In an earlier comment, I indicated that I would post something on my blog to help clarify the distinction between direct and representationalist perception.
I have now done that, in a series of four posts. The first of those is
and it, in turn, contains links to the other posts.
Your turn.
“In other words, the requirements become that of consistency of the calibration over the various instruments that we might use for measuring temperature. This is more of a pragmatic requirement than a truth requirement.”
Then you know what calibration means, i.e. you need to know that there is a relation, between your two signals temperature and length of the mercury column. You then need to know what a relation means, and what your signals mean.
“The direct perception viewpoint is that, via a process of perceptual learning, our brains build cat transducers and tune them to do an adequately good job of picking out cats. As our brains build these transducers we are, in effect, gaining implicit knowledge on what distinguishes cats from other things. I take it that this implicit knowledge becomes part of our system of meanings. The learning needed to build these transducers is, I believe, part of what explains intentionality.”
There you have a lot actions that means you already have intentionality or knowledge. Why you do not start with a sensorial signal like a neurone activated by a photon and only with neurons you try to reach the concept of light?
You don’t even need a concept of length for this to work. You need only know that the column changes. You put calibration marks against the column for specific temperature. Whether you know the meaning of length is of no relevance.
What changes in the column? The color? the weight? the wide?
How much it change? Each mark is representative of how much?
You need the concept of lenght and measure.
Try to explain how do you build concept from a signal only with neurons.
No, you don’t.
Let me compare to the internet. A packet gets from here to where you see it, by a series of hops. The general principle on which the internet is built, is that the details of all of those hops do not matter. The only thing that matters is the end-to-end result, namely that the packet gets to the destination one way or the other.
It’s the same with my example of measuring temperature. What matters is the end-to-end result. That is, what matters is that a particular real world property is reflected in the reading of the thermometer. How that is achieved does not matter.
The length is important to understand the physics of the entire procedure. But it plays no conceptual role. Only the end-to-end result is involved in the conceptual role. We can use the temperature readings without a concept of length.
That’s a question about neural implementation, and outside the intended scope of this thread.
Thanks for these posts, Neil.
I’ll have more to say in coming days, but I wanted to post a link to an online book I found useful in understanding what is claimed by “direct perceptionists”.
Direct Perception, by Claire F. Michaels and Claudia Corello.
Well that is the point. Direct perception doesn´t exist. To have the perception of something you have to process the signal. With out processing the signal the only you can have is answers/reactions to the signal. Not perceptions.
Without processing the signal the only thing you can have is a pavlovian dog.
Do you understand that the dog “perceived” the bell? If yes we are calling perception in different ways, and probably it is not the way Plantinga understand it.
I should have mentioned that I do not claim to speak for all direct perceptionists.
I did not read about direct perception, and decide “that’s great; it explains everything.” Rather, I was trying to understand perception, and someone pointed out that my views were close to those of direct perception.
When I read what is written by direct perceptionists, some of them seem far too doctrinaire to me. My interest is in what works, not in adhering to some official doctrine.
My first encounter with direct perception was before I had much of a worked out idea. And the book that mentioned direct perception was an epistemology book by a Randroid philosopher (or “objectivist philosopher”). He was clearly using “direct perception” as a doctrine that allowed him to evade all of the hard questions.
This criticism is very common, and based on a misunderstanding of what direct perception is (or should be) claiming.
It seems your posts are about three separate levels of explanation:
1. A neuroscience level: for example, your references to Hebbian learning in the last post of your series.
2. A psychological level: I think the term “direct perception” is used mainly in psychology; for example, based on a quick skim, that seems to be the main level of explanation in the reference Keith links to on direct perception. It is also the subject of experiments like those on mental representation and image rotation
3. A philosophical level: at its core, Plantinga’s EAAN is a philosophical argument based on the mind/body problem as studied by philosophers, specifically mental content and mental causation and how they relate to physical behavior.
It is possible to be a representationalist at each of these levels without being one at the other. So I think it is important to identify what level one is addressing in any specific criticism of representationalism.
I’ll say a bit more about 1 and 3 in separate posts.
But I do have a question about how you saw the markings getting on the thermometer in your third post. Also on how the analogy relates to what the brain or mind is doing.
Based on your reference to using a signal generator to calibrate other types of measurement, my take would be that you are saying that the brain/organism calibrates its perceptions by successful interaction with the environment. So that environment is in some sense that reference signal generator. Is that what you meant? If so, that would make sense to me.
To clarity this point: I don’t think Plantinga has any issue with the claim that evolution can affect brains and behavior. He would accept that physical brains/organims which are successful in an evolutionary sense can correctly categorize the world and act accordingly.
