The title is from a blog post by Brian Leiter. Leiter links to an article in the LA Review of books: Imitation and Extinction: The Case Against Reality. The article is written by Donald Hoffman.
We have discussed the general topic before, in several threads. So maybe this is a good time to revisit the topic.
Hoffman asks: “I see a green pear. Does the shape and color that I experience match the true shape and color of the real pear?”
My take is that there is no such thing as the “true shape and color of the pear.”
It is a common presumption, that there is an external standard of truth. Here, I mean “external to humans”. Truth is presumed to come from somewhere else. And our perceptual systems evolved to present us with what is true.
As I see it, this is backwards. Yes, our perceptions are mostly true. But this is not because perception is based on truth. Rather, it is because our human ideas of truth are based on what we perceive.
Open for discussion.
I think that from a philosophical viewpoint, misrepresentation and non-veridicality are, in essence, addressing the same issue.
A lot of the discussion so far misses the point that Hoffman is making because he has a clear, mathematical definition of veridicality and his modelling address why he thinks evolution will not address it.
The introductory sections of the followings paper are helpful, even if you (like me) did not want to take the time to work through the math in detail.
https://link.springer.com/article/10.3758/s13423-015-0890-8
ETA: On the red-green issue: misrepresentation is usually discussed by assuming a standard or normal perceiver; “standard” then needs to be elaborated in the context.
Cue a moron who thinks that is what he is saying.
Corneel,
The OP pushes Neil’s idiosyncratic views regarding perception and truth, using Hoffman’s article as a starting point.
The article argues that any veridical perceptual strategy is always outcompeted by at least some non-veridical strategies — but it uses a narrow and unrepresentative notion of veridicality to make that argument.
By Hoffman’s standard, a perceptual system that highlights fitness-enhancing (or fitness-harming) aspects of the environment is actually non-veridical, precisely because of those highlights.
I think that’s silly.
You are absolutely right! Although I think you meant to write Siberian Fox, not the Husky. In Novosibirsk, Beelyayev and Trut clearly showed the effect of pleiotrophy in breeding programs, most famously in foxes, but also in rats and mink. Select merely for tamer foxes, and you get raised tails and mottled fur, too. Amazing effects in just a few generations.
keiths,
And it assumes that natural selection is everything. I think that view is probably 50 years too old.
Right. Exactly what natural selection DOESN’T promise.
Natural selection has not made any promises to me — have you been misled again?
So if you try selecting for a bunch of traits simultaneously, you may find it rather easy (if pleiotropy is working in your favor) or you may find it rather difficult (if pleiotropy is working against you). Of course evolution doesn’t care: if the ‘bundle’ is adaptive, it will do well; if not, it won’t. Spandrels and such.
DNA_Jock,
Well, I agree, the guy did have a tendency to fib a lot.
keiths,
Furthermore, if we are to take Hoffman at his words, the reality that humans exist and breed is also not real, so why should natural selection be real. He is trapped by his own logic.
Hoffman says that nothing is real, including space-time, so how could natural selection be real.
Seems there are a lot of fruitcakes with a PhD after their names.
phoodoo,
So your complaint is that Natural Selection is acting too fast in the Siberian Fox experiment? You complain that NS is over-delivering?
Oooookay.
Or the environment changes. Been known to happen.
Bear in mind that the foxes were bred for their fur, and that selecting for tameness destroyed their monitory value. Traits are often linked. Dogs and cats bred for shape and color often have health problems.
In the wild, shaping traits takes longer, due to linkages. The population must survive the changes. This is why I say selection is multidimensional.
petrushka,
Agreed. And that aspect (pleiotropy) was the extent of my agreement with phoodoo. But then his complaining about “what NS promises” and the 160-year-old quote appears to be a dig at Darwin’s gradualism. Which, of course, the foxes showed in spades. The evolution was a lot quicker than expected, but still pretty continuous.
Whilst we are discussing pleiotropy, the words “linked” and “linkage” are potentially subject to misinterpretation, as they are terms of art for geneticists.
