Dissolving the Question

Followup to:  How an Algorithm Feels From the Inside, Feel the Meaning, Replace the Symbol with the Substance

"If a tree falls in the forest, but no one hears it, does it make a sound?"

I didn't answer that question.  I didn't pick a position, "Yes!" or "No!", and defend it.  Instead I went off and deconstructed the human algorithm for processing words, even going so far as to sketch an illustration of a neural network.  At the end, I hope, there was no question left - not even the feeling of a question.

Many philosophers - particularly amateur philosophers, and ancient philosophers - share a dangerous instinct:  If you give them a question, they try to answer it.

Like, say, "Do we have free will?"

The dangerous instinct of philosophy is to marshal the arguments in favor, and marshal the arguments against, and weigh them up, and publish them in a prestigious journal of philosophy, and so finally conclude:  "Yes, we must have free will," or "No, we cannot possibly have free will."

Some philosophers are wise enough to recall the warning that most philosophical disputes are really disputes over the meaning of a word, or confusions generated by using different meanings for the same word in different places.  So they try to define very precisely what they mean by "free will", and then ask again, "Do we have free will?  Yes or no?"

A philosopher wiser yet, may suspect that the confusion about "free will" shows the notion itself is flawed.  So they pursue the Traditional Rationalist course:  They argue that "free will" is inherently self-contradictory, or meaningless because it has no testable consequences.  And then they publish these devastating observations in a prestigious philosophy journal.

But proving that you are confused may not make you feel any less confused.  Proving that a question is meaningless may not help you any more than answering it.

The philosopher's instinct is to find the most defensible position, publish it, and move on.  But the "naive" view, the instinctive view, is a fact about human psychology.  You can prove that free will is impossible until the Sun goes cold, but this leaves an unexplained fact of cognitive science:  If free will doesn't exist, what goes on inside the head of a human being who thinks it does?  This is not a rhetorical question!

It is a fact about human psychology that people think they have free will.  Finding a more defensible philosophical position doesn't change, or explain, that psychological fact.  Philosophy may lead you to reject the concept, but rejecting a concept is not the same as understanding the cognitive algorithms behind it.

You could look at the Standard Dispute over "If a tree falls in the forest, and no one hears it, does it make a sound?", and you could do the Traditional Rationalist thing:  Observe that the two don't disagree on any point of anticipated experience, and triumphantly declare the argument pointless.  That happens to be correct in this particular case; but, as a question of cognitive science, why did the arguers make that mistake in the first place?

The key idea of the heuristics and biases program is that the mistakes we make, often reveal far more about our underlying cognitive algorithms than our correct answers.  So (I asked myself, once upon a time) what kind of mind design corresponds to the mistake of arguing about trees falling in deserted forests?

The cognitive algorithms we use, are the way the world feels.  And these cognitive algorithms may not have a one-to-one correspondence with reality - not even macroscopic reality, to say nothing of the true quarks.  There can be things in the mind that cut skew to the world.

For example, there can be a dangling unit in the center of a neural network, which does not correspond to any real thing, or any real property of any real thing, existent anywhere in the real world.  This dangling unit is often useful as a shortcut in computation, which is why we have them.  (Metaphorically speaking.  Human neurobiology is surely far more complex.)

This dangling unit feels like an unresolved question, even after every answerable query is answered.  No matter how much anyone proves to you that no difference of anticipated experience depends on the question, you're left wondering:  "But does the falling tree really make a sound, or not?"

But once you understand in detail how your brain generates the feeling of the question - once you realize that your feeling of an unanswered question, corresponds to an illusory central unit wanting to know whether it should fire, even after all the edge units are clamped at known values - or better yet, you understand the technical workings of Naive Bayes - then you're done.  Then there's no lingering feeling of confusion, no vague sense of dissatisfaction.

If there is any lingering feeling of a remaining unanswered question, or of having been fast-talked into something, then this is a sign that you have not dissolved the question.  A vague dissatisfaction should be as much warning as a shout.  Really dissolving the question doesn't leave anything behind.

A triumphant thundering refutation of free will, an absolutely unarguable proof that free will cannot exist, feels very satisfying - a grand cheer for the home team.    And so you may not notice that - as a point of cognitive science - you do not have a full and satisfactory descriptive explanation of how each intuitive sensation arises, point by point.

You may not even want to admit your ignorance, of this point of cognitive science, because that would feel like a score against Your Team.  In the midst of smashing all foolish beliefs of free will, it would seem like a concession to the opposing side to concede that you've left anything unexplained.

