An excellent question Abolfazl. What does integration mean in this theory? Information, in a traditional Shannon reading at least, consists of signals which must have senders and receivers. To be integrated I think the signals must all impinge on at least one receiving unit. Otherwise nothing gets 'the whole message'. But what is a receiving unit in a brain? I think the only valid answer here is a cell, since if some signals are received by some cells in a 'system' and others by others then they have no causal relation that can account for their contents (meanings) being integrated. The problem is that it is doubtful that this leaves us with a useful definition for integration within a 'system'. Most brain cells send signals to about 10,000 cells. For any group of signals there are likely to be cells that get some and not others, others that get the converse and some that get none, with maybe a few getting all. We seem to be left with the only watertight definition being a group of signals for which at least one cell gets all of them. You do not need any synchronisation for that because if all the signals arrive together no other form of 'binding' is needed. You may nevertheless need synchrony in earlier stages to ensure that this set of signals is set off in the first place.

To my mind, although phi is a neat idea in some ways the only situation in which it can actually be defined is in terms of all those signals arriving at at least one cell. So the 'system' is most usefully seen as that cell. Almost everyone says this cannot possibly explain the richness of conscious experience but without any cogent supporting arguments. If a cell receives 40,000 signals and nothing else receives more then a cell is our best bet for richness.

One day the penny will drop!!

That's an idea, But a few comments:

1_ if we argue that consciousness arises from activities in a "cell" we would then have to say we need "new cells" for our unprecedented experiences.(since these experiences happen in order of milliseconds this idea seems unlikely). But I don't think you meant that.

2_Other possibility is to say different activities of a "NEURON" (i.e. different firing patterns) accounts for consciousness. if it is the case, we should be able to "Isolate" "consciousness producing nucleus" (or nuclei, say) in the brain. sth that seems to be unlikely. But I don't think you meant it neither.

3_"different neurons" in "different places" account for "different conscious experiences". that would be a fine assumption. But it is still suffers from the concept of "one neuron alone" can produce consciousness.

or maybe I'm totally wrong!

I found that my question has been proposed by Seth et al(1) previously. as they say based on IIT: "The focus on capacity leads to the counterintuitive prediction that a brain with a high value of Φ but displaying no activity at all would be conscious. "

one point : why don't we look at the consciousness as an emergent property of a system as whole?

1:

Seth, Anil K., et al. "Theories and measures of consciousness: an extended framework." Proceedings of the National Academy of Sciences 103.28 (2006): 10799-10804.

P.S sorry for the delay

Yes, Abolfazl, I think you are totally wrong, but entertaining that possibility is surely the key to progress in science.

I think the key mistake is to think that what we are conscious of relates to outputs. Consciousness is being aware of the effects of the world on us, it is not an output of us. We are only aware of our outputs indirectly through further feedback inputs. Consciousness, or experience, more or less MEANS input. Moreover, Tononi's idea derives from arguments about combinatorial possibilities of INPUTS. If you consider several signals that all input into the same thing then the number of meanings possible grows exponentially with the number of signals. That is what Tononi latches on to. Yet like so many other people in neuroscience he then tells the story as if it was a pattern of outputs that was of interest. But outputs do not 'integrate' except in the sense that they converge to form some further input next along. This is a very basic point that so many people seem to miss - even Tononi maybe.

So if what hosts consciousness is a neuron then we do not need a new neuron for each new thought. With 50,000 input channels and 20msec cycles no neuron in the brain will ever have time to receive all its possible combinations of inputs in a lifetime there are so many options. So we are not looking for cells that 'produce consciousness' either. We just need some cells with multiple input channels. You would not want different cells in different places to host different conscious experiences because that would severely restrict the computational routes that could be associated with an idea. What is much more plausible is that millions of cells all get a pattern of inputs determined by your current sensations and are conscious of that pattern and those cells that have some useful inference to make about that pattern fire. Nobody is suggesting that one neuron alone in a head produces consciousness. The idea is that each of a very large number of neurons hosts a 'copy' of an experience of the world. This experience of the world would not include the idea that other cells were hosting something similar (it would be silly to have this inference about one's inside functions get in the way of watching a good movie) unless some very very complex inference process had been put in train - which is what I am doing here but maybe nobody did before 2001 when Steve Sevush and I both stumbled on the idea.

I agree that Seth points out the essential paradox in Tononi's idea if you base it on correlations between cell firings. If you base it on inputs, which, as I say, is what the idea actually is based on originally there is no problem but you can only consider one neuron at a time.