His argument is that there is nothing in this that allows naturalists to justify the reliability of beliefs, which are part of the mind, not the brain. For he believes that
(a) a naturalist must claim that beliefs and other mental contents must supervene on the physical brain and
(b) that there is no generally accepted explanation of how they do so and of how the causation of physical behavior by mental contents can work.
Since it is physical behavior that interacts with evolution by NS, such an explanation is needed to justify the evolution of reliable belief generators solely by naturalistic evolution.
For example, in this comment on the issue he says:
This point has come up before, eg in Dr Liddle’s posts about the use of EAAN at UD. She and others on that thread seemed to take it as obvious and a given that beliefs cause behavior. I think that is true, and so does Plantinga, but his argument is that it is not a given of naturalistic evolution.
(BTW, I think is closely related to the point Blas is trying to make).
Probably true. The division into academic disciplines often doesn’t fit very well.
Not quite.
Yes, the organism calibrates on the basis of interaction with the environment. But, from time to time, you want to check the consistency of calibration when the environment is not doing anything important. And that’s what I was thinking of in terms of a test signal and dreaming.
Yes, that might be right. I don’t find it easy to follow Blas.
Of course, Blas is correct, that processing is required. The point of my discussion was that the thermometer already does most of the processing. It isn’t something that we have to do after reading the thermometer.
On Gibson’s view: he was a proponent of perceptual learning, and his wife wrote a book on that. For Gibson, perceptual learning would include the construction of neural devices (his tranducers), which could be said to do the processing.
We’ve been having a lively discussion of direct perception on the Thus I refute Johnson thread. It should probably move here, now that this thread is available.
My question for Neil or any other defenders of direct perception: If perception is direct and inference is absent, how do you explain these motion illusions? There is no actual motion, so any perceived motion is inferred motion.
To clarify this point, Connectionist models can be said to include representations of the categories they recognize. Such models have been used in neuroscience, in particular for perception and categorization I believe, although it is true they are not thought to be completely biologically accurate. But connectionist models may capture enough biological features to be a scientifically useful model. Only scientific research can address whether they do or not and whether better models might exist.
Under connectionist models, categorization in perception would involve, among other things, using the representations encoded in neural configurations to categorize based on past experience. Then feedback from action resulting from that categorization results in update to those neural representations.
Do these models imply computation by the brain? That depends on one’s definition of physical computation. If one restricts computation to the symbol manipulations used, eg, by non-connectionist AI, then connectionist models are not computational. But as I posted elsewhere at TSZ and as discussed here, there is a better definition of physical computation; further one can argue that both connectionist models and the physical brain conform to that definition.
So the argument that the brain might be shown by scientific research to be computational and representational, at a neuroscience level, would involve using experiments and analysis to show:
1. the connectionist model or something along that lines provides the best explanation ( which would include predictive power), and
2. the model is implementing something which can be called a representation,
3. the model conforms to a physical definition of computation in how it captures processes in the brain, and
4. the physical brain processes which are being modeled also conform to that definition of computation, and in particular, in the way the definition is used by the model.
Point 4 is meant to address an argument that modelling that brain as computation is not the same as showing the physical brain is computing.
That seems to be a mistake. Proponents of direct perception do not claim that perception is perfect. There’s no reason that a motion detector (transducer) could not be triggered even when there is no motion.
That said, there is also a question of where perception ends and inference begins. And I don’t think that is completely clear.
Neil,
I’m not trying to show that perception is imperfect. That’s obvious.
What’s interesting about the motion illusions is that the imperfection is systematic, and it reveals something about how motion perception works in veridical cases.
Specifically, the illusions show that in some cases motion is inferred, not directly perceived.
Do you dispute that?
As I see it, we overuse the term “computation.” I’ve come across people who insist that Venus is computing its orbit as it moves along that orbit. Presumably, a few hundred years ago, Venus was computing cycles and epicycles, but due to a more recent program update it is now computing ellipses.
I don’t find that way of using “computation” at all useful. So I prefer to stick to the idea of computation as operations on abstract symbols. From that perspective, a computer emulation of a neural network is computation (computational simulation), but the simulated network is not computing.
In a way, the arguments about computation are a side issue. The trouble with computationalism is that it emphasizes internal operations, while I see more importance in how we interact with the world. The internal operations are mostly an implementation detail. It’s the interaction that is mostly important.
I’m not persuaded that they show inference.
To take a different example, the cart wheels in old Westerns — I perceive them turning backwards, and I infer that they are really turning forward.
I don’t find it useful either, but for a different reason. The orbit of Venus isn’t really about anything other than the orbit of Venus. A physical computation is about something other than itself.
