Dan: OK Julia, I’m ready for the big IIT blog in this Ontology of Consciousness Series. You are making a case for the ontological fusion of subjective, first-person consciousness and the neurological workings of the brain, that they are the same thing. Neuronal firing, properly defined, IS consciousness, where “properly defined” is what Integrated Information Theory (IIT) provides. And now you are going to walk us through IIT.
Julia: Right, and as I mentioned last time, this journey, like Dante’s, has three parts: 1. The ontological unity you mentioned; 2. An IIT aspect unique to consciousness studies that 3. allows it to be applied for testing and for practical use, at least in principle (but some people contest its practicability, as I will mention).
Dan: How about starting with the first point, the ontological unity of the non-physical phenomena of subjective experience and the physical stuff that makes up the neural substrate of consciousness?
Julia: Right, but first I want to go back to our previous discussion when you said that my argument for the ontological union of subjective consciousness and patterns of neuronal firings, though persuasive, was – and here’s what you said — “not as conclusive as John Stewart Bell’s theorem eliminating hidden variables in quantum mechanics.” My argument was based on eliminating all options except neuronal firings, and you said that, unlike Bell, I – quoting you again – “have not shown with mathematics-like finality that ‘there is nothing else but this firing.’” Now I must confess that you are right. Tononi makes a crucial move that shifts the focus away from neurons firing and puts at center stage a special kind of information – integrated information. The ontological unity, he says, is between consciousness and this special kind of information. In fact, Tononi starts with this assertion. He calls it “the central identity of IIT.” Here is a quote from the 2008 Tononi article “Consciousness as Integrated Information: A Provisional Manifesto”
According to the IIT, consciousness is one and the same thing as integrated information. This identity… has ontological consequences. Consciousness exists beyond any doubt (indeed, it is the only thing whose existence is beyond doubt). If consciousness is integrated information, then integrated information exists. Moreover, according to the IIT, it exists as a fundamental quantity—as fundamental as mass, charge, or energy.
Dan: JULIA!! You have committed the classic crime of bait-and-switch – get us excited about the “the ontological identity of consciousness and neuronal firings” then switch out neuronal firings, which is easy to understand, and substitute “integrated information,” which renders all obscure. This is a Bait-and-Switch felony. I admit that compared to a conceptual structure that is maximally irreducible intrinsically, from the Tononi quote I gave last time, integrated information” sounds comprehensible, but, nonetheless, you are guilty. How do you explain this?
Julia: Ah, Dan, it’s sweet to have so responsive an interlocutor. And you have helpfully highlighted the most important point of this whole discussion. Tononi’s radical move away from neuronal firings to integrated information is what transforms his work from philosophy to science. This is an historic first. Before Tononi, all attempts at a scientific explanation of consciousness started with the obvious neurons-firing frame of mind, like I was espousing. For example, maybe it’s a certain combination of neurons or a certain part of the brain or a certain type of neuron or quantum entanglement happening within nanotubules inside neurons. They all failed in the sense that they did not achieve the level of what Thomas Kuhn in his most-referenced-of-all-academic-books, The Structure of Scientific Revolutions, calls “normal science.” That is, they could not create a paradigm within which novel problems could be formulated and a method provided that guarantees a solution, at least in principle. The search for a physics-based solution to the problem of consciousness became so desperate that the famous philosopher, Daniel Dennett, concluded that consciousness does not exist as a physical thing; and since all things that exist are physical, it must be an illusion.
Dan: Yes, I remember. That was in his book Consciousness Explained, which critics renamed Consciousness Explained Away. What is Tononi’s transforming trick?
Julia: Patience. Before unveiling the secret, I want to respond to the Bait-and-Switch charge. True, I thought in neurons-firing terms until reading Tononi, and for some time after because of the obscurity of the conceptual structure that is maximally irreducible intrinsically. But my main message throughout these discussions has been that there is an ontological unity between the non-physical phenomena of subjective, first-person consciousness and something having to do with the physical brain, and it is better to call this something the easy-to-grasp “neurons firing” rather than “integrated information,” the whole un-obfuscate-able explanation of which would have to come first. My interest is in exploring the world of subjective phenomena, in being able to address such questions as “Can we build conscious robots?” and “Are trees conscious?” with a ‘paradigm’ that would allow me to proceed in the manner of ‘normal science.’ Tononi is the first person to offer the possibility of being able to do so, and he did it by switching from “neurons firing” to “integrated information.”
Dan: I accept your plea of extenuating circumstances. You’re pardoned. Felony charge dropped. Now, please, what is Tononi’s transforming trick?
