Why I Am Not An Integrated Information Theorist (or, The Unconscious Expander)

http://www.scottaaronson.com/blog/?p=1799

by altro • 12 years ago

80 22 12 years ago

apl3• 12 years ago

I appreciate the notion of a "pretty-hard" problem of consciousness; it nicely captures what people in the field, philosophers and scientists alike, are actually looking for.

His counterarguments appear sound, especially the technical concerns. (That's not particularly exciting, though: formalisations of philosophical arguments often crumble under the mathematician's lens. Even the best ones!) They're not exactly new, I'd say, but their rigour is refreshing. My key problem with the overall approach remains the shitty test set we have for any theory of consciousness. It ultimately consists of two elements. If we're being generous, there's three. Namely, almost everyone except for the philosophical extremist agrees that we are conscious and that a stone isn't. IIT works quite well for these cases. Some people would include, say, a cat as a conscious being while maintaining that its "level" of consciousness is reduced. This boils down to

  C_stone < [C_cat <] C_human

which isn't much. There's no real hope for extending it. Any other cases like the ones Aaronson discusses are ones for which we don't even have strong intuitions. Sure, we'd like to think that none of the entities he describes are in fact conscious. But if that's the bullet I have to bite in order to get a decent theory of consciousness, then I might be OK with that.

Udo2• 12 years ago

There is little doubt that the cat is conscious, and the idea that consciousness is a gradient with more than one dimension is not really in dispute. In fact, you can build a very simple gradient model by observing nothing but humans: a human baby is scarcely conscious at all, a toddler is pretty much on par with the cat, and a grown human has a higher level still. The reason why we think this is true is based on at least two things: awareness and meta-awareness (which includes the ability to reason about one's own existence as well as the world in general).

The cat and the toddler are clearly conscious in the sense that they do have a recognizable life experience. They don't just react to the environment, they make models of it, and they have a limited ability to abstract observations. They experience life, and they have complex dreams and emotions. However, they still lack a deeper understanding of causalities, and they can grasp the subjective realities of other beings only on an instinctive level. Adult humans are less limited in this regard, but it's easy to imagine hypothetical beings who perform better still.

That's bad news for the validity of consciousness as a concept, since it seems to be interlinked with intelligence, which itself is a fuzzy idea at best.

So these observations amount to the position that "consciousness" isn't a fundamental property at all, it's the result of other processes and it can come in different flavors. Likewise, the concept of "intelligence" is similarly flawed. In both, we tend to make the mistake of looking at them as simple scalar values. They're not. They're a name we give for a collection of different capabilities. As such, I believe a mathematical model of consciousness is actually pretty unscientific, since consciousness is not an objective property. It's an artificial label we're obsessed with.

apl3• 12 years ago

That's not the notion of consciousness we're dealing with when talking about the "hard" problem of consciousness. You're describing its psycho-functional interpretation: the ability to see yourself as a "self".

Qualia are a largely orthogonal issue. Specifically, it's imaginable that an entity we'd consider unintelligent has a rich and detailed subjective internal life.

rictic3• 12 years ago

There's many more points on the gradient, some of which are of practical importance. A person in deep sleep is less conscious than someone in light sleep who is less conscious than someone blearily waking up who is less conscious than that person when they're wide awake. More practically speaking it is medically and ethically important for a person under anesthesia to have low consciousness.

mtrimpe2• 12 years ago

Namely, almost everyone except for the philosophical extremist agrees that we are conscious and that a stone isn't.

I really don't get why it's so hard for people to imagine consciousness as a property of the universe rather than people.

Everyone seems to intuitively agree that consciousness is a continuum with monkeys having quite a bit, cats having some; plants probably a bit too and stones virtually none.

Given that, it makes much more sense to me that consciousness is an inherent property of the universe, perhaps even the mere passing of time, and that what we traditionally call 'consciousness' is just the degree to which some subset of consciousness is similar to our own.

jerf• 12 years ago

"I really don't get why it's so hard for people to imagine consciousness..."

You name the crazy idea, it's been proposed and taken seriously by someone. I cut off your quote there quite deliberately.

However, as Scott observes, most of them (and I'd daresay including yours) are so vague that it's difficult to even imagine what would falsify them. What does "consciousness is an inherent property of the universe" really mean? Can I bottle some of this inherent consciousness? Can I pour it out of one person and put it in another? Could a universe have exactly equivalent physical laws but lack this "inherent consciousness"? (If not, where in our physical laws does this "inherent consciousness" reside?) Can I talk to disembodied consciousness somehow? Can I build a device (with any theoretical level of complexity you like) that can concretely measure the amount of consciousness embedded in some system? Can you answer these questions in some way that can actually be manifested in some device or practice which can be demonstrated to produce real information of some sort, and not just more handwaving that plays shell games with the fact that it never ever quite gets down to the level of a theory that could even be slightly proved or falsified?

That's the problem. In practice, all vague murmurings about consciousness that have absolutely no ability whatsoever to be used for anything are, ultimately, equivalent. "Pretty words strung in a row with no reference to reality" is one of the most venerable of philosophical traps.

NoMoreNicksLeft2• 12 years ago

almost everyone except for the philosophical extremist agrees that we are conscious and that a stone isn't.

It is possible for me to imagine a conscious being that, mute and immotile, can't give any outward indications of consciousness. I am not an extremist.