Then you’re back to arguing that computers don’t compute. You never did respond to my questions about that from the other thread:
I’m afraid I’m not getting the relevance of the motion illusion to the question of the directness of perception. The general take on this is that there must always be some epistemically relevant non-physical intermediary that must be perceived for a physical object to be perceived in order for directness to fail. What does that mean?
An entity (or process) E is relevantly common both to some perceiver S’s actual perception of some (intentional) object O and to his merely ostensible perceptual experience as of O = df. E is such that both (i) it is only in virtue of S’s perceiving or otherwise apprehending E that S actually perceives O; and (ii) S’s perceiving or otherwise apprehending E is sufficient for S to be in a condition that is merely to him as if he were perceiving O.
What epistemically relevant intermediary do you think must be apprehended by S (not TAKE PLACE in or around S) in order for the the motion illusion to work for S?
Neil,
They’re both inferences.
The TV presents you with a rapidfire series of static images. In each successive image, the spokes are displaced from where they were before. Your brain infers a motion to explain the displacement, and the inference is that the wheels are turning backwards. Just as in the motion illusions I cited, there is no real motion, just a series of static images. The motion is inferred, and it is inferred preconsciously.
You are then able to consciously infer that the wheels must really be turning forward, due to your knowledge of how physics works in the real world.
walto,
Consider the second illusion on this page.
Two static images are being shown repeatedly to your visual system. There are at least three possible interpretations of those image sequences:
1. The balls poof out of existence, and then new balls poof into existence in different positions.
2. Each ball is moving right, then left, then right…
3. Each ball is moving up, then down, then up…
Each of those three interpretations is compatible with the stimulus. The stimulus is ambiguous.
If you’re like me, your perception will “flip” between (2) and (3), but excludes (1) entirely. Why? I think it’s because in the real world, objects typically don’t poof in and out of existence. Perceiving them as moving, rather than poofing, is more likely to be correct, so our visual systems perceive them that way. It’s an inference, and the inference is biased in a way that tends to track reality.
The motion isn’t directly perceived, it’s inferred — and the inference is biased in a way that tends to match the behavior of real-world objects.
They are silly questions.
No, that’s not according to me. That’s according to your misunderstanding of the point.
You are imputing an inference to the brain, but not for any reason that I can discern.
Neil,
Not so silly, evidently. They expose a fundamental problem with your “computers don’t compute” claim, and you seem to be aware of that. When you’re backed into a corner, it’s much easier to dismiss questions as “silly” than to actually answer them, isn’t it?
Then what is your interpretation of the scenario I described? Who is doing the computation, and when?
Neil,
Because there is no actual motion, just static images. Whence the perceived motion if it is not being inferred from the static images?
Okay. So there is no reason at all. You are just making it up because it suits your ideology.
Neil,
You didn’t answer the question:
My reason for thinking that the motion is inferred is because I can’t see where else it could be coming from, given that there is no motion in the images themselves. What is your answer to my question?
You also didn’t answer these questions:
Keith
There’s no rule that anyone must answer questions.
Alan,
Of course there isn’t. There’s also no rule stating that I can’t ask them or point out when they go unanswered. All of us, including you, have pointed out that William J Murray fails to answer certain questions.
Neil is free to ignore my questions, and if he does, readers will have to weigh at least two competing hypotheses: 1) Neil has excellent answers to my questions, but has for some reason chosen not to state them; or 2) Neil can’t answer my questions without conceding the point being disputed, and so has chosen not to answer.
Given that my questions are obviously relevant, I suspect that readers will see #2 as the likelier explanation. They will make that inference.
Or Neil can choose to defend his claims. It’s up to him.
Sure!
Alan,
Really? After asking a question just once in the midst of a busy thread and getting no reply, you’d assume that #2 was the likelier explanation?
You might want to rethink that.
OK, rethought. You overlooked it. Sorry for not pressing the point.
ETA we’re all fallible, Keith.
I’m happy to answer if you’re still curious.
keiths,
I’ll look out for it.
Alan,
I agree. Is anyone disputing that?
keiths,
The Pope?
Heh. Even the Pope admits he’s fallible — just not when speaking ex cathedra.
It would be much handier to have a Pope who’s infallible all of the time, but I guess God can’t be bothered.
Apologies for diversion. Back on topic:
I like this. As a kid, I can remember seeing the illusion of the buggy wheels turning backwards and having no problem realising it was not real; my general knowledge of how the world turns (forwards mostly 🙂 ) informed me that this was some sort of artefact generated by the filming process to which one could quickly adapt. (A more extreme example would be the upside-down goggles that have been used in experiments to which subjects apparently quickly become accustomed).
Another of Neil’s metaphors about scaling a mountain using different techniques and handholds resonates. Not quite the Tower of Babel but…
Or (3) – the questions were intended as ridicule, rather than as a serious attempt to understand the issues.
