Posts tagged ‘Tononi’

You can’t build experience out of mere information. Not, at any rate, the way the Integrated Information Theory (IIT) seeks to do it. So says Garrett Mindt in a forthcoming paper for the JCS.

‘Information’ is notoriously a slippery term, and much depends on how you’re using it. Commonly people distinguish the everyday meaning, which makes information a matter of meaning or semantics, and the sense defined by Shannon, which is statistical and excludes meaning, but is rigorous and tractable.

It is a fairly common sceptical claim that you cannot get consciousness, or anything like intentionality or meaning, out of Shannon-style information. Mindt describes in his paper a couple of views that attack IIT on similar grounds. One is by Cerullo, who says:

‘Only by including syntactic, and most importantly semantic, concepts can a theory of information hope to model the causal properties of the brain…’

The other is by Searle, who argues that information, correctly understood, is observer dependent. The fact that this post, for example, contains information depends on conscious entities interpreting it as such, or it would be mere digital noise. Since information, defined this way, requires consciousness, any attempt to derive consciousness from it must be circular.

Although Mindt is ultimately rather sympathetic to both these cases, he says they fail because they assume that IIT is working with a Shannonian conception of information: but that’s not right. In fact IIT invokes a distinct causal conception of information as being ‘a difference that makes a difference’. A system conveys information, in this sense, if it can induce a change in the state of another system. Mindt likens this to the concept of information introduced by Bateson.

Mindt makes the interesting point that Searle and others tend to carve the problem up by separating syntax from semantics; but it’s not clear that semantics is required for hard-problem style conscious experience (in fact I think the question of what, if any, connection there is between the two is puzzling and potentially quite interesting). Better to use the distinction favoured by Tononi in the context of IIT, between extrinsic information – which covers both syntax and semantics – and intrinsic, which covers structure, dynamics, and phenomenal aspects.

Still, Mindt finds IIT vulnerable to a slightly different attack. Even with the clarifications he has made, the theory remains one of structure and dynamics, and physicalist structure and dynamics just don’t look like the sort of thing that could ever account for the phenomenal qualities of experience. There is no theoretical bridge arising from IIT that could take us across the explanatory gap.

I think the case is well made, although unfortunately it may be a case for despair. If this objection stands for IIT then it most likely stands for all physicalist theories. This is a little depressing because on one point of view, non-physicalist theories look unattractive. From that perspective, coming up with a physical explanation of phenomenal experience is exactly the point of the whole enquiry; if no such explanation is possible, no decent answer can ever be given.

It might still be the case that IIT is the best theory of its kind, and that it is capable of explaining many aspects of consciousness. We might even hope to squeeze the essential Hard Problem to one side. What if IIT could never explain why the integration of information gives rise to experience, but could explain everything, or most things, about the character of experience? Might we not then come to regard the Hard Problem as one of those knotty tangles that philosophers can mull over indefinitely, while the rest of us put together a perfectly good practical understanding of how mind and brain work?

I don’t know what Mindt would think about that, but he rounds out his case by addressing one claimed prediction of IIT; namely that if a large information complex is split, the attendant consciousness will also divide. This looks like what we might see in split-brain cases, although so far as I can see, nobody knows whether split-brain patients have two separate sets of phenomenal experiences, and I’m not sure there’s any way of testing the matter. Mindt points out that the prediction is really a matter of  ‘Easy Problem’ issues and doesn’t help otherwise: it’s also not an especially impressive prediction, as many other possible theories would predict the same thing.

Mindt’s prescription is that we should go back and have another try at that definition of information; without attempting to do that he smiles on dual aspect theories. I’m afraid I am left scowling at all of them; as always in this field the arguments against any idea seem so much better than the ones for.


doorknob‘…stupid as a doorknob…’ Just part of Luboš Motl’s vigorous attack on Scott Aaronson’s critique of IIT, the Integrated Information Theory of Giulio Tononi.

To begin at the beginning. IIT says that consciousness arises from integrated information, and proposes a mathematical approach to quantifying the level of integrated information in a system, a quantity it names Phi (actually there are several variant ways to define Phi that differ in various details, which is perhaps unfortunate). Aaronson and Motl both describe this idea as a worthy effort but both have various reservations about it – though Aaronson thinks the problems are fatal while Motl thinks IIT offers a promising direction for further work.

Both pieces contain a lot of interesting side discussion, including Aaronson’s speculation that approximating phi for a real brain might be an NP-hard problem. This is the digression that prompted the doorknob comment: so what if it were NP-hard, demands Motl; you think nature is barred from containing NP-hard problems?

