Donald Hoffman: "Evolution is like a videogame" https://www.ted.com/talks/donald_hoffman_do_we_see_reality_as_it_is (Quote is from , which I'm not sure if I'll finish.)

Reminds me of “The Interpreter” in Gazzaniga’s Gifford Lectures. IIRC, he says that the brain makes up an interpretation of what happened - after the fact. https://www.ed.ac.uk/news/all-news/gifford-071009

Yeah, going to be honest, not a fan at all. In the most charitable version of this, it is just Kant's view dressed up with some "science" to back it. In the worst it is a global skepticism that is self defeating. He definitely isn't blowing any empiricists out of the water, but just repeating things philosophers have explored in much more depth and rigor elsewhere. To try to be concrete, the argument in the TED talk is just bad. I ran a game theory simulation and found that in my simulation reality tracking was not the most fit setup. In the simulation, the non-reality tracking "organisms" out competed all. Therefore, in actual reality we don't track reality." There are so many steps missing here. hopefully it is clear that this isn't a valid argument. But it also becomes a mess when applied to itself. The evolution that "follows these equations" is the evolution in the phenomenal world, the world as we see it (the desktop in his metaphor). But what we are interested in is not that evolution, but the real evolution, the one that actually shapes our cognitive faculties. Why think it obeys these equations? By this very commitment, we have no access to that evolution. So, we have no reason to think it follows the equations. So when coming to accept this view, we get an undercutting defeater (a belief that undermines the support we had) for the very belief we just adopted. Once we adopt this interface view, we realize we didn't see Evolution, we saw the Interface that we call evolution and have no reason to think this is the evolution of reality, it is merely the evolution of our interface. Therefore, it does not follow that Evolution causes us not to see reality.

(Were you already aware of this?) There is a you and a me, we aren't just interface constructs. We can't trust our perceptions, but consciousness is consciousness and death is death. And if you have life and death, you still have evolution and fitness. Organisms reproducing to create new consciousness is not just an interface construct. I don't (yet) see how this is self-defeating. If somebody showed how the laws of physics emerge from the laws of a cellular automaton, I'd call that science. It's a model that explains observations, that you can use to make predictions. Did Newton or Einstein do anything more? An equation is just a kind of model. A lot of modern models have a lot of variables, which brings in questions of are we just p-hacking. But that doesn't refute model-building as an endeavor. It seems to me that persuasive models are the currency of science, along with data. Now, to be fair, in the second link he says he's trying to find the underpinnings of evolution, which feels a bit woo. I'm not sure how you would do science on that question. He gives up all the experiments and theorems he's built up over the last 20 years and now has to start from scratch. I'm gonna read his book to better understand the plan of work.

I was not aware of this.

Organisms reproducing to create new consciousness is not just an interface construct.

On his view everything is interface, because that's the only thing we have access to. We have no access to the world at all. Our cognitive faculties are not mapped to truth at all. Nothing we know is true. As for why this statement is interface? Organisms are things that exist in space and time. As he tells us, space and time don't exist. They are interfaces. So if organisms exist in space and time and space and time don't exist, organisms don't exist. If I'm honest, this kind of stuff bothers me so much. I see it as pure sophistry. I truly recommend reading Kant or Berkeley, or James or Pierce or Rorty, or Putnum or any philosopher in the Kantian or pragmatist bent over this stuff.

Organisms don't exist in space and time. We do have access to the world -- through math. He's making the strongest case I've yet seen for model building over empiricism, reason over the senses. Why would I read philosophers over this? They don't make testable predictions, he does. It's apples and oranges. It's super useful to see points of agreement and contradiction, but I don't see one side obviating the other. I guess I'll stop here. (Though really if it bothers you, shouldn't you just stop reading this thread and control your own reality? 😂)

This is literally just a bad rehash of what philosophy has talked about for so long. It is literally a strawman version of what Kant said.

Except Kant could be right, Kant could be wrong. With this guy we can check. Seems worth rehashing.

This is an old argument that people have been making for decades now. There’s nothing new. It is an evolutionary debunking argument. You can find tons of them in the literature. Maybe you are right that I shouldn’t have such a negative reaction. But I do. This is too much of a nerd-snipe topic for me. If you want to read him go ahead. But I’d at least recommend reading critiques of his work as well.

Donald also discussed this idea a few months ago on the Lex Fridman podcast

It's very weird to me how many intelligent people with technical training have not gotten past the "what if we are brains in vats?" stage of philosophical investigation. I'm with @Jimmy Miller on this one.

