not specifically than have a bias for depriving someone of the stages of sleep when dreams are most likely and when dreams tend to be most hallucinatory, and most sort of confabulate Tory in their, their, their, in their content? And, you know, one example that comes to mind? I'm not sure if you thought about this, but my understanding is that SSRIs are actually reasonably good at suppressing REM sleep. And, you know, I remember reading papers as a graduate student, where this was actually an argument made as an argument for why, you know, some type of learning mechanism was unlikely to be tied to dreams, because, you know, their argument was basically, well, there's been people that have been on SSRIs for years or decades, and there's no obvious learning deficit with them. People go to law school people do, you know, things that require learning, even though they're not getting REM sleep and potentially not getting as many dreams as they would otherwise? Is there some kind of prediction or something that this hypothesis would say about a case like that, where maybe they're not getting overt learning deficits, but maybe they would have some other deficit that we just didn't test for?


Erik Hoel 53:01

Yeah, so it's a great question. And I think it's one that all like theories of dreams need to face, which are just these examples of people who seem to dream quite little. They don't really occur in the natural population, although some people claim to sleep quite little, and sort of claim to sleep and dream fight little. You know, so first, there's the obvious question, which is that do they really knock out all dreams, right? And as simple example to this is that most people actually don't remember the dream. So if I asked you what you dreamed about last night, most of the time, you'll be like, I have no idea. It'd be like, maybe I maybe I didn't dream. And actually, you almost certainly did. And we can tell that by that if we take people and do some wakeup experiments. So if you were just waking up by an experimenter, and they're like, what was just going on? Right, you would actually have an answer throughout a lot of the night, you would be able to come up with an answer to that. So other this is a huge amount of dreaming, that kind of doesn't make it into 2012, just like long term memory store. And then the and furthermore, right, that you can actually improve that. So if you if you make people dream journal, which is that when you wake up and you have a dream, and you start writing it down, what you'll find very common effect is that you'll remember your dreams a lot more. Right? So it may be that some of these people who are reporting that they dream very, very little, I don't know if there's ever been any sudden wake up experiments done on them. It would be fascinating if they were so that's just to put that out there as like, we shouldn't take it as gospel that there are actually people out there who are just never dreaming.


Nick Jikomes 54:31

Right, right. I can I mean, just to give an anecdote from my own life. I wouldn't necessarily encourage people to do this. But something that happened to me recently was my partner started waking up for work about an hour and a half earlier than so the alarm was going off in the house about an hour and a half earlier than I was used to it getting up and I was not getting up earlier. So basically, the alarm wakes me up an hour and a half before I normally wake up, I immediately go back to sleep and then I wake up, you know, an hour and a half or two later. What I noticed when that that schedule change happened for us is, I started remembering, dreams that both times. So I would wake up for the day and remember the dream that I just came out of. And I would remember the one that I had about an hour and a half before that when that alarm first went off. And for whatever reason, I'm having good recall at both times. And you know what, one interpretation I could have had of that was, while I wasn't dreaming for months, and months before that now, suddenly, I'm dreaming again. But in fact, what's likely happening, I think, is I was always dreaming. It's just that that interruption is somehow allowing my memory to capture that in a way that it could not before.