The real crux as I understand it is Aaronson’s argument that we can give examples of systems with high scores for Phi that we know intuitively could not be conscious. Eric Schwitzgebel has given a somewhat similar argument but cast in more approachable terms; Aaronson uses a Vandermonde matrix for his example of a high-phi but intuitively non-conscious entity, whereas Schwitzgebel uses the United States.

Motl takes exception to Aaronson’s use of intuition here. How does he know that his matrix lacks consciousness? If Aaronson’s intuition is the test, what’s the point of having a theory? The whole point of a theory is to improve on and correct our intuitive judgements, isn’t it? If we’re going to fall back on our intuitions argument is pointless.

I think appeals to intuition are rare in physics, where it is probably natural to regard them as illegitimate, but they’re not that unusual in philosophy, especially in ethics. You could argue that G.E. Moore’s approach was essentially to give up on ethical theory and rely on intuition instead. Often intuition limits what we regard as acceptable theorising, but our theories can also ‘tutor’ and change our intuitions. My impression is that real world beliefs about death, for example, have changed substantially in recent decades under the influence of utilitarian reasoning; we’re now much less likely to think that death is simply forbidden and more likely to accept calculations about the value of lives. We still, however, rule out as unintuitive (‘just obviously wrong’) such utilitarian conclusions as the propriety of sometimes punishing the innocent.

There’s an interesting question as to whether there actually is, in itself, such a thing as intuition. Myself I’d suggest the word covers any appealing pre-rational thought; we use it in several ways. One is indeed to test our conclusions where no other means is available; it’s interesting that Motl actually remarks that the absence of a reliable objective test of consciousness is one of IIT’s problems; he obviously does not accept that intuition could be a fall-back, so he is presumably left with the gap (which must surely affect all theories of consciousness). Philosophers also use an appeal to intuition to help cut to the chase, by implicitly invoking shared axioms and assumptions; and often enough ‘thought experiments’ which are not really experiments at all but in the Dennettian phrase ‘intuition pumps’ are used for persuasive effect; they’re not proofs but they may help to get people to agree.

Now as a matter of fact I think in Aaronson’s case we can actually supply a partial argument to replace pure intuition. In this discussion we are mainly talking about subjective consciousness, the ‘something it is like’ to experience things. But I think many people would argue that that Hard Problem consciousness requires the Easy Problem kind to be in place first as a basis. Subjective experience, we might argue, requires the less mysterious apparatus of normal sensory or cognitive experience; and Aaronson (or Schwitzgebel) could argue that their example structures definitely don’t have the sort of structure needed for that, a conclusion we can reach through functional argument without the need for intuition,

Not everybody would agree, though; some, especially those who lean towards panpsychism and related theories of ‘consciousness everywhere’ might see nothing wrong with the idea of subjective consciousness without the ‘mechanical’ kind. The standard philosophical zombie has Easy Problem consciousness without qualia; these people would accept an inverted zombie who has qualia with no brain function. It seems a bit odd to me to pair such a view with IIT (if you don’t think functional properties are required I’d have thought you would think that integrating information was also dispensable) but there’s nothing strictly illogical about it. Perhaps the dispute over intuition really masks a different disagreement, over the plausibility of such inverted zombies, obviously impossible in  Aaronson’s eyes, but potentially viable in Motl’s?

Motl goes on to offer what I think is a rather good objection to IIT as it stands; ie that it seems to award consciousness to ‘frozen’ or static structures if they have a high enough Phi score. He thinks it’s necessary to reformulate the idea to capture the point that consciousness is a process. I agree – but how does Motl know consciousness requires a process? Could it be that it’s just…  intuitively obvious?

boxers…for two theories?

Ihtio kindly drew my attention to an interesting paper which sets integrated information theory (IIT) against its own preferred set of ideas – semantic pointer competition (SPC). I’m not quite sure where this ‘one on one’ approach to theoretical discussion comes from. Perhaps the authors see IIT as gaining ground to the extent that any other theory must now take it on directly. The effect is rather of a single bout from some giant knock-out tournament of theories of consciousness (I would totally go for that, incidentally; set it up, somebody!).

We sort of know about IIT by now, but what is SPC? The authors of the paper, Paul Thagard and Terrence C Stewart, suggest that:

consciousness is a neural process resulting from three mechanisms: representation by firing patterns in neural populations, binding of representations into more complex representations called semantic pointers, and competition among semantic pointers to capture the most important aspects of an organism’s current state.

I like the sound of this, and from the start it looks like a contender. My main problem with IIT is that, as was suggested last time, it seems easy enough to imagine that a whole lot of information could be integrated but remain uniluminated by consciousness; it feels as if there needs to be some other functional element; but if we supply that element it looks as if it will end up doing most of the interesting work and relegate the integration process to something secondary or even less important. SPC looks to be foregrounding the kind of process we really need.