So "almost everyone" is not thinking rationally on the subject.

I doubt very much that stones are conscious, it's not worth exploring that question in a practical sense. But to dismiss a theory, even an outlandish one, based only on my admittedly-flawed intution... that's highly unreasonable.

The trouble is that this is no longer a philosophical question, that it can't be a scientific question until we have an objective definition, and worst of all... it's very much a political question now.

How is that you say? It's a political question for many reasons. The abortion controversy touches on it, animal rights touch on it, etc. Everyone's groping for an answer that proves chickens are soulful and shouldn't be eaten, or that it's ok to scrape away their unborn child with a curette because they couldn't be bothered to figure out how to use a condom or take a pill.

Sure, we'd like to think that none of the entities he describes are in fact conscious. But if that's the bullet I have to bite in order to get a decent theory of consciousness, then I might be OK with that.

What's decent about it? This theory is useless in a scientific sense. Can you construct an artificial consciousness if this theory is true? No. Can you test consciousness with it, if it is true? Again, no (though some will dispute this, naturally). Can it lead to other theories that are themselves useful? Well, that remains to be seen, but it's difficult to imagine why that might be so... this doesn't seem like a rough idea that merely needs refinement.

rictic3• 12 years ago

Even knowing the chemical makeup of a stone as being a simple lattice of inert unchanging material? I don't think it's anyone's position that something could be mute and immotile yet conscious, but we know a lot more about rocks than that.

As far as applications, one of the early uses of a practical theory of consciousness may be to determine the effectiveness of anesthesia on a moment by moment basis.

vilhelm_s• 12 years ago

Can you construct an artificial consciousness if this theory is true?

Yes, very easily! The linked blogpost describes how. :)

tunesmith• 12 years ago

Pretty dense article for what seems to be a really daffy definition of consciousness, so maybe someone can summarize? It seems to just be the difference between systems thinking and reductionism, but why would any irreducible concept be evidence of consciousness? Why on earth would that idea make any more sense than, say, a math problem that is too hard for a 4th-grader being evidence of consciousness? Or a locked machine, or a patented process, etc? There's nothing intrinsically special about an irreducible process other than it being irreducible - it's not like it's mystical or anything.

gone35• 12 years ago

Close. Aaronson's most devastating argument (he offers two more) is that Tonini's \Phi or "integrated information" of an input-output system (function), under a reasonable operationalization as a measure of how correlated subsets of inputs are with subsets of outputs, is just not good measure of "consciousness" because, as it happens, large families of rather mundane systems that one wouldn't think as "conscious" in fact have provably large "integrated information" or \Phi by design --such as Reed-Solomon codes, for instance.

Put in another way, if Tonini were right then your (say) portable DVD player ought to be "conscious" because, surprisingly, the amount of "integrated information" achieved by the scratch- and skip-tolerant error-correcting code it uses internally has to be huge in order for it to work. But that is prima facie ridiculous; so either we are wrong and the DVD player is in fact "conscious" or, more likely, Tononi's proposed definition of consciousness is lacking.

Again as I said Aaronson makes two other points, but I think this one alone is the most conclusive.

logicallee• 12 years ago

It's not prima facie ridiculous that the DVD player might be conscious, as let's assume it is prima facie ridiculous - then it would be prima facie ridiculuos in 1969 with a tape player, in 1979 with a VCR, in 1989 with a CD-ROM drive, 1999 with a DVD-ROM drive, in 2009 with a Blu-RAY drive, in 2019 with x, in 2029 with y, and there is no reason it shouldn't still be prima facie ridiculous even if at some point along the way the thing happens to be conscious as a side-effect, without this being necessarily visible in its outputs. So you can't just call it prima facie ridiculous and be done with it - you need some other argument.

andyjohnson0• 12 years ago

I wanted to like this, but IIT just seems like a lot of hand-waving.

"to hypothesize that a physical system is “conscious” if and only if it has a large value of Φ"

By this measure, would a long mathematical proof or an orchestral symphony be conscious? If so, how does this actually help us understand how subjectivity relates (or doesn't) to all this?

[1] http://en.wikipedia.org/wiki/List_of_long_proofs

alokm• 12 years ago

I have had the fortune of working on this IIT right after my IIT :). I worked on implementing the research software used by Giulio Tononi and his research team. I added a visualizer in OpenGL, and optimized the calculations for calculating integrated information.

byerley• 12 years ago

I'd argue that the author's views are too concerned with intuition. If we reduce consciousness to a scale rather than a binary property; of course a thermostat is somewhere on the scale, it simply has such a small value that it's not worth considering - leading to our philosophical intuition. There's nothing magical about our level of consciousness; as unintuitive as that may be. Aaronson argues that the model must yield to our intuition, but if the model is consistent and explains our observations, our intuition should yield to it. - the obvious analog here being quantum mechanics

In regards to saying "both that the Hard Problem is meaningless, and that progress in neuroscience will soon solve the problem if it hasn’t already," neuroscientists and mathematicians too often overlook the Turing test here. It's consistent (if not accurate) to say that the "Hard Problem" is ill-defined, but maintain that we've clearly solved it if we can comprehensively beat the Turing test. That's what the Turing test was designed for, knowing that we've created a conscious machine because it's indistinguishable from a human; even if we can't decide on what consciousness is.

Why I Am Not An Integrated Information Theorist (or, The Unconscious Expander)

Related Stories