And so, perhaps, you'll come up with a just-so evolutionary-psychological argument that hunter-gatherers who believed in free will, were more likely to take a positive outlook on life, and so outreproduce other hunter-gatherers - to give one example of a completely bogus explanation.  If you say this, you are arguing that the brain generates an illusion of free will - but you are not explaining how.  You are trying to dismiss the opposition by deconstructing its motives - but in the story you tell, the illusion of free will is a brute fact.  You have not taken the illusion apart to see the wheels and gears.

Imagine that in the Standard Dispute about a tree falling in a deserted forest, you first prove that no difference of anticipation exists, and then go on to hypothesize, "But perhaps people who said that arguments were meaningless were viewed as having conceded, and so lost social status, so now we have an instinct to argue about the meanings of words."  That's arguing that or explaining why a confusion exists.  Now look at the neural network structure in Feel the Meaning.  That's explaining how, disassembling the confusion into smaller pieces which are not themselves confusing.  See the difference?

Coming up with good hypotheses about cognitive algorithms (or even hypotheses that hold together for half a second) is a good deal harder than just refuting a philosophical confusion.  Indeed, it is an entirely different art.  Bear this in mind, and you should feel less embarrassed to say, "I know that what you say can't possibly be true, and I can prove it.  But I cannot write out a flowchart which shows how your brain makes the mistake, so I'm not done yet, and will continue investigating."

I say all this, because it sometimes seems to me that at least 20% of the real-world effectiveness of a skilled rationalist comes from not stopping too early.  If you keep asking questions, you'll get to your destination eventually.  If you decide too early that you've found an answer, you won't.

The challenge, above all, is to notice when you are confused - even if it just feels like a little tiny bit of confusion - and even if there's someone standing across from you, insisting that humans have free will, and smirking at you, and the fact that you don't know exactly how the cognitive algorithms work, has nothing to do with the searing folly of their position...

But when you can lay out the cognitive algorithm in sufficient detail that you can walk through the thought process, step by step, and describe how each intuitive perception arises - decompose the confusion into smaller pieces not themselves confusing - then you're done.

So be warned that you may believe you're done, when all you have is a mere triumphant refutation of a mistake.

But when you're really done, you'll know you're done.   Dissolving the question is an unmistakable feeling - once you experience it, and, having experienced it, resolve not to be fooled again.  Those who dream do not know they dream, but when you wake you know you are awake.

Which is to say:  When you're done, you'll know you're done, but unfortunately the reverse implication does not hold.

So here's your homework problem:  What kind of cognitive algorithm, as felt from the inside, would generate the observed debate about "free will"?

Your assignment is not to argue about whether people have free will, or not.

Your assignment is not to argue that free will is compatible with determinism, or not.

Your assignment is not to argue that the question is ill-posed, or that the concept is self-contradictory, or that it has no testable consequences.

You are not asked to invent an evolutionary explanation of how people who believed in free will would have reproduced; nor an account of how the concept of free will seems suspiciously congruent with bias X.  Such are mere attempts to explain why people believe in "free will", not explain how.

Your homework assignment is to write a stack trace of the internal algorithms of the human mind as they produce the intuitions that power the whole damn philosophical argument.

The correct answer will not be given tomorrow.  This is one of the first real challenges I tried as an aspiring rationalist, once upon a time.  One of the easier conundrums, relatively speaking.  May it serve you likewise.

Comments

[Tom McCabe]

I have no idea why or how someone first thought up this question. People ask each other silly questions all the time, and I don't think very much effort has gone into discovering how people invent them.

However, note that most of the silly questions people ask have either quietly gone away, or have been printed in children's books to quiet their curiosity. This type of question- along with many additional errors in rationality- seems to attract people. It gets asked over and over again, from generation unto generation, without any obvious, conclusive results.

The answer to most questions is either obvious, or obviously discoverable- some easy examples are "Does 2 + 2 = 4?", or "Is there a tiger behind the bush?". This question, however, creates a category error in the human linguistic system, by forcibly prying apart the concepts of "sound" and "mental experience of sound". Few people will independently discover that a miscategorization error has occurred; at first, it just seems confusing. And so people start coming up with incorrect explanations, they confuse a debate about the definition of the word "sound" with a debate about some external fact (most questions are about external facts, so this occurs by default), they start dividing into "yes" and "no" tribes, etc.