Different traits may be correlated [?] with one another, either due to pleiotropic effects at one locus OR due to physical linkage between different loci. Or both.
I agree that Hoffman’s standard of veridicality is far too high, because it requires that we bracket away anything involving actual behavior or interactions with the environment.
At the same time, I like the ways that Anil Seth and others using predictive processing use that model to argue for the non-veridicality of cognition — in a one sense of that much-vexed term.
Hoffman’s argument for the non-veridicality of cognition may be a bad argument but that doesn’t mean there aren’t good arguments as well.
KN,
It’s also worth noting that veridicality isn’t all-or-nothing, and that there are tradeoffs. Allowing non-veridicality in one area may more than compensate by enhancing veridicality in another, more crucial one.
Hmm. Maybe.
One of the things I like about Sellars on this issue is that he talks about “adequacy” with the explicit point that it’s intuitive how adequacy admits of degrees.
It’s somewhat more difficult to see how truth admits of degrees — to do that one would need to reject bivalence of truth values, though we do get a model of that in fuzzy logic.
But talking about adequacy of cognitive mapping allows of degrees while leaving bivalence about truth values intact — and we might very well need that to make sense of our ordinary discourse involving truth-talk.
I do appreciate the emphasis on competing trade-offs between cognitive mapping systems — neurocomputational resources are finite and a system may very well be excellent in mapping along one set of dimensions and not good at others. Bats are much better at mapping acoustic space than visual space, and humans are not well-known for their olfaction compared with other mammals.
Perhaps a cow morphed into a whale in one go too. Turned out it was a lot of beneficial mutations.
Anything is possible with evolution, right.
KN,
I would say that veridicality, like accuracy, does admit of degrees. After all, plenty of people talk about human perception being basically veridical without meaning that it’s perfect, flawless, free of illusions, etc.
That’s backed up by definition #2 below:
A great many philosophical issues regarding fitness would more simply be expressed as probabilities.
For a population of frogs catching flies, or of grape eaters contemplating purple things, the issue is not truth, but likelihood that an action will be rewarded.
From the perspective of the researcher, the issue is the probability of a stimulus triggering a behavior.
petrushka,
Hoffman is just batshit crazy. He will tell you the moon doesn’t exist when you don’t look at it, because, well natural selection.
So nothing exists like we think it does, space time doesn’t exist (he actually says this) and yet, natural selection, well of course that exists exactly like we think it does.
But in another universe he is not crazy.
That’s certainly true enough, and Hoffman might be one of those very confused individuals.
I also happen to find this idea, that things stop existing when we’re not looking at them, completely fatuous.
Many interpret the PP framework by saying the Bayesian models built by acting/sensing have a structure which somehow is similar to the aspect of the hidden causal structure of reality which is relevant to the organism’s survival and reproduction.
In the Psychonomic paper I linked upthread, Hoffman does test a structural mapping idea which he calls “critical realism”. But he limits the mapping from representation to reality to be “a homomorphism that preserves all structures”. So he is still limited to a yes/no situation — either the mapping preserves all structures or it does not.
In the PP case, the similarity of structure can be captured by a range of values, not merely yes/no. One way to quantify the similarity uses KL divergence. Another use a success semantics so that greater similarity is captured by saying more accurate representations mean more successful actions. Success is relative to the organism and involves both survival and reproducing in its niche.
But Hoffman in essence questions what we can conclude about the nature of reality based on our theory of perception. What can we conclude about reality from PP? It seems the best we can do is structural realism of some sort.
Kantian Naturalist,
I think this short essay on the 25th anniversary of McDowell’s Mind and World is on topic in that it also relates to perception and reality. In any event, you may enjoy it and you will certainly get more out of it than I did.
Schwenkler starts his essay with Kant; there was a brief exchange at Leiter on whether Hoffman is just Kant updated with evolution:
https://leiterreports.typepad.com/blog/2019/09/evolution-does-not-select-for-veridical-perception.html
(comments 6, 7, 8)
Rumraket,
Look at that, there are some things we can agree on.