Erik Hoel 55:33

Yeah, and it probably has a lot to do with I mean, I'm not a huge fan of these sort of, like modular explanations of how the brain functions. But it could very well have to do with like the ordering of what brain regions are coming on. Right. So it's kind of some like long term storage areas and like to prefrontal cortex is coming on quickly, as you wake up, you might really be able to remember the gift of the dream. But if but if that's not happening, if you have a different ordering, it might be that you're dreaming. But the way that you wake up means that you're basically getting to all the parts responsible for figuring out what just happened last, maybe with SSRIs, right? You you could have a switch, you might have some subtle difference in exactly what what's booting up there. So you get really drastically reduced free reports. But let me go let's take it seriously for a second and say like, maybe there are people who start taking your drug, and they just like shutdown dreams completely. I think actually, the overfit of brain hypothesis is probably one of the very, very few hypotheses, that would actually still make sense. So first of all, because I expect the effect to be cumulative. And this is something that is very, it is very first of all different from other theories about how how dreams function, which are generally this notion of something happens during the day. And then at night, the dreams do something to the to what happened during the day like it, it helps memories move to long term storage or something like that. It's which is not true, but like, probably not true. But it doesn't quite make sense. But you know, let's assume that that's your theory, then you have this big problem if someone stops dreaming. But now if you take someone who let's say they have a very kind of like healthy, well adjusted brain, so basically, you have a well trained, well suited brain to your environment, right? And you suddenly stopped dreaming and your stay 45. Right? To what degree would that really show up as cognitive deficits, it would probably take a while, it'll probably take a long time to really show up as a serious cognitive deficit, because quite simply, you're well fitted. So and your vitamins may be not changing that much. So really, you're not going to notice this loss for a while. So, you know, I have to be careful there because I don't want to make the theories, this sort of, like unfalsifiable thing. Yeah, I think neuroscience does not do well, with hypotheses that it can't figure out, like, within a very brief pier of time, because that's our experimental window, right? If the effect is subtle, and cumulative over over an organism's life, right? Maybe not even subtle, but just subtle, at a, at a at like a daily basis, right, but very powerful and accumulates at long run, just as data augmentation to keep you from becoming overfit. That's very difficult to figure out empirically. Because let's say I go to test it, and I find that actually, you know, I can, I can basically make people not dream, and they, they they generalize pretty much just as statistically Well, on the test the next day, it's like, well, but they already have very good well fitted brains, and they're all kind of starting from the same spot. And the effect you're looking for is some tiny, cumulative, you know, add on. And so I, you know, this is an example of why testing anything in neuroscience and coming to concrete answers is, is is very difficult. But I think that there's a room for just let's, let's at least create a set of theories that are that are out there. And that kind of fit well, with the data that we that we have.


Nick Jikomes 58:48

Yeah, another thing that I think is interesting here that you mentioned is, you know, if it's, in fact true, that dreams have this, have this function where they're trying to prevent overfitting, and therefore help with generalization. The effects of the dreams or the lack of dreams should be particularly prominent in cases where the animal or the human has to actually go into or out of a brand new context or environment. Right. So you'd expect very different results from Dream deprivation. If someone was meant to do a task, they already know how to do inside of a context they're already well fit to, versus one that requires them to move into a new context and sort of learn the gist and the rules of that new context. And it reminds me of this thing that you mentioned, these anecdotes you mentioned in the paper, which which also happened to me, which is that near to the beginning of COVID, right when COVID happened and sort of lockdowns came upon us. We were you know, effectively what that means. In other words, is we were moved out of one context and put into another right we didn't have to take the same route to work. We were maybe working from home, maybe you're not waking up at the same time of day. Your entire routine is different. In that sense. Your context is different and a lot of people apparently reported near to the beginning of lockdowns that their dreams were much more I think vivid and or weird. And so I'm wondering how you think about cases like that in the context of your hypothesis?


Erik Hoel 1:00:08

Yeah, I think it, I think it, it explains it quite well, right? Because you're, you're going through a period of, you know, new intense learning. But also in a sense, you're you've moved into an environment, which are then very much under sampling, right. So you just like a new undersampled environment, because you're stuck in your house and so on, it would be fascinating to try to dig down into that by looking at like, Dream reports from for example, prisoner, it's like new prisoners, or something like that. Or people go into solitary confinement, you know, and so on. Yeah. Again, I think that it's quite difficult to say for sure, right. So you can find some supporting you can find supporting evidence for this theory and a lot of places, figuring out ways to to falsify it is a bit more difficult. But that actually is not necessarily a problem. I mean, it might be it might be good that it might be an indication, we're on the right track, that that it seems to kind of fit well, with various things that you can kind of quickly come up with, with explanations for a lot of the phenomenon that we observe.