At this point, the viral meme-spreading process begins. An ordinary question ("Is the sky green?") makes reference to concepts we are already familiar with, and interrelates them using standard methodology. A nonsensical question either makes reference to nonexistent concepts ("Are rynithers a type of plawistre?"), or uses existing concepts in ways that are obviously incorrect ("Is up circular?"). Our mind can deal with these kinds of questions fairly effectively. However, notice the form of a question asked by the tribal chief/teacher/professor/boss: things like "Does electromagnetism affect objects with no net charge?". Even at large inferential distances, the audience will probably pick up on some of the concepts. Most laymen have heard of "electromagnetism" before, and they have a vague idea of what a "charge" is. But they lack the underlying complexity- the stuff beneath the token "electromagnetism"- needed to give a correct answer.

From the inside, this sounds pretty much like the makes-a-sound question: familiar concepts ("tree", "falling", "sound") are mixed together in ways which aren't obviously nonsense, but don't have a clearly defined answer. The brain assumes that it must lack the necessary "underlying knowledge" to get past the confusion, and goes on a quest to discover the nonexistent "knowledge". At the same time, the question conveys an impression of intelligence, and so the new convert tells it to all of his friends and co-workers in an attempt to sound smarter. Many moons ago, this exact question even appeared in a cartoon I saw, as some sort of attempt to get kids to "think critically" or whatever the buzzword was.

[Maksym Taran]

I think a brain architecture/algorithm that would debate about free will would have been adapted for large amounts of social interaction in its daily life. This interaction would use markedly different skills (eg language) from those of more mundane activities. More importantly it would require a different level of modeling to achieve any kind of good results. One brain would have to contain models for complicated human social, kin and friendly relationships, as well as models for individuals' personalities.

At the center of the mesh of social interactions would be the tightest wad connections. That would be the brain/person, interacting with and modeling all the other members of their tribe/society. However, their brain cannot model itself modeling others with perfect fidelity, and so many simplifications are made even there. These simplifications pile on top of the perceptual differences that a human sees between (itself, other humans) and (everything else). A whole different mental vocabulary arises between descriptions/models of fellow humans and descriptions/models of everything else. Only in humans does it make predictive sense to talk about intent, capability, and inclination, and the wide gap between these kinds of perceived "properties" of fellow socially interacting humans, and the generally much simpler properties seen in inanimate objects and animals, leads the brain to allocate them to widely separated groups of buckets. It is this perceived separation in mental thing-space that leads to the the a free-will boundary being drawn around the cluster of socially interacting humans. When this boundary is objected to, people go their natural arguing ways.

This is just a first attempt, so I think I may have fallen for some of the traps specifically proscribed against in the post. I hope others will attempt to put up their own explanations and maybe even poke some holes in mine :)

I'll definitely pay attention to further comments on this homework assignment.

[Z. M. Davis]

"What kind of cognitive algorithm, as felt from the inside, would generate the observed debate about 'free will'?"

I would say: people have mechanisms for causally modeling the outside world, and for choosing a course of action based on its imagined consequences, but we don't have a mechanism for causally modeling the mechanism within us that makes the choice, so it seems as if our own choices aren't subject to causality (and are thus "freely willed").

However, this is likely to be wrong or incomplete, firstly because it is merely a rephrasing of what I understand to be the standard philosophical answer, and secondly because I'm not sure that I feel done.

[Z. M. Davis]

A difference of predictions between Maksym's proposed answer and mine occurs to me. If the sense of free will comes from not being able to model one's own decision process, rather than from taking the intentional stance towards people but not other things, then I would think that each individual would tend to think that she has free will, but other people don't. Since this is not the default view, my answer must be wrong or very incomplete.

[Chris Hallquist]

"Many philosophers - particularly amateur philosophers, and ancient philosophers - share a dangerous instinct: If you give them a question, they try to answer it."

This line goes in that book you're going to write.

[Ron Hardin]

You will find that the need to nail things down is mostly a male thing. Women are more driven to add complexity, as more interesting to them.

http://home.att.net/~rhhardin9/vickihearne.womenmath.txt

And ``algorithm'' is a picture in Wittgenstein's sense.

[mitchell porter]

A warning to those who would dissolve all their questions:

Why does anything at all exist? Why does this possibility exist? Why do things have causes? Why does a certain cause have its particular effect?

[Paul Crowley]

I don't think this answer meets the standards of rigour that you set above, but I'm increasingly convinced that the idea of free will arises out of punishment. Punishment plays a central role in relations among apes, but once you reach the level of sophistication where you can ask "are we machines", the answer "no" gives the most straightforward philosophical path to justifying your punishing behaviour.