If we had Mung back imagine how much unity this site could achieve. :}
That helps a bit, but I still have trouble grasping their point. My best guess so far is that Hoffman and colleagues claim that our perceptions of the world are in a way arbitrary; It could have been represented in a completely different way and it would still have served us equally well (in terms of fitness, as well as otherwise). At least that is how I interpret the “computer display” metaphor. I am still at a loss how to interpret a perceptual strategy that generates a veridical perception of the “true” structure of the world.
That relates to Objections 1 and 3 in the paper, but that’s all fine and dandy with me. What bugs me is that, with or without those biases, we always need some mapping from the structures in the world to our perceptions. So why the big fuss that what we observe are “icons” rather than the true structure of the world?
There are no structures in the world. The structures that we see come from us.
Yes, ants build anthills. But part of what makes the anthill a structure (rather than a bunch of grains of sand) is the way that our perceptual systems pick it out as a structure. And, for that matter, what makes something a grain of sand (rather than an interferer in light transmission) is the way that our perceptual systems pick it out as a grain of sand.
I’m not a fan of the “icons” idea. However the point ought to be that there is no true structure of the world. There is a true structure of how we perceive the world. But the “true” and “structure” parts come from us.
This is part of why there is a hard problem of consciousness. The “hard problem” people want an explanation of how perception gives us the true structure that is already there. But the true structure is not already there. So the “hard problem” is a bogus problem. We need to abandon the idea that perception is passive, and instead we need to understand the creativity of perception.
I think you have to start by assuming perceptions produce mental representations* (see note at end). Given that, by definition a “veridical” perception produces a “true” representation.
But now we get into what exactly ‘true’ can mean in this case. The usual starting point is that a true representation corresponds somehow to an actual state of affairs in the world. That’s the sort of perception that Hoffman is attacking as I understand him. He says that his simulations show that fitness does not lead to perceptions that produce true representations of states of affairs. Hence we have no reason to draw conclusions on the nature of reality from our perceptions.
He then goes on to develop his own theory of the nature of reality based somehow on consciousness, but those claims are something completely different (the allusion to Monty Python is intentional).
———————————-
* I have not got into the nature of mental representation. Old-fashioned views from philosophy and AI assume it is a series of beliefs in some kind of internal language of thinking. Modern approaches, like the Predictive Processing theory of perception, and the Deep Learning approach in AI, do not use not language-like approaches to representation.
Didn’t reference Hoffman. Haven’t read Hoffman. I’m just stating my own understanding of things. Don’t claim it to be true or important. But I appreciate responses, if they show some sign of being related to what I said.
My undergraduate major included working as a lab assistant to a guy doing experiments on color vision. My knowledge of the subject is rather shallow, but my interest is high. I find it interesting that so many people expound on perception without having designed, or tried to design, a well controlled experiment on the subject.
Just my opinion, but the task of trying to do something leads to a different kind of understanding from reading about it.
I agree with that. We cannot start by assuming there is an objective structure of reality that we perceive. In other words, there is no god’s-eye viewpoint that perception let’s us take.
Where I see us as parting ways in the past on scientific realism, and where I guess we part ways here on everyday realism, is on what we can conclude from the success of our scientific theories involving unobservable, theoretical entities and similarly from the success of our action/sensing for everyday entities.
I think there is an IBE argument that both latch onto some aspect of reality. That aspect could be set of objects and their relations, or only structure, or set of processes. On even days of the week, I go with structural realism. On odd days, I work out and try not to think about it.
On the hard problem: I think it is about phenomenal experience and is different from the issues of perception that Hoffman is making claims about in his simulations. I agree that his further work claims on consciousness covers the hard problem. I don’t see that the two need be related, but maybe I need to read him again. Unlikely, that, however.