The authors provide three basic hypotheses on which SPC rests;

H1. Consciousness is a brain process resulting from neural mechanisms.
H2. The crucial mechanisms for consciousness are: representation by patterns of firing in neural populations, binding of these representations into semantic pointers, and competition among semantic pointers.
H3. Qualitative experiences result from the competition won by semantic pointers that unpack into neural representations of sensory, motor, emotional, and verbal activity.

The particular mention of the brain in H1 is no accident. The authors stress that they are offering a theory of how brains work. Perhaps one day we’ll find aliens or robots who manage some form of consciousness without needing brains, but for now we’re just doing the stuff we know about. “…a theory of consciousness should not be expected to apply to all possible conscious entities.”

Well, actually, I’d sort of like it to – otherwise it raises questions about whether it really is consciousness itself we’re explaining. The real point here, I think, is meant to be a criticism of IIT, namely that it is so entirely substrate-neutral that it happily assigns consciousness to anything that is sufficiently filled with integrated information. Thagard and Stewart want to distance themselves from that, claiming it as a merit of their theory that it only offers consciousness to brains. I sympathise with that to a degree, but if it were me I’d take a slightly different line, resting on the actual functional features they describe rather than simple braininess. The substrate does have to be capable of doing certain things, but there’s no need to assume that only neurons could conceivably do them.

The idea of binding representations into ‘semantic pointers’ is intriguing and seems like the right kind of way to be going; what bothers me most here is how we get the representations in the first place. Not much attention is given to this in the current paper: Thagard and Stewart say neurons that interact with the world and with each other become “tuned” to regularities in the environment. That’s OK, but not really enough. It can’t be that mere interaction is enough, or everything would be a prolific representation of everything around it; but picking out the right “regularities” is a non-trivial task, arguably the real essence of representation.

Competition is the way particular pointers get selected to enter consciousness, according to H2; I’m not exactly sure how that works and I have doubts about whether open competition will do the job. One remarkable thing about consciousness is its coherence and direction, and unregulated competition seems unlikely to produce that, any more than a crowd of people struggling for access to a microphone would produce a fluent monologue. We can imagine that a requirement for coherence is built in, but the mechanism that judges coherence turns out to be rather important and rather difficult to explain.

So does SPC deliver? H3 claims that it gives rise to qualitative experience: the paper splits the issue into two questions: first, why are there all these different experiences, and second, why is there any experience at all? On the first, the answers are fairly good, but not particularly novel or surprising; a diverse range of sensory inputs and patterns of neural firing naturally give rise to a diversity of experience. On the second question, the real Hard Problem, we don’t really get anywhere; it’s suggested that actual experience is an emergent property of the three processes of consciousness. Maybe it is, but that doesn’t really explain it. I can’t seriously criticise Thagard and Stewart because no-one has really done any better with this; but I don’t see that SPC has a particular edge over IIT in this respect either.

Not that their claim to superiority rests on qualia; in fact they bring a range of arguments to suggest that SPC is better at explaining various normal features of consciousness. These vary in strength, in my opinion. First feature up is  how consciousness starts and stops. SPC has a good account, but I think IIT could do a reasonable job, too. The second feature is how consciousness shifts, and this seems a far stronger case; pointers naturally lend themselves better to thus than the gradual shifts you would at first sight expect from a mass of integrated information. Next we have a claim that SPC is better at explaining the different kinds or grades of consciousness that fifteen organisms presumably have. I suppose the natural assumption, given IIT, would be that you either have enough integration for consciousness or you don’t. Finally, it’s claimed that SPC is the winner when it comes to explaining the curious unity/disunity of consciousness. Clearly SPC has some built-in tools for binding, and the authors suggest that competition provides a natural source of fragmentation. They contrast this with Tononi’s concept of quantity of consciousness, an idea they disparage as meaningless in the face of the mental diversity of the organisms in the world.

As I say, I find some of these points stronger than others, but on the whole I think the broad claim that SPC gives a better picture is well founded. To me it seems the advantages of SPC mainly flow from putting representation and pointers at the centre. The dynamic quality this provides, and the spark of intentionality, make it better equipped to explain mental functions than the more austere apparatus of IIT. To me SPC is like a vehicle that needs overhauling and some additional components (some of those not readily available); it doesn’t run just now but you can sort of see how it would. IIT is more like an elegant sculptural form which doesn’t seem to have a place for the wheels.

jennifer2Are ideas conscious at all? Neuroscience of Consciousness is a promising new journal from OUP introduced by the editor Anil Seth here. It has an interesting opinion piece from David Kemmerer which asks – are we ever aware of concepts, or is conscious experience restricted to sensory, motor and affective states?