There are no emergent properties of systems. That is a dead idea I think. If systems with parts can be simulated on computers then their behaviour is no more than the behaviour of their parts and so invoking a system adds nothing to the analysis. What we actually want are emergent receiving entities, that can handle large numbers of inputs indivisibly in a way that is different from many entities each having some of the inputs. That is what Tononi is struggling to create but confusing it with a system. Modern physics actually shows where to find such entities quite nicely. All fundamental entities are like that but systems with parts are never like that, as Leibniz pointed out.

Best wishes

Jo Edwards

( The basis for all this is discussed at http://www.ucl.ac.uk/jonathan-edwards )

Dear Vittorio,

IIT is about the advantage of combinatoriality. That needs combining and the only sites of combining in brains are single cells. It is such a fundamental point it is difficult to know how to say it otherwise. What I thin may be the problem is that it is unclear what 'integrated' means.

What could you mean by 'integrated outputs'. I do not understand that. As indicated before, the only thing I can think of it might mean is synchronised, but that is only causally relevant to the subsequent co-inputs to individual cells further along. Synchrony of output spikes themselves is of no causal relevance. You cannot make use of synchrony to combine more signals than can combine (in an individual dendritic tree.) There is a strange tendency in neuroscience today to start out with a causal physical story and then let it drift off into something that sounds neat but is causally impossible.

So what could you mean by 'integrated outputs'. I am intrigued. And where do thalamic nuclei come in?

Best wishes

Jo

Dear Jonathan,

I think that "integration" in IIT can be understood more clearly in terms of structural dissociation of human personality. The aim of therapy in the case of structural dissociation is integration of different parts of patient's personality. It doesn't mean, that there is some "superpersonality", that must integrate all the parts. Instead, different parts become more reciprocally penetrated by enhancing mutual communication, they share memories, skills, knowledge about each other and, eventually, the identity of the patient. At the same time, without therapy, each part expresses much lower level of consciousness (the lower psychic tendencies after Janet).

I mean that "integration" in IIT is not about some further processing, but it is about internal communication -- an exchange of information within a system.

Dear Ksenilia,

I don't think that can be right. IIT has a specific mathematical basis relating to combinatorial richness with increasing numbers of degrees of freedom. That richness is then attributed to a set of axonal spikes (each with one degree of freedom) that are considered to be 'integrated'. Integration in this sense cannot be communication, unless of course you read it the way I suggest - that it is the integration that occurs when a set of neurons communicate with another neuron receiving the spikes together. I don't think this has anything to do with personality dissociation. Integration means lots of different things to different people! The problem for IIT is in working out exactly what it does mean for that theory.

Hi Abolfazl,

I agree we cannot rely on IIT too much but I am not sure that the conclusion that a Hopfield network would be conscious is crucial. I think the deeper problem is that any assignation of phi is arbitrary except for a set of signals that are convergent on a single event, as Leibniz pointed out. (They knew enough neuroanatomy in the seventeenth century to have the same conversation we are having but people forget historical literature.)

The 'well-establishment' of an idea is never a scientific argument, Abolfazl. The received wisdom is usually wrong. And in philosophical circles emergence is highly contentious. I personally think that emergence does exist, but not in complex aggregate systems. These are explained by their parts. Examples of emergence I would give would be chirality or the acquisition of Goldstone modes by crystals. Goldstone modes can be completely unrelated to any parts.

I am not sure that you really mean 'disambiguate'. Maybe you just mean 'say what you mean'. But I think I did. It was just a bit surprising. The idea that one neuron can be conscious has been suggested by many people for three hundred years (since cells were seen). There is nothing new about it and in the nineteenth century it was commonplace, when it was called polyzoism. The theory is very much alive and well, with some recent modifications variously by myself, Steven Sevush and Erhardt Bieberich.

My proposal would certainly entail that connecting a single neuron to very carefully designed inputs could give it a sense of watching 'Star Wars' on DVD, but the problem is not that this is implausible, simply that ascertainment of the existence of the experience is impossible. We need more subtle approaches. They may pose intractable problems but maybe no more than the Higgs boson.

You think it unlikely that a single cell could host an experience of Star Wars. But an intuitive sense of implausibility needs to backed by scientific argument. Why should this be implausible, other than in intuitive terms? William James said in 1890 that it was the only solution that was not self-contradictory. I don't think much has changed since then. We need to be 'open minded'.

Very best wishes

Jo