Why I Am Not An Integrated Information Theorist (or, The Unconscious Expander)
scottaaronson.com
scottaaronson.com
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
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.
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.
> I would call it duck typing
> It's not clear that we need a theory of consciousness at
> all, how is something as ill-defined as consciousness even
> viable in a mathematical context?
Yes, just as its possible that an entity we'd consider intelligent doesn't, but since it's subjective, its not subject to empirical -- and, hence, scientific -- verification. Anything we are going to say objectively on issues related to consciousness isn't going to address that, because that's a question outside the scope of science.
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.
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.
Essentially consciousness is nothing more than the mere passage of time; change itself. And if the universe is not deterministic that the required 'magic spark' is unlikely to be reserved to just (our) brains and is more likely a universal property of the universe.
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.
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.
Until someone tells me what consciousness is, that's impossible to say one way or the other.
Do you know what it is? I can make some vague descriptions myself that I think are supportable (it's related to intelligence), but past that I'm willing to admit we don't even have a starting point to work with.
It seems to me to be the modern not-quite-atheist-hipster soul. The term is used most often when someone wants to prove to me that it's a sin to eat meat (that chicken was a conscious being!). I see quite a few headlines that are mangled, claiming that a neurologist has proven the honeybee or an orca to be conscious, but when I read the articles themselves it always turns out that the MRI machine has similar readings to those of a human.
None of this is meaningful.
> 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.
If we're daydreaming, sure. But that's one of dozens. We're probably looking at the basis for AI too, if we get one. Whether or not that can be accomplished in software. Education will benefit... you'll be able to determine whether a student is learning, or how best to accomplish that. Intelligence testing, personality tests (don't elect him, he only scored an 8 on the non-sociopath test!), lie detection, etc.
But I've yet to see a theory of consciousness (debunked or not) that could lead to any of this. Even if the author is wrong, and that theory is correct... even if it is only preliminary and needs to be polished, where will it give us any of this? It doesn't even point us in the direction of a more complete theory, and none of them do.
Yes, very easily! The linked blogpost describes how. :)
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.
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.
"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?