On the face of it a rather strange question? According to Kemmerer there are basically two positions. The ‘liberal’ one says yes, we can be aware of concepts in pretty much the same kind of way we’re aware of anything. Just as there is a subjective experience when we see a red rose, there is another kind of subjective experience when we simply think of the concept of roses. There are qualia that relate to concepts just as there are qualia that relate to colours or smells, and there is something it is like to think of an idea. Kemmerer identifies an august history for this kind of thinking stretching back to Descartes.

The conservative position denies that concepts enter our awareness. While our behaviour may be influenced by concepts, they actually operate below the level of conscious experience. While we may have the strong impression that we are aware of concepts, this is really a mistake based on awareness of the relevant words, symbols, or images. The intellectual tradition behind this line of thought is apparently a little less stellar – Kemmerer can only push it back as far as Wundt – but it is the view he leans towards himself.

So far so good – an interesting philosophical/psychological issue. What’s special here is that in line with the new journal’s orientation Kemmerer is concerned with the neurological implications of the debate and looks for empirical evidence. This is an unexpected but surely commendable project.

To do it he addresses three particular theories. Representing the liberal side he looks at Global Neural Workspace Theory (GNWT) as set out by Dehaene, and Tononi’s Integrated information Theory (IIT)’ on the conservative side he picks the Attended Intermediate-Level Representation Theory (AIRT) of Prinz. He finds that none of the three is fully in harmony with the neurological evidence, but contends that the conservative view has distinct advantages.

Dehaene points to research that identified specific neurons in a subject’s anterior temporal lobes that fire when the subject is shown a picture of, say, Jennifer Aniston (mentioned on CE – rather vaguely). The same neuron fires when shown photographs, drawing, or other images, and even when the subject is reporting seeing a picture of Aniston. Surely then, the neuron in some sense represents not an image but the concept of Jennifer Aniston?  against theconservative view Kemmerer argues that while a concept may be at work, imagery is always present in the conscious mind; indeed, he contends,  you cannot think of ‘Anistonicity’ in itself without a particular image of Aniston coming to mind. Secondly he quotes further research which shows that deterioration of this portion of the brain impairs our ability to recognise, but not to see, faces. This, he contends, is good evidence that while these neurons are indeed dealing with general concepts at some level, they are contributing nothing to conscious awareness, reinforcing the idea that concepts operate outside awareness. According to Tononi we can be conscious of the idea of a triangle, but how can we think of a triangle without supposing it to be equilateral, isosceles, or scalene?

Turning to the conservative view, Kemmerer notes that AIRT has awareness at a middle level, between the jumble of impressions delivered by raw sensory input on the one hand, and the invariant concepts which appear at the high level. Conscious information must be accessible but need not always be accessed.  It is implemented as gamma vector waves. This is apparently easier to square with the empirical data than the global workspace, which implies that conscious attention would involve a shift into the processing system in the lateral prefrontal cortex where there is access to working memory – something not actually observed in practice. Unfortunately although the AIRT has a good deal of data on its side the observed gamma responses don’t in fact line up with reported experience in the way you would expect if it’s correct.

I think the discussion is slightly hampered by the way Kemmerer uses ‘awareness’ and ‘consciousness’ as synonyms. I’d be tempted to reserve awareness for what he is talking about, and allow that concepts could enter consciousness without our being (subjectively) aware of them. I do think there’s a third possibility being overlooked in his discussion – that concepts are indeed in our easy-problem consciousness while lacking the hard-problem qualia that go with phenomenal experience. Kemmerer alludes to this possibility at one point when he raises Ned Block’s distinction between access and phenomenal  (a- and p-consciousness), but doesn’t make much of it.

Whatever you think of Kemmerer’s ambivalent;y conservative conclusion, I think the way the paper seeks to create a bridge between the philosophical and the neurological is really welcome and, to a degree, surprisingly successful. If the new journal is going to give us more like that it will definitely be a publication to look forward to.


measureThere were reports recently of a study which tested different methods for telling whether a paralysed patient retained some consciousness. In essence, PET scans seemed to be the best, better than fMRI or traditional, less technically advanced tests. PET scans could also pick out some patients who were not conscious now, but had a good chance of returning to consciousness later; though it has to be said a 74% success rate is not that comforting when it comes to questions of life and death.

In recent years doctors have attempted to diagnose a persistent vegetative state in unresponsive patients, a state i which a patient would remain alive indefinitely (with life support) but never resume consciousness; there seems to be room for doubt, though about whether this is really a distinct clinical syndrome or just a label for the doctor’s best guess.