Neil,
Not only are my questions serious, they also get to the heart of the issues being discussed.
I asked:
This is a crucial question. We perceive motion, yet there is no motion in the actual stimulus. Where is the perceived motion coming from if it is not being inferred? As a direct perceptionist, how do you explain the perceived motion without invoking an inferential process? A ‘motion transducer’ doesn’t solve the problem because the ‘transducer’ itself is inferring motion where there is none.
Regarding my computer scenario, I asked:
You’ve stated that computation is the manipulation of abstract symbols, and that computers therefore cannot compute. If you are right, then the computer in my scenario is not computing during the eight hours that I’m asleep. If so, then who is doing the computing, and when, exactly?
These are legitimate questions that show why it doesn’t make sense to insist, as you do, that computers can’t compute.
But I answered that long ago.
A transducer can be a simple crude neural circuit. It isn’t expected to be a complete perceptual system. If I walk out of my garage while the garage door is closing, the door stops closing. But there isn’t a person detector. Does it follow that the door opener is making an inference? Or is the door opener just a crude detector (a light cell detecting an interrupted beam)?
The transducers do not have to be accurate. They only need to be statistically correct on most occasions.
Neil,
The inferences we’re talking about are far more complicated than the “inferences” of your garage door sensor.
Look at the last illusion on this page. Not only do we infer motion, we also infer that one of the balls is hidden behind the green square!
Neither the motion nor the hiddenness are explicit in the stimulus. They must be inferred.
I can’t think of a clearer demonstration that the theory of direct perception is wrong. No wonder it is so widely rejected.
I see only a static image that disappears after an instant. Presumably, the video doesn’t like linux.
So can I ask you a question?
How does your post relate to the question walto asked at the end of his post?
I am not trying to be argumentative, I am just trying to understand the exchange between you two. In particular, the dense, precise prose of walto’s post is natural for trained philosophers but is hard to fully engage with for me. But I understand what you are saying, just not how it relates to what he asked.
For example, are you saying the the subconsicous brain processes are the relevant intermediary. If so, how would you say they being apprehended, as opposed to taking place within.
Our visual perception system developed in an evolutionary manner and has been designed by the environment in which we evolved. This did not usually contain optical illusions (other than camouflaged prey and predators) such as computer-generated images, holograms etc. The reason given for moths spiralling into candle flames is that their sensory system is interpreting the light as a far fixed (parallax-free) point and is trying to navigate a straight line. The candle is outside its evolutionary experience.
Neil,
There are two static images in the video. In the first image (the one you can presumably see), there is a constellation of red balls including a central ball just to the left of the green square. In the second image, the green square remains in place and the entire constellation of red balls has been shifted to the right, except that the central red ball disappears.
The visual system 1) infers motion from the relative displacement of the two almost identical constellations, 2) infers that the green square is in front of the constellation, and 3) infers that in the second image, the central red ball is still there, but it is behind the green square.
Yet in the images themselves there is no motion, no in-front-of or behindness, and no central red ball in the second image. They are all inferred, and the inferences are not trivial.
keiths,
I still see no basis for concluding that there is inference in the visual system.
I’m wondering what you even mean by “inference.” To me, an inference involves weighing possibilities and making a judgment. There isn’t time for the visual system to do that. It might be reasonable to say that you are consciously making an inference based on the information that the visual system provides. Since I am not seeing anything other than the initial image (which promptly goes blank), I don’t have an opinion on that.
BruceS,
My reply to walto was in response to his first sentence:
My answer was an attempt to convey the relevance to him.
The rest of his comment didn’t make any sense to me.
First he stipulated that the epistemic intermediary must be non-physical:
Then he asked:
Put those together, and in effect he is asking: “What non-physical intermediary outside of S must S apprehend in order to experience the motion illusion?”
Surely he doesn’t mean to imply that all critics of direct perception are dualists who believe we must apprehend ghostly non-physical intermediaries floating outside of us, but I don’t see another interpretation of what he wrote. I hope he’ll clarify.
Even neglecting the “non-physical” part, why would the intermediaries necessarily be outside of us? As walto put it, they must be “…apprehended by S (not TAKE PLACE in or around S)…”.
Where else are “epistemically relevant intermediaries” going to take place if not inside of us?
keiths:
Alan:
Agreed, but with scare-quotes around “designed”.
Agreed.
I agree with all of that. Were you hoping I wouldn’t? 🙂
I would only add that what is most interesting about illusions is what they tell us about how perception works under normal conditions.
It is easy to see why the same inferences that lead to the motion illusions we’ve been analyzing are useful inferences under normal conditions. I can elaborate if the reasons aren’t clear to everyone.