Oh yeah. My undergrad included a year of “experimental psychology” with a strong practical emphasis on perception. As a consequence, when Hoffman asks “Does the shape and color that I experience match the true shape and color of the real pear?” my first reaction is “Whoa there! Those are two completely different things.” But then I calm down a bit, and acknowledge that the second best example of adaptive visual processing is probably size constancy…
Following that logic, there are no ants, or grains of sand, or a world, nor any “us”. Yet, in using those words and concepts, we are labeling something outside of us, and that something must be “truly” there and “truly” consist of ordered and interacting parts, i. e. be structured.
That seems overthought. Sure, what we actually perceive is photons hitting our retina, vibrations in the air, pressure changes on our skin, etc, and everything else is just us making sense of all that. If Hoffman et al. wish to call those internal concepts “icons”, I am fine with that. What I find hard to understand is why you say that means the outside objects and events those perceptions relate to don’t have “true” structure apart from what we attribute to it. Ants do not spontaneously desintegrate into elementary particles, so there must be true interactions keeping the ant together. It is not our minds doing that.
Sorry, I fail to see how that helps solving the hard problem of consciousness.
I have not much hope of anyone solving the hard problem of consciousness in my lifetime. I think there’s a difference between solving it and finding a useful approach.
It amuses me that we are trying to solve the problem by mimicking or replicating what brains do. Is that the only kind of consciousness possible?
The interpretation of what ‘true’ means here is the hard part all right. In the end a mental representation is just a mental representation. For example, people with synesthesia experience stimulations in one sensory pathway as experiences in another, e.g. some perceive sound as visual images. Suppose that the mapping of amplitude, tone and pitch of an airwave to visual sensations was isomorphic, would that constitute a true representation?
The entire part about natural selection creating biases in our sensory perceptions is of course quite valid, but it’s not that original and it doesn’t seem to be the focus of most of the discussion here.
In preparing classes, I once tried to find an example of phenotypic variation that went beyond the hoary blue-brown eye colour polymorphism. So I came across a website on optical illusions here, where the author stated that some subjects were incapable of perceiving certain illusions:
I wrote an e-mail and got a nice reply that the frequency was probably too low to reliably detect in a group of students. A real shame, that would have been so cool.
Thanks for that! I did go down the McDowell rabbit hole once upon a time (and published three papers on him) but eventually concluded that there’s a lot less there that meets the eye.
Jerry Fodor has a rather infamous review of Mind and World that ends as follows:
Ever since Descartes, a lot of the very best philosophers have thought of science as an invading army from whose depredations safe havens have somehow to be constructed. Philosophy patrols the borders, keeping the sciences ‘intellectually respectable’ by keeping them ‘within … proper bounds’. But you have to look outside these bounds if what you care about is the life of the spirit or the life of the mind. McDowell’s is as good a contemporary representative of this kind of philosophical sensibility as you could hope to find. But it’s all wrong-headed. Science isn’t an enemy, it’s just us. And our problem isn’t to make a place in the world for the mind. The mind is already in the world; our problem is to understand it.
I agree about ants, grains of sand and us. But I disagree about “world”.
There is a reality apart from us. And we can talk about that reality as a whole (i.e. the world). But in order to talk about parts of the world (such as ants or grains of sand), we first have to divide it up into parts. And that dividing up comes from us.
The whole can be said to be truly there. But there are no parts until we divide that whole into parts. And there is no structure until we provide the structure by means of our dividing.
If what we actually perceive is photons, then why didn’t Aristotle write about photons?
It ought to be obvious that we do not perceive photons at all, except perhaps with the aid of sophisticated apparatus. You are stretching “perceive” way beyond where it will go.
According to current physical theories, photons mediate our perception. But that is not the same as perceiving photons. And photons only exist in our physical theories because of the way that we have structured the world.
Well, okay. So you are a theist, and your god tells you about structure and truth. And you are a dualist, so that your spiritual soul has a channel to your god to find out what is this structure.