"Can we know reality as it is?" is indeed a question philosophers have been discussing for quite a while. And in this general form, it's outside of science - you can't check this empirically. But specific aspects of how perception (human or otherwise) is based on both incoming signals and a priori models are very well accessible to scientific inquiry, and add a useful perspective in my opinion (but then, I am biased by being a card-carrying scientist).

@Konrad Hinsen... which is, itself, a philosophical position. A position I happen to share! That said, the logical conclusion of Hoffman's argument is that you can't reliably test Hoffman's argument, having no transhuman means to do so. The same situation occurs with "brain in vat"/simulationist (and similar back to Plato) thought experiments. So, some of us decide (philosophically) that we're going to rely on the evidence of our senses because it's the best we can do as agents in the world (Heidegger). After which, all the standard cog-sci limits of perception/cognition stuff that's already in the textbooks applies, plus a steady stream of new and fascinating findings. 🤷🏻‍♂️

@Jack Rusher That's indeed the weak point in Hoffman's arguments - he should have applied them to themselves, or at least mentioned this issue. But... TED is TED. My main issue with "brain in a vat" is that it's uninteresting. A dead end in thinking and exploring. At the opposite end (as I see it), there's Terence Deacon's teleodynamics theory (see his book "Incomplete Nature"). Not quite empirically testable right now, there's a lot of preliminary work to be done, but empirically testable in principle, and I am confident that the road to there will yield interesting insights.

Jimmy and I went back and forth on it a bunch in private, and I'm not eager to do that again. We did find that our disagreement hinged on different interpretations of the words "reality", "true" and "false". I'm not going to tag him in case he's (understandably) muted this thread. Jack: • Were you aware of this talk already? Is your position coming from having thought about it years ago when he did the rounds? Or did you actually watch the whole TED talk after I posted it? (Just the 20-minute talk. I'm not expecting others to watch the 2-hour interview.) • Can you first state what you think his argument is in your own words? In particular, do you think he considered your previous comments? I draw attention in particular to the experiments he describes, and the analogies with the train, the cliff, the lion. • If you think he's obviously vapid, can you try to imagine what nearby consistent+testable thing he might be saying? It's totally fine if that seems like too much effort, and you're sure there's no there there. We can just agree to disagree and move on.

@Kartik Agaram I ran into his stuff years ago, when it first made the rounds. I'll watch the 20-minute talk at 2x out of fondness for you and continue this convo in another channel. 🙂

After reading the paper [1] (which I hadn't gotten around to locating until @Jack Rusher shoved it under my nose), I still don't quite see (heh) the philosophical refutation. However, it seems clear to me that: The full generality of his theory is not needed by his experiments, and not testable with similar experiments. The screenshot below feels like a representative example of the sorts of perceptions his experiments evolve. It shows the organism's perceptions "folding" the real resource level around its optimum desired level of some resource. This seems reasonable. But it doesn't seem reasonable or testable to go from "the perceptual space is not identical to the world space" to "there is no world space," to focus on just one of his broader claims. As a concrete example, I have a hard time imagining the world not being a metric space (https://en.wikipedia.org/wiki/Metric_space). Hoffman would agree that even if there isn't really a lion, there's something that can affect your fitness -- and this something "moves" unlike the static resource. More precisely, your perception of where you are is decoupled from the fitness of this location. This bank of the river has high fitness if the lion's on the other side, but low fitness if the lion's on the same side. It's hard to imagine any static mapping from a non-metric space to a metric space being consistently adaptive, i.e. consistently yielding good fitness. I think the null hypothesis here should be that selection will tend towards noisy but unbiased mappings when it comes to space simply because of the wide variety of scenarios an organism will get into in its life. Perceived distance will correspond to real distance. It's hard to imagine a world in which this null hypothesis is false. (Though I did ping Greg Egan in hopes of nerdsniping him to come up with one[2].) Obviously, further comments and refutations most appreciated, both on the conclusions and on the means by which I arrived at them. I don't science/critical-think too good. [1] https://link.springer.com/article/10.3758/s13423-015-0890-8 is from 2015 and seems like a decent summary of all their past work. It covers all the ground in the TED talk. [2] https://merveilles.town/@akkartik/109856299739181138

@Kartik Agaram There have been various proposed scientific theories/models that include alternatives to metric spaces. The most recent and ambitious one is the Wolfram Physics Project: https://www.wolframphysics.org/ In all these proposals, metric spaces become pragmatic approximations to reality, rather than a fundamental part of it. On the philosophical side, just look at the number of Wikipedia pages on "realism" to see the importance of such questions: https://en.wikipedia.org/wiki/Realism

I tend to think Wolfram's theories and string theories all don't go so far as to say, "there is no space." The metric thing is just a concrete example to think about, though then again I don't know how to think about where in Wolfram's theory the space becomes metric..

image.png

Metrics are derived from measurements, and measurements at small scales (small lengths, small energies) have limited precision due to quantum effects (the “uncertainty principle”). Wolfram’s theories are for even smaller scales, so metrics are not relevant.