All medical methods use proxies, of course, whether they are behavioural or physiological; none of them aspire to measure consciousness directly. In some ways it may be best that this is so, because we do want to know what the longer term prognosis is, and for that a method which measures, say, the remaining blood supply in critical areas of the brain may be more useful than one which simply tells you whether the patient is conscious now. Although physiological tests are invaluable where a patient is incapable of responding physically, the real clincher for consciousness is always behavioural; communicative behaviour is especially convincing. The Turing test, it turns out, works for humans as well as robots.

Could there ever be a method by which we measure consciousness directly? Well, if Tononi’s theory of Phi is correct, then the consciousness meter he has proposed would arguably do that. On his view consciousness is generated by integrated information, and we could test how integratedly the brain was performing by measuring the effect of pulses sent through it. Another candidate mught be possible if we are convinced by the EM theories of Johnjoe McFadden; since on his view consciousness is a kind of electromagnetic field, it ought to be possible to detect it directly, although given the small scales involved it might not be easy.

How do we know whether any of these tests is working? As I said, the gold standard is always behavioural: if someone can talk to you, then there’s no longer any reasonable doubt; so if our tests pick out just those people who are able to communicate, we take it that they are working correctly. There is a snag here, though: behavioural tests can only measure one kind of consciousness: roughly what Ned Block called access consciousness, the kind which has to do with making decisions and governing behaviour. But it is widely believed that there is another kind, phenomenal consciousness, actual experience. Some people consider this the more important of the two (others, it must be added, dismiss it as a fantasy). Phenomenal consciousness cannot be measured scientifically, because it has no causal effects; it certainly cannot be measured behaviourally, because as we know from the famous thought-experiment about  philosophical ‘zombies’ who lack it, it has no effect on behaviour.

If someone lost their phenomenal consciousness and became such a zombie, would it matter? On one view their life would no longer be worth living (perhaps it would be a little like having an unconscious version of Cotard’s syndrome), but that would certainly not be their view, because they would express exactly the same view as they would if they still had full consciousness. They would be just as able to sue for their rights as a normal person, and if one asked whether there was still ‘someone in there’ there would be no real reason to doubt it. In the end, although the question is valid, it is a waste of time to worry about it because for all we know anyone could be a zombie anyway, whether they have suffered a period of coma or not.

We don’t need to go so far to have some doubts about tests that rely on communication, though. Is it conceivable that I could remain conscious but lose all my ability to communicate, perhaps even my ability to formulate explicitly articulated thoughts in my own mind?  I can’t see anything absurd about that possibility: indeed it resembles the state I imagine some animals live their whole lives in. The ability to talk is very important, but surely it is not constitutive of my personal existence?

If that’s so then we do have a problem, in principle at least, because if all of our tests are ultimately validated against behavioural criteria, they might be systematically missing conscious states which ought not to be overlooked.


Giulio Tononi’s Phi is an extraordinary book.  It’s heavy, and I mean that literally: presumably because of the high quality glossy paper, it is noticeably weighty in the hand; not one I’d want to hold up before my eyes for long without support, though perhaps my wrists have been weakened by habitual Kindle use.

That glossy paper is there for the vast number of sumptuous pictures with which the book is crammed; mainly great works of art, but also scientific scans and diagrams (towards the end a Pollock-style painting and a Golgi-Cox image of real neurons are amusingly juxtaposed: you really can barely tell which is which). What is going on with all this stuff?

My theory is that the book reflects a taste conditioned by internet use. The culture of the World Wide Web is quotational and referential: it favours links to good stuff and instant impact. In putting together a blog authors tend to gather striking bits and pieces rather the way a bower bird picks up brightly coloured objects to enhance its display ground, without worrying too much about coherence or context. (If we were pessimistic we might take this as a sign that our culture, like classical Greek culture before it, is moving away from the era of original thought into an age of encyclopedists; myself I’m not that gloomy –  I think that however frothy the Internet may get in places it’s all extra and mostly beneficial.) Anyway, that’s a bit what this book is like; a site decorated with tons of ‘good stuff’ trawled up from all over, and in that slightly uncomplimentary sense it’s very modern.