Nick Jikomes 1:01:10

The other interesting area that this gets into his aspects of human culture in particular, things like fiction, and stories, like why do we make movies? Why do we write novels, this is obviously something that's specific to humans. And, you know, historically, as, as an evolutionary biologist, or psychologist, it was, you know, I imagine a bit of a challenge to explain why human beings would spend so much time and energy that you could argue is, you know, wasted on something that's just untrue and made up and make believe. And so it's like, what really is the adaptive value of something like fiction or something like the various aspects of culture that that we can just think of as being like fiction? And, you know, as someone who has thought about sleep and dreaming as a scientist, and as someone who has written fiction himself that will come to how do you think about how some of the ideas we've been talking about with respect to dreams tie into how you think about why humans evolved the capacity to, to, to, to generate cultures in the way that we have, the more we're literally making up things that aren't true? What might those be doing for us?


Erik Hoel 1:02:15

Yeah, so that's, this is a great kind of summary. You know, like, this is something that I've written about, if people are kind of interested in some longer form discussion of this. There's an essay of mine, called enter the SuperS and Storium. Which is about the overfitting brain hypothesis and what it means for, for, for our consumption of fictions. My favourite thought experiment in regards to this is the notion of like, realist aliens. So So aliens come to Earth. And let's just say they're realists. And by realist, I mean that they just hate fictions. To them. Fictions are lies. So if someone says to them, you know, this person did x, you know, Harry Potter killed Voldemort, right? They'd be like, You're lying to me right now. Please don't do that. That's unacceptable. We should just speak truth things and facts about the world. Right? And you could you could is actually a pretty imaginable culture. That that does is they just don't have any movies or books or so on. Right. And we're meanwhile, like, obsessed with all these stories, these these crazy stories. And it is very strange, right? Because it seems like well, we have met selective pressure on us both from you know, evolution, but also just from the from the modern world. And, you know, for some reason, we spend a huge deal of our time just consuming fictional stories, and I certainly am. Someone who not only loves those, you know, I write I grew up in an independent bookstore and grew up in my mother's bookstore, I sold books when I was a when I was a teenager there. And so my whole life has revolved around books and our own stories, and our own fictions, which are, according to these realist aliens lives, right? And so what how might we explain to these real estate aliens? Why we so enjoy loving fictions? And I think one answer that maybe the realest aliens would come up with and I think maybe Steven Pinker would agree, is just that this is, it's all just cheesecake, right? Like, why it's like asking, Why do humans love cheesecake love cheesecake, because it's got a lot of fat in it, we've evolved to love fat, it's got all the fat, it supernormal stimuli, we love it. Similarly, you might say, well, fictions are just really interesting versions of life. We love that stuff. So we really dig into it. And this is kind of like the null hypothesis of it. But it kind of implies that when people are consuming a fiction, they're doing something kind of dumb, right? It's like, like, junk food. Exactly, exactly. That's a great term. And but this, I think, gives gives an alternative view on it. And again, I'm I'm not saying that this is the only justification for the production or consumption of fiction, right? So there might be all sorts of justifications, but like a very hard nosed scientific, biological one is very different than that. And in this case, it might just be that listen, when you're an organism and you're learning you're constantly becoming over fitted to your environment. And what you need is some out of distribution samples. And so when you go and you watch Dune, right You get the planet Arachis in front of you is a very different like novel experience that is literally combating the, the shrinking focus of your own cognitive abilities, right? Because every day you shrink a little bit because your brain is excluding the superfluous, and just really getting good at exactly what you do, which is, you know, mostly sit in front of a computer like, like all of us, or you work a job, but that job is not that variable, right? So no matter what, right all humans have this shrinking cognitive focus, and maybe fictions and the arts in general, are, you know, aesthetically pleasing and, and pleasing to us precisely because they combat us. And then that means that maybe someone who goes home and watches TV isn't being as slothful, as we we kind of might innately think that they are right, like, like, maybe there's actually some work. Now I'm not claiming that it's all good, or all work or anything like that. But maybe there's actually some, in a sense by homeostatic work going on cognitively, when you watch a TV show, because it's something that's or read a book, or consumed the fiction because it's, it's so out of distribution that you just experiencing, that is going to help prevent you from running into these cognitive deficits and not being able to generalize and so on. And so that's, I think, taking the overview brain hypothesis seriously and applying it to to literature and fiction. Now, that is very speculative. But I think it's I think it's quite interesting. I think it's worth talking about.


