[xnull]

Integrated Information Theory I think is a fad, and a bad one at that.

One could think of integration of information as a necessary but not sufficient condition for consciousness a la Scott Aaronson [1] who applies more rigorous mathematical examples of integrated systems - in fact ones close to maximal possible integrated information - that very clearly are unlikely have any measure of consciousness (like expander graphs and error correcting codes). What about ciphers, where small changes to a single bit in key, ciphertext or plaintext radically alters the evolution of the internals and output of the system?

But even that might be too strong a statement. Aren't minimally interacting systems such as Conway's Game of Life Turing Complete? There are tiny, tiny Universal Turing Machines, and UTMs are essentially free to encode/decode their tape inputs in any way they see fit.

Ultimately while IIT has a 'mathematical description' it mostly only serves to obfuscate a naive and trivial notion that 'the system has to be complicated and has to integrate and process local information in some holistic global way'.

No shit.

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

[SapphireSun]

Not quite. I attended a talk by Dr. Koch at MIT. I didn't get all of it, but it seemed to me that his definition tries to figure out what's included or excluded in conscious systems and how conscious they are. It may or may not be right, but it at least it's working towards something with predictive value.

The examples you gave are mostly abstract and not physical. If consciousness is a physical phenomenon, these algorithms must be manifest before they can qualify. As for the game of life, the pieces on a physical board aren't actually interacting with each other with feedback loops. In a computer, the circuits are mostly feed forward and the algorithm is simulated rather than physically realized.

[xnull]

The examples though _do_ have physical manifestations - everything that exists in the universe has a physical manifestation (by tautology). How is a feed forward circuit carrying information (to be integated) not a physical realization? It's not clear why neurons carrying potentials and mixing them count but not an FPGA (no longer an ALU simulation) doing cryptographic operations for a bitcoin mining rig. Furthermore as an FPGA more directly implements the information integration system than an ALU fetch-execute cycle, is it 'more conscious'? For me that sounds absurd.

Presume the Game of Life _did_ have a direct physical implementation, let's say neurons fire according to GOL rules. Would you then be forced to call it conscious or not? Wouldn't it depend on the cell configuration? An all blank cell configuration certainly isn't conscious but the system would have the same low IIT score as a manifestation running some UTM-equivalent setup.

Finally, what about physical systems that do highly integrate information? N-body problems exhibit highly integrated behavior manifest in the so-called butterfly effect. Why do the air molecules in a balloon not count as conscious? This physical example escapes the "America" criticism - every molecule applies a force on every other and due to n-body mechanics the system is highly integrated.

My worry, which I think is mostly corroborated, is that the definite clarity of the IIT definition will serve to obscure the fact that underlying definition has no real definite clarity or derivation.

Ultimately the IIT formula was imagined to capture some heuristic notions - but it was not derived. For this reason, added to those above, I am very skeptical.

Or maybe it's my CS background that makes me biased toward thinking that _how_ information is integrated matters purely more than some measure of how much mixing there is (e.g. by one of many versions [5 now?] of the IIT formula).

[SapphireSun]

So I'm not exactly sure how an FPGA is wired, so I won't comment on that. However, I feel, but cannot prove that the GOL neurons might have some kind of internal experience.

Your air example is a good one, however, the kind of interactions they have depend on temperature. Also, the information a air molecules transmit has little explicit encoding other than a few attributes, so I imagine that it wouldn't feel like much to be air. Feedback loops are probably frequently formed and broken, so I'm not sure what to make of that.

I think the IIT formula is interesting. I'm not sure if it's right (it probably isn't and will need modifications at least. Dr. Koch talked about this version of it being invariant across time, which sounded off to me). I'm planning on learning more about it though because it sounds like might have at least part of the answer and it might be on the path to finding something more experimentally amenable which is more than I can say of most theories I hear about.

I'm also attracted to the physicality requirements of IIT because it's clear to me that abstract computation alone is not enough to reify physical phenomena. For instance, mathematics often returns imaginary solutions to physical problems. Some math represents reality, but not all of it. It's important to stick close to physical processes in physics.

I haven't yet heard that there were five versions of the formula, but that doesn't surprise me. It's actively under development and it hasn't hit the point where we can conduct experiments to narrow it down. As far as I'm concerned, it's an interesting idea at this point, and I hope I can take away some useful tools from it.