You may have guessed that I’m not sure I like this magpie approach.  The pictures are forced into a new context unrelated to what the original artist had in mind, one in which they jostle for space, and many are edited or changed, sometimes rather crudely (I know: I should talk, when it comes to crude editing of borrowed images – but there we are). The choice of image skews towards serious art (no cartoons here) and towards the erotic, scary, or grotesque. Poor Maddalena Svenuta gets tipped on her back, perhaps to emphasise the sexual overtones of the painting – although they are unignorable enough in the original orientation. This may seem to suggest a certain lack of respect for sources and certainly produces a rather indigestible result; but perhaps we ought to cut Tononi a bit of slack. The overflowing cornucopia of images seems to reflect his honest excitement and enthusiasm: he may, who knows, be pioneering a new form which we need time to get used to; and like an over-stuffed blog, the overwhelming gallimaufry is likely here and there to introduce any reader to new things worth knowing about. Besides the images the text itself is crammed with disguised quotes and allusions.  Prepare to be shocked: there is no index.

I’m late to the party here. Gary Williams has made some sharp-eyed observations on the implicit panpsychism of Tononi’s views;  Scott Bakker rather liked the book and the way some parts of Tononi’s theory chimed with his own Blind Brain theory (more on that another time, btw). Scott, however, raised a ‘quibble’ about sexism: I think he must have in mind this hair-raising sentence in the notes to Chapter 29:

At the end, Emily Dickinson saves the day with one of those pronouncements that show how poets (or women) have deeper intuition of what is real than scientists (or men) ever might: internal difference, where all the meanings are.

Ouch, indeed: but I don’t think this is meant to be Tononi speaking.

The book is arranged to resemble Dante’s Divine Comedy in a loose way: Galileo is presented as the main character, being led through dreamlike but enlightening encounters in three main Parts, which in this case present in turn, more or less, the facts about brain and mind – the evidence, the theory of Phi, and the implications. Galileo has a different guide in each Part: first someone who is more or less Francis Crick, then someone who is more or less Alan Turing, and finally for reasons I couldn’t really fathom, someone who is more or less Charles Darwin (a bit of an English selection, as the notes point out); typically each chapter involves an encounter with some notable personality in the midst of an illuminating experience or experiment; quite often, as Tononi frankly explains, one that probably did not feature in their real lives. Each chapter ends with notes that set out the source of images and quotations and give warnings about any alterations: the notes also criticise the chapter, its presentation, and the attitudes of the personalities involved, often accusing them of arrogance and taking a very negative view of the presumed author’s choices. I presume the note writer is, as it were, a sock puppet, and I suppose this provides an entertaining way for Tononi to voice the reservations he feels about the main text, backing up the dialogues within that text with a kind of one-sided meta-textual critique.

Dialogue is a long-established format for philosophy and has certain definite advantages: in particular it allows an author to set out different cases with full vigour without a lot of circumlocution and potential confusion. I think on the whole it works here, though I must admit some reservation about having Galileo put into the role of the naive explorer. I sort of revere Galileo as a man whose powers of observation and analysis were truly extraordinary, and personally I wouldn’t dare put words into his mouth, let alone thoughts into his head: I’d have preferred someone else: perhaps a fictional Lemuel Gulliver figure. It makes it worse that while other characters have their names lightly disguised (which I take to be in part a graceful acknowledgement that they are really figments of Tononi) Galileo is plain Galileo.

Why has Tononi produced such an unusual book? Isn’t there a danger that this will actually cause his Integrated Information Theory to be taken less seriously in some quarters? I think to Tononi the theory is both exciting and illuminating, with the widest of implications, and that’s what he wants to share with us. At times I’m afraid that enthusiasm and the torrent of one damn thing after another became wearing for me and made the book harder to read: but in general it cannot help but be engaging.

The theory, moreover, has a lot of good qualities. We’ve discussed it before: in essence Tononi suggests that consciousness arises where sufficient information is integrated. Even a small amount may yield a spark of awareness, but the more we integrate, the greater the level of consciousness. Integrated potential is as important as integrated activity: the fact that darkness is not blue and not loud and not sweet-tasting makes it, for us, a far more complex thing that it could ever be to an entity that lacked the capacity for those perceptions.  It’s this role of absent or subliminal qualities that make qualia seem so ineffable.

This makes more sense than some theories I’ve read but for me it’s still somewhat problematic. I’m unclear about the status of the ‘information’ we’re integrating and I don’t really understand what the integration amounts to, either. Tononi starts out with information in the unobjectionable sense defined by Shannon, but he seems to want it to do things that Shannon was clear it couldn’t. He talks about information having meaning when seen from the inside, but what’s this inside and how did it get there? He says that when a lot of information is aggregated it generates new information – hard to resist the idea that in the guise of ‘new information’ he is smuggling in a meaningfulness that Shannonian information simply doesn’t have.  The suggestion that inactive bits of the system may be making important contributions just seems to make it worse. It’s one thing for some neural activity to be subliminal or outside the zone of consciousness: it’s quite different for neurons that don’t fire to be contributing to experience. What’s the functional difference between neurons that don’t fire and those that don’t exist? Is it about the possibility that the existing ones could have fired? I don’t even want to think about the problems that raises.