[tcpkac]

Things in thingspace commonly coming within the boundary 'free will' :

moral responsibility
could have done otherwise
possible irrational action
possible self-sacrificial action
gallantry and style (thanks to Kurt Vonnegut for that one)
non-caused agency
I am a point in spacetime and my vector at t+1 has no determinant outside myself
whimsy
'car c'est mon bon désir'
absolute monarchy
you can put a gun at my head and I'll still say 'no'
idealistic non-dualism
consciousness subtending matter
disagreeing with Mum & Dad
disagreeing with the big Mom & Pop up there in the White House
armed response
no taxation without representation
... no taxation even with representation (daft )
'No dear not tonight I've got a headache....'
....

aw hell, just go read Dennett : 'Elbow Room', he did it better than I could.

[anonymous]

Only in humans does it make predictive sense to talk about intent, capability, and inclination, and the wide gap between these kinds of perceived "properties" of fellow socially interacting humans, and the generally much simpler properties seen in inanimate objects and animals, leads the brain to allocate them to widely separated groups of buckets. It is this perceived separation in mental thing-space that leads to the the a free-will boundary being drawn around the cluster of socially interacting humans.

careful there. animistic beliefs are quite widespread in tribal societies, so the notion that the brain allocates two entirely distinct clusters to humans and animals vs. inanimate objects is quite suspect.

[Robin Hanson]

When you're done, you'll know you're done, but unfortunately the reverse implication does not hold.

So when you have the impression you are done, you are not necessarily done because some have this impression without really being done. But then when you are really done, you won't actually know you are done, because you will realize that this impression of being done can be misleading.

[Roko]

So here's your homework problem: What kind of cognitive algorithm, as felt from the inside, would generate the observed debate about "free will"?

I've written up my answer to this on my blog.

I claim that the reason we posit a thing called free will is that almost all of our decision-making processes are amenable to monitoring, analysis and even reversal by “critic” algorithms that reside one (or more) levels higher up. [I say almost all, because the top level has no level above it. The buck really does stop there]. There would probably be no feeling of free will if there were only two levels, but I think that with three or more, you have situations where ... continue reading

[Eliezer Yudkowsky]

Robin: So when you have the impression you are done, you are not necessarily done because some have this impression without really being done. But then when you are really done, you won't actually know you are done, because you will realize that this impression of being done can be misleading.

You'd think it would work that way, but it doesn't. Are you awake or asleep right now? When you're asleep and dreaming, you don't know you're dreaming, so how do you know you're awake?

If you claim you don't know you're awake, there's a series of bets I'd like to make with you...

[Ron Hardin]

Then there's Edmond Jabes, on freedom and how words come to mean anything.

[Robin Hanson]

Eliezer, you seem to be saying that the impression you get when you are really done feels different from the impression you get when you ordinarily seem to be done. But then it should be possible to tell when you just seem to be done, as this impression is different. I can imagine that sometimes our brains just fail to make use of this distinction, but it is quite another to claim that we could not tell when we just seem to be done, no matter how hard we tried.

[Robin Hanson]

Eliezer, also, the bet your proposed would only be enforced in situations where I am not dreaming, so it would really be a bet conditional on not dreaming, which defeats the purpose.

[Caledonian]

When you're asleep and dreaming, you don't know you're dreaming, so how do you know you're awake?

If you claim you don't know you're awake, there's a series of bets I'd like to make with you...

1) Some people claim they can recognize that they're in a dream state.

2) The quoted claims are an example of the rhetorical fallacy known as equivocation.

[Unknown]

When I'm dreaming, I always know I'm dreaming, and when I'm awake I always know I'm awake.

I realize that this doesn't apply to many other people, however... even the second part.

[Unknown]

A fuller explanation of the preceding: As an example of Robin's point, "I can imagine that sometimes our brains just fail to make use of this distinction," the reason that some people don't know when they're dreaming is that that are unable, at that time, to pay attention to all the aspects of their experience; otherwise they would be able to easily distinguish their state from the state of being awake, because the two states are very different, even subjectively. I pay attention to these aspects even while dreaming, and so I recognize that I'm dreaming.

[PK]

Ughh more homework. Overcoming bias should have a sister blog called Overcoming laziness.

[Eliezer Yudkowsky]

Eliezer, you seem to be saying that the impression you get when you are really done feels different from the impression you get when you ordinarily seem to be done. But then it should be possible to tell when you just seem to be done, as this impression is different.

Yes, exactly; it feels different and you can tell the difference - but first you have to have experienced both states, and then you have to consciously distinguish the difference and stay on your guard. Like, someone who understands even classical mechanics on a mathematical level should not be fooled into believing that they understand string theory, if they are at all on their guard against false understanding; but someone who's never understood any physics at all can easily be fooled into thinking they understand string theory.