Nick Jikomes 1:06:30

Yeah, I mean, if we take you know, one of the things that I think is attractive about your hypothesis here about dreams is it's probably safe to assume and there's, I guess, probably some evidence to suggest that, you know, humans are probably drinking a dreaming more than most other animals. And so you might say, Well, why is that? Well, in the context of everything you talked about, about, you know, overfitting and generalization and how that connects to the ability to learn across contexts. You know, as I've discussed on this podcast with others before, you know, the human niche is really niche switching, more so than I think any other species, humans are adapted to being able to literally move from one habitat to another and adapt to a brand new context. And to the extent that an animal species has that sort of ability to switch niches, you would expect it to have more need for being concerned for having a brain be concerned about overfitting and generalization.


Erik Hoel 1:07:28

Yeah, precisely. And maybe, you know, we, we, like my, the analogy that I've given for this is that, yeah, there's a hypothesis that one of the reasons why humans are able to have such big brains is that we move our digestive systems outside of our stomachs, every every other creature has to digest inside themselves, right. So they have to have really big stomachs, which is a really big problem, because you only have so much energy to go around. You can't you can't, you know, you can't even get them out. Like if humans, big stomachs, you can even get them out right? When they're born. But what we do is that we cook our food, and that's an artificial digestion that we've, we've we've made into technology. And so similarly, it might be the case that dreams are such an incredibly important cognitive function. We simply don't even dream enough in life. And so we kind of artificially supplement by dream having people dedicated dreamers, like filmmakers, like shamans, like poets, you know, who dream these artificial dreams, and then we consume them outside of it in such a way that we have moved our digestive systems outside.


Nick Jikomes 1:08:37

So, you know, all while, while we are on the topic of fiction, my understanding is you've actually written or you're about to publish a novel. Can you talk about what the novel is and what it's about? But also, why or why have you written a novel as someone who is a scientist?


Erik Hoel 1:08:54

Ah, yeah, so the book is, is out, it came out a little bit earlier this year, it's called the revelations, you can find it possibly in your in your local bookstore, you can certainly find it at like a Barnes and Noble. And it's, of course, it's on Amazon. And it's a novel that's actually set in the world of science. It's a it's a murder mystery, set in the world of consciousness research, and one of a young scientist dies and the other scientists kind of formed this amateurish new Irish investigation into their death. And so, so as to as to why a novel, you know, sure, surely, I could have written a nonfiction book. You know, I think one novels, I can give a couple reasons, right. So one novels are, I think, much better at giving people a feel of of something, right. So novels evoke the transmission of conscious experience. And there's something very interesting about about science itself. I think science itself is a worthy subject of literature. You know, in a similar way, that, you know, forgive the comparison because it's deeply unfair to Melville, but innocent citizens are way that like, you know, Herman Melville might set a novel inside the world of whaling. And that becomes itself a literary enterprise. I think that science itself is a literary enterprise and, and not like science fiction, but like, the actual process of discovery itself has all sorts of kind of human drama and human moments in it. And that was something that I really wanted to capture. And, you know, with that in mind, I'm also a big proponent of, of not separating the arts and the sciences too much. So this the subject that I run, the intrinsic perspective, covers, you know, sciences and the arts, in various in kind of equal equal measures. And I've, you know, I've always, I've always written and, you know, what, I won a couple awards for my writing, I was lucky enough to win in my early 20s. And that led to me being able to get a get a novel published if I wanted to, and, and I just felt like, I had to write it, and I had a big kick out of writing it as well. And frankly, the relationship that you have with a reader, when it's a novel is, frankly, a lot better than when it's nonfiction than fiction. You're kind of like the departure of inflammation, right? But, uh, you know, you get to be a little bit of an artificial dreamer, right? When you're when you're a fiction writer, and I think that it's a bit more of an intimate and more than interesting relationship to me.


Nick Jikomes 1:11:25

How long did it take you to write it?