I mean metric spaces in the topological sense. Basically I'm trying to get at least a rough sense for how Hoffman's lay language "there's no space" might be meaningful. Wolfram wouldn't say there's no space, I think, just that 3D space at large scales is an emergent property of a very different space. If space just has more dimensions than we perceive, that doesn't feel like justification to say there is no space in the world. Projection feels like a familiar transform. However, if the transform is much more complicated, such as the one described in http://akkartik.name/post/wangs-carpets, then the statement starts to seem meaningful.

The “there’s no space” part is the reasonable bit about his stuff. That’s just Kant. (Space is a category we impose just like causality)

My understanding of Hoffman's "there is no space" is: "our perception makes us see space, but that's no reason to believe space is real". From a scientific point of view, the statement "there is no space" makes no sense, as it is not a testable hypothesis. As @Jimmy Miller points out, such statements belong to the realm of philosophy.

The space we perceive is not real, sure. We've learned this multiple times in the past. Babies learn depth perception to go from 2D to 3D. Humans figured out our 3D is an approximation for 4D spacetime. We also figured out solids are all mostly empty space. Maybe string theory will tell us there are 7 or 11 or 438 dimensions. So, "your senses are lying to you" is fairly tame at this point. A word like "reality" gets in the way here. If there's something that affects fitness that our perceptions are transforming for us, it seems more fruitful to ask how complex the transform is. We agree that it's not testable. I'm trying to see if it's even coherent+new. Is there an interesting fiction to be written here? Perhaps I should go read Kant. But I tend to make more progress muddling through by myself or thinking things through with lucky folks like you all 😄

Reminder: all evidence we have for evolution and fitness existing in the world at all has arrived via these senses that we cannot trust.

We agree that it's not testable. I'm trying to see if it's even coherent+new.

(Hoffman doesn't agree that it's not testable, which is one of his major problems) The incoherent part are not his metaphysical claims (claims about how the world is) but his epistemological claims (claims about how/what we can know). The epistemological claims he advocates for are incoherent. The metaphysical claims are not new. He is just an idealist in the Berkeleyan sense. The material isn't real, What is real is the mental. Objects aren't material things, they are things of consciousness. If you want a very readable survey of idealism, Chalmers has a nice paper on it (even mentions Hoffmans view) https://philpapers.org/archive/CHAIAT-11.pdf

(Hoffman doesn't agree that it's not testable, which is one of his major problems)

Yes. I changed my mind at https://futureofcoding.slack.com/archives/C5U3SEW6A/p1676736793668239?thread_ts=1676267575.200009&cid=C5U3SEW6A. He has some claims and he's testing something. But he's not testing his claims, and his claims are not testable.

I guess I wasn’t confused because you said you didn’t see the refutation. His major claim of innovation is that this is all testable and therefore “science”. That was also your major point early on in the conversation. If you agree he can’t do that, isn’t that the refutation? That isn’t to say some of what he says can’t be right. But his project is supposed to different from all these earlier philosophers by being testable.

I think he's wrong, but I also don't understand the refutation Jack and you made about there being something incoherent about even attempting to find an experiment to test here. I find my comment to be a simpler refutation. I'm not trying to find a contradiction, just a counterexample. But it seems nobody understands my refutation either. So it goes.

I agree with you that his argument for no space doesn’t follow from his premises. That there are many alternative schemes that could be true. He hasn’t given us any reason to prefer that there is “no space” over “there is space that is different from what we perceive”. My point is that the same thing is true for all of his arguments. Does he have a good an argument for consciousness being fundamental? Does he have a good argument for game theory simulations warranting claims about what happened in the actual world? Does he have a good argument for why we can know about the fundamental nature of reality, despite the fact that we can’t perceive the external world? The flaw you found with his argument for no space is rooted in a methodology flaw he uses throughout his work. His grounding for his view is that it is scientific and testable, but he undermines all of what would make his claims testable. Someone might be able to test claims that sound like his, but not given all his commitments. How can he even interpret the results of his computer simulation if he can’t trust his senses? How does he know that he sees the right numbers? How does he know that evolution occurred at all? How does he know that game theory mathematics corresponds to the reality we can’t see? Couldn’t that itself be a lossy interface? All the things he wants to claim are true, he undermines

Even if it fails as science it seems interesting as philosophy, which doesn't have to be testable[1]. You're probably right that all the good bits are from Kant, and that I lack the philosophical sophistication to understand the contradictions caused where he diverges from Kant. [1] I think of philosophy as akin to fiction. E.g. separating metaphysics from epistemology. I'm unconvinced after this conversation that that's a fundamental distinction. It is at best one of several frames. But obviously I don't know much here and I'm even less interested in getting into it. So it's an extremely low-confidence, low-priority thought.