[Sebastian Hagen]

I think I'll give this a try. Let's start with what a simple non-introspective mind might do:

Init (probably recomputed sometimes, but cached most of the time):
I1. Draws a border around itself, separating itself from the "outside world" in its world model. In humans and similarly embodied intelligences you could get away with defining the own body as "inside", if internal muscle control works completely without inspection.

Whenever deciding on what to output:
A1. Generates a list of all possible next actions of itself, as determined in I1. For human-like embodieds, could be a list of available body movements.
A2. Computes a probability distribution about the likely future states of the world resulting from each action, consulting the internal world-model for prediction. Resolution, temporal range and world-model building are beyond the scope of this answer.
A3. Assigns utilities to each considered future state.
A4. Assigns preferences to the probability distribution of futures. This could e.g. use Expected Utility or some satisficing algorithm.
A5. Chooses the possible next action with the most-prefered distribution of futures. Tie breaking is implementation defined.
A6. Execute that action.

As part of A2, the reactions of other intelligent entites are modeled as part of the general world model. This kind of mind architecture does not model itself in the present; that'd lead to infinite recursion: "I'm thinking about myself thinking about myself thinking about ...". It also wouldn't achieve anything, since the mind as instantiated necessarily has a higher resolution than any model of itself stored inside itself. It will, however, model past (for lost data) or future versions of itself.

The important point here is that this mind doesn't model itself while computing the next action. In the extreme case it needn't have facilities for introspection at all.
Humans obviously have some such facilities. Either inbetween deciding on output, inbetween the individual steps A1-A6, completely in parallel, or some combination of those, humans spend cputime to analyze the algorithm they're executing, to determine systematic errors or possibilities for optimization. I'll call the neural module / program that does this the introspector.
When introspecting a thread which generates motor output (A1-A6 above), one (very effective) assumption of the introspected algorithm will always turn out to be "My own next action is to-be-determined. It'd be ineffective for me to run a model of myself to determine it.". For a mind that doesn't intuitively understand the self-reference and infinite-recursion parts, this turns into "My own actions can't be modeled in advance. I have free will.".

In the cases where the introspected thread is running another instance of the introspector, the introspector still isn't attached to its own thread; doing that would lead to infinite recursion. Each introspector will work similarly to the motor-output-choosing algorithm described above, except that the generated output will be in the form of new mental heuristics. Therefore, the same "It'd be ineffective to run a model of myself to determine my next action." assumption in the algorithm can be observed, and "Free will." is still the likely conclusion of a mind that doesn't understand the rationale behind the design.

[Rolf Nelson]

Most of the proposed models in this thread seem reasonable.

I would write down all the odd things people say about free will, pick the simplest model that explained 90% of it, and then see if I could make novel and accurate predictions based on the model. But, I'm too lazy to do that. So I'll just guess.

Evolution hardwired our cognition to contain two mutually-exclusive categories, call them "actions" and "events."

"Actions" match: [rational, has no understandable prior cause]. "Rational" means they are often influenced by reward and punishment. Synonyms for 'has no understandable prior cause' include 'free will', 'caused by elan vitale' and 'unpredictable, at least by the prediction process we use for things-in-general like rocks'.

"Events" match: [not rational, always directly caused by some previous and intuitively comprehendable physical event or action]. If you throw a rock up, it will come back down, no matter how much you threaten or plead with it.

We are born to axiomatically believe actions we take of this innate 'free will' category have no physical cause. In this model, symptoms might include:

* believing there is an interesting category called 'free will'

* believing that arguing whether humans either belong to, or don't belong to, this 'free will' category, is an interesting question

* believing that if we don't have 'free will', it's wrong to punish people

* believing that if we don't have 'free will', we are marionettes, zombies, or in some other way 'subhuman'.

* believing that if we don't understand what causes a thunderstorm or a crop failure or an eclipse, it is the will of a rational agent who can be appeased through the appropriate sacrifices

* believing that if our actions are caused by God's will, fate, spiritual possession, an ancient prophesy, Newtonian dynamics, or some other simple and easily-understandable cause, we do not have 'free will'. However, if our actions are caused by an immaterial soul, spooky quantum mechanics, or anything else that 'lives in another dimension beyond the grasp of intuitive reason', then we retain 'free will'.

I'm not particularly confident my model is correct, the human capacity to spot patterns where there are none works against me here.

[Tom Breton]

Great post, Rolf Nelson.