Erik Hoel 1:11:28

Man, you could, you could almost judge it as 10 years. I mean, there is a sense to which I was, I was writing the book before I even became a neuroscientist, because I was I was looking around for topics for novels. And I thought, you know, what I would love to read, I love to read like a novel really, like deeply set in the world of science. That would, that would appeal to me, like hugely, and I just couldn't really find other than maybe a couple authors like Richard powers, anyone doing doing it the way that I would want it done. And so you know, and it says, I chose what I thought was one of the most interesting and biggest problems in science, which is the problem of how the human brain generates conscious experience. And, and I got very lucky and ended up working in one of the top labs for that during my PhD, but I was at the same time, you know, I was writing and, you know, probably the main text took about took about four to five years, but then all the notes and associated other stuff, you know, it all ends up being, you know, about a decade.


Nick Jikomes 1:12:28

Yeah, it'll be interesting to read, it's sort of reminded me of the book, which probably, I'm guessing most listeners will not have heard of, but it's very popular among scientists. It's called the eighth day of creation by Horace Freeland. Jetson, which is a nonfiction book about the development of the field of molecular biology in the early mid 20th century, and the discovery of the genetic code and all that. And even though it's nonfiction, I think the reason it's so popular among scientists is, is that it does have this kind of literary quality to it, where he really did a good job of making the human drama. And the characters come out in a way that does feel a little bit more like a fictional story than just sort of a dry regurgitation of the facts. So if you're interested in the history of science, I recommend that one for sure.


Erik Hoel 1:13:14

Yeah, that's really interesting. I definitely think that, you know, at least some of the motivation to me to write the book is that scientists, you know, if you look at like the average depiction, scientists on TV, they're basically these like sexless brainiacs, who have, you know, no personality and are generally quite boring. And then like, weirdly experts at like, you know, every subject, right, like, somehow they have like this perfect knowledge of all these different things, versus science. Whereas I generally have found that I really at the top levels of science, you have people who are like, inspired to do science because of their love of music, right? Or they have you have people who, who, their, their reasons are just as kind of human and understandable and fallible and interesting, as kind of the more robotic automaton that kind of populate traditional culture. So, so at least in part, you know, this is a book that's, you know, set within science where there's like, jealousies, there's competition, there's, you know, a lot of kind of intrigue, because, frankly, I mean, outside of the murder, obviously, you know, that, that that's been my experience that that at the top level science is very, very human. And I'm not sure it could be any other way. Like I do you think that there's kind of like a mad poetic instinct behind science that is shared with with the arts in particular. And that this is something that we lose sight of nowadays, and maybe over rely on sort of thinking of scientists as rational and select for people who are kind of super rational when actually maybe you want to select for people who are like, really creative and can come up with like, really new and strange ideas, because nature doesn't always kind of work rationally or proceed kind of linearly. Yeah,


Nick Jikomes 1:14:59

yeah. The Dream imagery there sort of suggests itself. But that's been my experience as well, especially at the top levels of science. It's very, there's a lot of interesting personalities, and there's all of the jealousies and ambitions and other human emotions that you would expect in any good. Any good drama.


Erik Hoel 1:15:18

Yeah, there's certainly certainly a big, big egos as well. But again, I'm not entirely sure that this that, that, you know, stuff at a very high human level doesn't involve all that sort of stuff, right? Like, it's, it's like the human human emotions are an engine, right. And the engine produces various things can produce science, it can produce the arts and so on. But like, we shouldn't forget that the engine, despite the fact that science looks dry, and rational, it's an engine, it's driven by and its components, human human emotions. So you know, to me, that's what made it so literary. And that's what I was trying to capture in the revelations.


Nick Jikomes 1:15:59

The other thing I wanted to talk about was, your writing more generally, so So you do a lot of writing that isn't directly, you know, to do with your professional academic science work. You've written the novel, as we've just discussed, but you you're also on substack. That's, that's a place where I do a little bit of writing as well. And a lot of people have migrated to places like substack, where you're sort of writing independently outside of the normal structure of a journalistic institution or some other institution. And a lot of people have talked about this concept of unbundling and re bundling that we're seeing, you know, across industries, it's being driven by by internet technology. And so on the topic of what's sometimes called the Great unbundling, and the death of mainstream media. Can you just start talking about why you decided to cut the cord as a writer and kind of go to substack? And do something entirely on your own like that?