I don’t like the idea of qualia space, another of Tononi’s concepts, either. As Dennett nearly said, what qualia space? To have an orderly space of this kind you must be able to reduce the phenomenon in question to a set of numerical variables which can be plotted along axes. Nobody can do this with qualia; nobody knows if it is even possible in principle. When Wundt and his successors set out to map the basic units of subjective experience, they failed to reach agreement, as Tononi mentions. As an aspiration qualia space might be reasonable, but you cannot just assume it’s OK, and doing so raises a fear that Tononi has unconsciously slipped from thinking about real qualia to thinking about sense-data or some other tractable proxy. People do that a lot, I’m afraid.

One implication of the theory which I don’t much like is the sliding scale of consciousness it provides. If the level of consciousness relates to the quantity of information integrated, then it is infinitely variable, from the extremely dim awareness of a photodiode up through flatworms to birds, humans and – why not – to hypothetical beings whose consciousness far exceeds our own. Without denying that consciousness can be clear or dim, I prefer to think that in certain important respects there are plateaux: that for moral purposes, in particular, enough is enough. A certain level of consciousness is necessary for the awareness of pain, but being twice as bright doesn’t make my feelings twice as great. I need a certain level of consciousness to be responsible for my own actions, but having a more massive brain doesn’t thereafter make me more responsible. Not, of course, that Tononi is saying that, exactly: but if super-brained aliens land one day and tell us that their massive information capacity means their interests take precedence over ours, I hope Tononi isn’t World President.

All that said, I ought to concede that in broad terms I think it’s quite likely Tononi is right: it probably is the integration of information that gives rise to consciousness. We just need more clarity about how – and about what that actually means.

Picture: meter. It has been reported in various places recently that Giulio Tononi is developing a consciousness meter.  I think this all stems from a New York Times article by the excellent Carl Zimmer where, to be tediously accurate, Tononi said “The theory has to be developed a bit more before I worry about what’s the best consciousness meter you could develop.”   Wired discussed the ethical implications of such a meter, suggesting it could be problematic for those who espouse euthanasia but reject abortion.

I think a casual reader could be forgiven for dismissing this talk of a consciousness meter. Over the last few years there have been regular reports of scientific mind-reading: usually what it amounts to is that the subject has been asked to think of x while undergoing a scan; then having recorded the characteristic pattern of activity the researchers have been able to spot from scans with passable accuracy the cases where the subject is thinking of x rather than y or z.  In all cases the ability to spot thoughts about x are confined to a single individual on a single occasion, with no suggestion that the researchers could identify thoughts of x in anyone else, or even in the same individual a day later. This is still a notable achievement; it resembles (can’t remember who originally said this) working out what’s going on in town by observing the pattern of lights from an orbiting spaceship; but it falls a long way short of mind-reading.

But in Tononi’s case we’re dealing with something far more sophisticated.  We discussed a few months ago Tononi’s Integrated Information Theory (IIT), which holds that consciousness is a graduated phenomenon which corresponds to Phi: the quantity of information integrated. If true, the theory would provide a reasonable basis for assessing levels of consciousness, and might indeed conceivably lead to something that could be called a consciousness meter; although it seems likely that measuring the level of integration of information would provide a good rule-of-thumb measure of consciousness even if in fact that wasn’t what constituted consciousness. There are some reasons to be doubtful about Tononi’s theory: wouldn’t contemplating a very complex object lead to a lot of integration of information? Would that mean you were more conscious? Is someone gazing at the ceiling of the Sistine Chapel necessarily more conscious than someone in a whitewashed cell?

Tononi has in fact gone much further than this: in a paper with David Balduzzi he suggested the notion of qualia space. The idea here is that unique patterns of neuronal activation define unique subjective experiences.  There is some sophisticated maths going on here to define qualia space, far beyond my clear comprehension; yet I feel confident that it’s all misguided.  In the first place, qualia are not patterns of neuronal activation; the word was defined precisely to identify those aspects of experience which are over and above simple physics;  the defining text of Mary the colour scientist is meant to tell us that whatever qualia are, they are not information. You may want to reject that view; you may want to say that in the end qualia are just aspects of neuron firing; but you can’t have that conclusion as an assumption. To take it as such is like writing an alchemical text which begins: “OK, so this lead is gold; now here are some really neat ways to shape it up into ingots”.