I am neither a theist nor a dualist — and, as far as I know, you are neither a theist nor a dualist. So when I look for the source of structure and truth, I see that it seems to be coming from us.
You have completely changed the subject. I have not said anything about disintegration.
I agree that we consider ants to be objects. But the integrity of the ant as a whole comes from us.
That we consider the ant to be an object comes from our natural pragmatism. There are no rules of logic based on external truths that would tell you that an ant is an object. But we make better sense of the world by taking it to be an object.
Perception is not veridical; it is pragmatic.
That’s not what I have been doing. I have been looking at what abilities an organism needs in order to survive. I have occasionally called it (privately) “the big toe theory of intelligence” on the principle that intelligence might just as well come from the big toe. It doesn’t matter where it comes from. What matters is what it does.
At least we start with some agreement.
This is where I see you as inconsistent.
On one occasion you will criticize pragmatic theories of truth. And on another occasion, you infer truth from pragmatics.
The success of our action in sensing everyday entities is pragmatism at work. It has nothing to do with truth and logic.
I try to make a very careful distinction between decisions based on pragmatism and decisions based on truth and logic. And that’s why we often disagree.
There’s a chair on my patio. I know that it is an object, rather than something obtruding from the ground. And I know that because I can pick it up and move it around. And I know that it is chair because, among other things, I can sit on it.
But ants and birds can also be there on the patio. The ants and birds are unable to sit on the chair. So they cannot know that it is a chair. And they cannot pick it up and move it around, so perhaps they have no reason to consider it to be an object (rather than something obtruding from the ground). Pragmatics derives from our biology. Different organisms will have different pragmatics. If there is an external truth (external to humans), then it should be the same for all organisms.
In ordinary life, it might not matter whether we make these fine distinctions (such as between pragmatism and truth). But if we are trying to understand consciousness, then it becomes essential that we make such distinctions.
IBE (inference to the best explanation) is unavoidably tied to pragmatics rather than just truth and logic.
Then why should we expect our perceptions of natural selection to be accurate?
Accuracy is a measure. We expect to improve on such accuracy as science advances.
I have a hard time thinking there is a hard problem of consciousness. How human brains work, on the other hand, is a problem humans cannot solve
You first have to decide what consciousness is and whether things either have it or not. I don’t think it is binary. Human awareness evolved.
Exactly!
I agree with the first half of this statement! 😉
I’ll pinch that!
Neil,
Your position is inconsistent.
In this thread you write:
But earlier (in April) you acknowledged that that there was lumpiness to reality:
That lumpiness, and the different characteristics of different “lumps”, constitutes structure in reality.
I replied at the time:
When have I ever done that? Definitely not on purpose, at least not lately. In fact, insofar as I have a well thought out theory of truth, it is Misak’s version of Peiice’s pragmatism about truth. (With some help from Blackburn’s latest book on truth).
I understand your posts deriding certain viewpoints in AI or in philosophy of mental representations as assuming that representations are based on language, ie propositions, and that truth is correspondence between the propositions and some god’s-evey viewpoint of the facts. I agree such positions are wrong. They are also out of touch with modern approaches.
ETA: to the extent that I understand your ideas on truth, they involve what is accepted by a language community. If that is right, then perhaps that is close to Rorty’s version of pragmatism, which I do not agree with (but I have only a very limited understanding of his views based mainly on writings of his detractors, so caveat reader.)
ETA 2: I should emphasize that truth is not part of the PP theory of perception. That’s KN’s point as I understand him which I elaborated on in my reply to him.
My previous posts on truth have been involved communities of inquirers engaging in eg scientific practice.
Neil:
That isn’t what the hard problem is about.
Lumpiness does not equate to parts, unless we seperate out a lump and take it to be a part.
The lumpiness is still structure, Neil. The “dough” isn’t an undifferentiated, featureless mass.
Corneel,
It’s important because Hoffman considers those biases to be examples of non-veridical perception. I think they aren’t, because they aren’t telling us anything false. To emphasize something as worth paying attention to is not to mislead.