Erik Hoel 1:16:50

Yeah, sure. I think it's a great time to do something like that. Because, well, just in a sense, just because everyone's doing it, right. But like, as for me, personally, I had written for, you know, I've written essays, I love the form of the essay, I think it's, I think it's one of the most elegant kind of ways ways to do prose I like, I just a huge fan of essays and essays. And the Internet is a perfect is a perfect place for that, like the essay is the art form on the internet, right? I mean, when people wake up, they'll read an email, they'll go through the social media, but you know what they'll also do, like right there on the phone, they'll read an essay, and they are reading fiction, they aren't watching a movie, they aren't doing these other things, but they'll they'll read an essay. And that's very, very powerful. And so I really been trying to get into the forum more, because I think it's a form of my time, right? It's like, like, like, you should adapt yourself to what's around you, right? Like, if you're, if you're a playwright and lisabeth in England, right? You should probably do sonnets, right, you probably do, you probably do what the forms that are popular, right. And so I think now is the golden age of the essay. So I've really been trying to get into more, and I originally used to write for various publications, I published stuff in the Atlantic, I published stuff in The Daily Beast, and I would work with these organizations. But frankly, the experience of working with them was generally not very great. And the hoops that I would have to jump through were immense, just to get something that I didn't hate, published and out there. And I think with substack, you know, I publish once every other week, I publish basically on a topic that that interests me, I have one coming out tomorrow, on on on on Alec Baldwin's unfortunate, accidental shooting of that woman on his movie set and more more moral moral on the, the innate suffering of human life, like just the fact that the lot of a lot of human humanity has to suffer. And we kind of distract ourselves with with media. And I'm aware that you know, that is not something you know, that I'm speaking about as an athlete, that academic I'm thinking about it as, as a writer, but I do try to try to keep that generality as a writer because again, I think maybe it's important to not become too over fitted.


Nick Jikomes 1:19:15

Why do you think the essay is the native written art form of the internet?


Erik Hoel 1:19:22

Well, because if people will read it with their head on a pillow, so we nowadays most people, they wake up, they reach from everywhere, your phone to their face, right? And we can all like, we can all both, either do it and complain about it, or just maybe not do it out of sheer spite, right? And then and complain about it, right? And we can like try to, you know, pretend that that's not the world we live in. But that is the world that we live in. And it has its big negatives, but it also has some positives. And one big positive is that there is an art form that people 1000s of hours If people will read if you put it out there, and that's a good article or essay, and people will really spend time on that. And so in that sense, I mean, you could say the blog is the is the native art form, but I'm not sure that blogs by themselves, but I say blog people mostly think of like, you know, life updates, you know, stuff like that. Kind of like minor things. But like, really well crafted pieces of prose, that are kind of reflective and essayistic, and literary, but draw from different sources that I think is an art form capable of really thriving on the internet in a way that almost nothing else can. And so to me, it's like just put, put your energy into the thing that's going to work, like, no one's going to read my, you know, experimental poetry, right? Like, nobody cares, right? And that's, that's fine. Maybe we're in a completely different culture, where everyone was doing experimental poetry at a very different age. Like maybe we were in Roman times, putting up poetry pamphlets, everybody's doing it. Right Niro's doing it, right? Like, every, like, maybe the photolysis doing it, right? Like, maybe that would be great. But it's like, adapt yourself to your time. So in this time, the best thing to be is an s&s because that's where you get the most attention. And, and, and I don't mean that as in, like, you just want the most attention. But that's, that's when you can actually reach readers and give them something back that's relevant to them, and that they're receptive to. And that's really, really important as a writer, I mean, with a writer, you're constantly, you're constantly serving, and you don't really get anything back. Right. And and, and the the online essay, and the series of essays that I think I'm going to kind of try to keep doing my entire life, I'm just going to try to keep writing them every other week for the rest of my life. Yeah,


Nick Jikomes 1:21:45

one of the things I've enjoyed about reading people on substack is, you know, even though there's a very large spectrum of writers who I'm reading, and their work is about different things, there's kind of like this overall flavor that's different about going through that avenue, as opposed to like a mainstream media source. And I assume it has to do with not having to go through the same editorial filters that one would have to go through, if you're working through an editor at, say, The Atlantic, or the New York Times or whatever. Can you talk a little bit about that experience? How does like the content and the feel and the attitude of your writing differ on substack? Simply because it kind of goes directly from you to the reader, rather than going through that kind of filter? I assume there's actually maybe both pros and cons to that.