And alas, that’s not all. The idea of qualia space, if I’ve understood it correctly, rests on the idea that subjective experience can be reduced to combinations of activation along a number of different axes.  We know that colour can be reduced to the combination of three independent values (though experienced colour is of course a large can of worms which I will not open here) ; maybe experience as a whole just needs more scales of value.  Well, probably not.  Many people have tried to reduce the scope of human thought to an orderly categorisation: encyclopediae;  Dewey’s decimal index; and the international customs tariff to name but three; and it never works without capacious ‘other’ categories.  I mean, read Borges, dude:

I have registered the arbitrarities of Wilkins, of the unknown (or false) Chinese encyclopaedia writer and of the Bibliographic Institute of Brussels; it is clear that there is no classification of the Universe not being arbitrary and full of conjectures. The reason for this is very simple: we do not know what thing the universe is. “The world – David Hume writes – is perhaps the rudimentary sketch of a childish god, who left it half done, ashamed by his deficient work; it is created by a subordinate god, at whom the superior gods laugh; it is the confused production of a decrepit and retiring divinity, who has already died” (‘Dialogues Concerning Natural Religion’, V. 1779). We are allowed to go further; we can suspect that there is no universe in the organic, unifying sense of this ambitious term. If there is a universe, its aim is not conjectured yet; we have not yet conjectured the words, the definitions, the etymologies, the synonyms, from the secret dictionary of God.

The metaphor of ‘x-space’ is only useful where you can guarantee that the interesting features of x are exhausted and exemplified by linear relationships; and that’s not the case with experience.  Think of a large digital TV screen: we can easily define a space of all possible pictures by simply mapping out all possible values of each pixel. Does that exhaust television? Does it even tell us anything useful about the relationship of one picture to another? Does the set of frames from Coronation Street describe an intelligible trajectory through screen space? I may be missing the point, but it seems to me it’s not that simple.


Picture: Phi. I was wondering recently what we could do with all the new computing power which is becoming available.  One answer might be calculating phi, effectively a measure of consciousness, which was very kindly drawn to my attention by Christof Koch. Phi is actually a time- and state-dependent measure of integrated information developed by Giulio Tononi in support of the Integrated Information Theory (IIT) of consciousness which he and Koch have championed.  Some readable expositions of the theory are here and here with the manifesto here and a formal paper presenting phi here. Koch says the theory is the most exciting conceptual development he’s seen in “the inchoate science of consciousness”, and I can certainly see why.

The basic premise of the theory is simply that consciousness is constituted by integrated information. It stems from the phenomenological observations that there are vast numbers of possible conscious states, and that each of them appears to unify or integrate a very large number of items of information. What really lifts the theory above the level of most others in this area is the detailed mathematical under-pinning, which means phi is not a vague concept but a clear and possibly even a practically useful indicator.

One implication of the theory is that consciousness lies on a continuum: rather than being an on-or-off matter, it comes in degrees. The idea that lower levels of consciousness may occur when we are half-awake, or in dogs or other animals, is plausible and appealing. Perhaps a little less intuitive is the implication that there must be in theory be higher states of consciousness than any existing human being could ever have attained. I don’t think this means states of greater intelligence or enlightenment, necessarily; it’s more  a matter of being more awake than awake, an idea which (naturally enough, I suppose) is difficult to get one’s head around, but has a tantalising appeal.

Equally, the theory implies that some minimal level of consciousness goes a long way down to systems with only a small quantity of integrated information. As Koch points out, this looks like a variety of panpsychism or panexperientialism, though I think the most natural interpretation is that real consciousness probably does not extend all that far beyond observably animate entities.

One congenial aspect of the theory for me is that it puts causal relations at the centre of things: while a system with complex causal interactions may generate a high value of phi, a ‘replay’ of its surface dynamics would not. This seems to capture in a clearer form the hand-waving intuitive point I was making recently in discussion of Mark Muhlestein’s ideas.  Unfortunately calculation of Phi for the human brain remains beyond reach at the moment due to the unmanageable levels of complexity involved;  this is disappointing, but in a way it’s only what you would expect. Nevertheless, there is, unusually in this field, some hope of empirical corroboration.

I think I’m convinced that phi measures something interesting and highly relevant to consciousness; perhaps it remains to be finally established that what it measures is consciousness itself, rather than some closely associated phenomenon, some necessary but not sufficient condition. Your view about this, pending further evidence, may be determined by how far you think phenomenal experience can be identified with information. Is consciousness in the end what information – integrated information – just feels like from the inside? Could this be the final answer to the insoluble question of qualia? The idea doesn’t strike me with the ‘aha!’ feeling of the blinding insight, but (and this is pretty good going in this field) it doesn’t seem obviously wrong either.  It seems the right kind of answer, the kind that could be correct.

Could it?