I want to defend philosophy here, but I doubt I can do that justice. Rather than try to defend it I will just say what I find it in. Philosophy, particularly Analytic Philosophy, contains by far the clearest, most precise thinking I have seen anywhere. No where can you find people who care more deeply and are willing to consider any an all objections, counter-examples, and possibilities. The thinking that goes on the programming world looks weak next the rigor of these arguments. Reading works of philosophy has (in my opinion) been the single best thing for sharpening my thinking. For being a better programmer. For being a better thinker. Philosophy is for some reason seen as “old fashion” or what we did “before we had science” when in fact it is an incredibly rich field full of interesting insights and continue growth and change. We all have thoughts about the way the world is, the way things relate to each other, for example your thought about metaphysics and epistemology being on of several frames. That itself is a question of philosophy and one much explored. As you dive deeper and deeper into philosophical arguments, you begin to see ways in which commitments you have made are connected in ways you had never considered. Your views of ontological matters affect your views of ethics, your views on the foundations of logic, have affect your views possibility. It is truly hard for to me to understand why people have such a negative attitude towards philosophy. How could we possibly do science without it? Why wouldn’t we want to explore questions like, what is the criteria to know something? How should I change my beliefs if I find someone I consider to be an epistemic peer disagrees with me? What are the foundations of math and logic? Are there any? What are some ways language functions? What are numbers? Does the external world exist? Why think that only things that are testable matter? Is testability a good criteria for what counts as science? (it general isn’t considered to be) Philosophy is to me by far the most interesting thing we can spend our time reading and thinking about. It birthed science, it birthed programming. It continues to march forward ever expanding into new territory, ever changing, always questioning.

This is getting off thread, but I'd love to hear what initial experience spurred your love for philosophy.

@Jimmy Miller My personal (fully untestable!) hypothesis on how and why people got into "science is great, we don't need philosophy" is the enormous growth of science since the 1950s. Based on simple and unconsciously held notions of realism, science made enormous progress simply by injecting more resources. Philosophy is a bit like religion in that people turn it to when they face problems, but downplay it when all goes well.

@Konrad Hinsen I would have dated that phenomenon a bit earlier, to the rise of https://en.wikipedia.org/wiki/Logical_positivism. Success in 19th century physics led many disciplines to assume that everything was as easy, only to find that they haven't the predictive power they hoped (economics is a good example here).

@Jack Rusher The rise of positivism was indeed the preparation, but it was a time of active philosophical debate around science, in particular following relativity and quantum theory. After WW II, science shifted into industrial production mode. That was also the time when terms like "research unit" and "knowledge production" started to be used.

There is no theory we may hold and no observation we can make that will retain so much as its old defective reference to the facts if the net be altered. Tinitus, paraesthesias, hallucinations, delusions, confusions and disorientation intervene. Thus empiry confirms that if our nets are undefined, our facts are undefined, and to the “real” we can attribute not so much as one quality or “form.” With determination of the net, the unknowable object of knowledge, the “thing in itself,” ceases to be unknowable.

I'm writing something about artificial neurons at the moment, and thought of this thread when I noticed that McCulloch and Pitts were smoking some dank philosophy in 1943. 😹 https://www.cs.cmu.edu/~./epxing/Class/10715/reading/McCulloch.and.Pitts.pdf

Yeah Pitts was a child prodigy that pointed out flaws in Russels Principia Mathematica at age 12. At 15 he went to go study under Rudolf Carnap. This paper I think is published when he was 20. Thanks for the quote. That papers always been in the background for me. But maybe worth cracking open sooner.

Yeah, the only three references are Carnap, Hilbert, and Russell/Whitehead. 😆 The paper itself is quaint in retrospect, tbh. They go through quite a bit of logical self torture to demonstrate that their grossly incorrect model of neural networks is isomorphic to a subset of logic gates (using other frames to arrive at this), then make some grandiose statements about the importance of this work to philosophy and psychology. 🤷🏻‍♂️