Erik Hoel 1:22:31

Yeah, there are definitely some cons. There are definitely the occasional con but I think in general, it's it's an amazing ability to avoid the like terrible baleen filter that is, you know, the editor that you're working with, at the New York Times of the Atlantic, or whomever? Because, first of all, those like, what is the new york times it's just a blog, it's a blog, a lot of people work on it, but it's just one big blog, it's just the substack. Like, there's no, there's no difference between those two things not really. And, you know, once you realize that, you realize you're just in the hands of this, like cadre of editors, who, you know, like I've met, that editors who work in New York Times, I've gone to parties with them, they're just people, they have no, you know, they're different good at their jobs, but they have no like, innate special insight or ontological access to the truth, more so than anyone else. So in the end, you're just being filtered through someone else's perspective. And that can be a very good thing, I think, in in, in, particularly in people who are maybe overly ambitious, and can't curb themselves and can't edit themselves. But even if even if that's the case for you, there are various things that you can do like writing pieces ahead of time and sitting on them. And then reading them again, there are various kind of common practices you can do. And frankly, most for if you have a good first reader, write your first reader. And I have people in my life like that you've just have a good first reader, they can give you good feedback, the sort of feedback an editor would give of like, yeah, this paragraph isn't working, or this is a bit too long to shorten it a bit. And they're going to give it to you a lot, a lot nicer than working with editors. But I'm, I'm very hopeful about this contemporary moment and, and getting getting, you know, enough people online that you can actually do stuff like that. And then finally, one last thing about that I'm very hopeful about is that, you know, there is there is a sense in which people have begun to make their livings off of this because it turns out that people will pay for quality writing, I don't have any in planning to go paid on myself second, it's free substack so anyone can sign up for it. But but you know, one day, if possible years now that I might and it's amazing that people can now actually make direct money off of that you might not think that writing on the essays can make you money. Some people make a lot of money writing essays on the internet and that's because there's no bureaucratic filter right? If I if I write for The Daily Beast, I get a $400 check. You know, eventually and I go through, you know, hell, basically making sure that what comes out is not something I'm embarrassed by. And again, that sounds like I'm just producing amazing, I realized that this sounds like I'm just producing amazing stuff and it's getting like destroyed by, like the establishment or whatever. But to be honest, there is some there is. This simple truth is, is that most people are doing stuff, many, many people that is more ambitious or interesting. But they, they have to fit it into this very particular, you know, square round square hole that has to apply to their entire audience. So of course, anything you do, that's a little bit more literary, you know, like, if you look at what writers have had to do historically, to get interesting, well written prose out there in big magazines, and you should see the editors that David Foster Wallace used to send to like the editor at Harper's, write their stuff, like I will find your grave. And like dig it up, if you edit this like one more time. And that's because you know, you're, you're going you're going through this, you're going through like a great a great intestine.


Nick Jikomes 1:26:07

Well, one more time, what, what's the name of your substack? What's your book? And where can people find you?


Erik Hoel 1:26:13

The easiest way for people to find me is just look up Eric Kohelet, E, Ri, K. H. o al on Google, and my website and links to the substack the intrinsic perspective, I really encourage people to apply this to try because it's only once every other week. So you just get an email every other week. It's really it doesn't bother you. And generally, it's something pretty interesting that or at least high thing, certainly, obviously. And then the book The revelations, if you want like a deep dive into science, if you like, you know the name of the rose by Umberto Eco or you like kind of like big kind of intellectual books that are set in some sort of detailed world like here, the world of science, then that's something for you.


Nick Jikomes 1:26:58

Excellent. Well, Eric. Well, thanks for joining us.


Erik Hoel 1:27:01

Thank you so much for having me. This has been really fun.



8 views0 comments