Julia: It is this: Instead of trying to squeeze consciousness out of neurons firing – the universal-until-Tononi bottom-up approach – Tononi takes consciousness as given and asks what must be the properties of the physical substrate that gives rise to it – a top-down approach. The idea that consciousness is given – what he calls in the earlier quote the only thing whose existence is beyond doubt – comes from Descartes’ 1637 work Discourse on Method. As a philosopher, Dan, you should appreciate this. To answer his question – What must be the properties, etc.? – he follows Descartes’ analytical method and starts with axioms that state the incontrovertible properties of phenomenal consciousness that we experience directly and, hence, are known with cartesian-like certainty. Tononi’s first axiom is the same as Descartes’ – consciousness exists. Axiom 2 says that it has parts – sight, sound, touch etc., and each of these modes has structure. Axiom 2 is kind of obvious. Axiom 3 says that consciousness is informative, where he is using ‘information’ in the technical sense of ‘a reduction in uncertainty.’ Out of all possible subjective experiences that you are capable of having, you are having one particular experience now. That is an enormous reduction in uncertainty – out of a bazillion possibilities, just this particular one. One can in principle use information theory to quantify the amount of reduction, and hence the amount of information or equivalently, in IIT, the amount of consciousness. Tononi labels the informational amount of consciousness by the Greek letter phi. But there is more. Axiom 4 says that this information is integrated, by which he means that the whole conscious experience is perceived as being just one thing, the total experience itself. Even though it has modalities and these have structure, it is one experience. The fifth and last axiom is that consciousness is exclusive.
Dan: Wait a minute! You just said that consciousness is integrated. That means it’s inclusive. How, then, can it also be exclusive?
Julia: Yes, you are not alone in asking this question. Comprehending the exclusion axiom gives people the most trouble. What Tononi should have said is that it is selectively exclusive. To understand how the selection is done takes us into the technical details of the theory, the conceptual structure that is maximally irreducible intrinsically business. Selection is all those things you can throw away before you begin to reduce the quantitative measure of consciousness as integrated information, phi. For example, you can throw away the retina and the optic nerve without reducing phi. This is why, according to IIT, we can dream with full color, motion-picture vision.
Dan: All right, I’ll buy that axiom 5 must be understood retrospectively. I am still waiting to see Tononi’s transforming trick? It is not obvious from this list of axioms.
Julia: Maybe it isn’t obvious, but it is there in axiom 3, the information axiom. Instead of jumping straight to axiom 3, however, let us follow Tononi and proceed systematically starting with axiom 1. To each axiom Tononi gives what he calls a postulate stating what is the property of the physical substrate that corresponds to the axiom. For axiom 1, the existence axiom, you have the corresponding existence postulate. A physical substrate must exist. I know it sounds trivial, but notice, it does not say brains must exist, and for someone like me with an eye on conscious robots and trees, the difference is crucial. To axiom 2, the parts axiom, you have a corresponding parts postulate. The physical substructure must have parts too, but more than that, they must interact with each other in such a way as to maintain a continuous conscious experience.
What I just said about the parts must interact in a special way is key here, so let me illustrate it using neurons. Neurons fire when they are prodded by the firings of other neurons to which they are connected as receivers. Think again of the optic nerve – about a million nerve fibers connect neurons in the retina to neurons in the visual cortex where multiple levels of processing occur. You are tempted to think of this as external light stimulating the brain into action in a one-way neuron-to-neuron chain starting in the eye and ending in conscious vision. But this cannot be right as our dream example shows. Visions in our dreams are fully articulated in terms of shapes, colors, and motions, but the eyes are closed. There is no one-way neuron-to-neuron stream of neuronal firing leading from the eye to vision. For us to be conscious while asleep and dreaming with no external stimulation the brain must stimulate itself. Therefore, the process must be circular – neurons connect to neurons connect to neurons etc., etc. until some of these connections get back to stimulate the ones where we started. The process sustains itself through interacting with itself. Tononi calls this feedback-loops concept “reentrant architecture.” The self-sustaining aspect he refers to as “intrinsic physical cause-effect power.”
Dan: I understand the self-sustaining feedback-loops idea, so let’s move on to axiom 3, the magical information axiom that metamorphoses philosophy into science.
Julia: Don’t be glib; we have reached a point of high seriousness and difficulty. I already gave a clue as to how looking at consciousness as information leads to science because information can be quantified, and, of course, quantification is the essence of science. Tononi uses devices called – here comes the obfuscation – logic gates to illustrate the idea of information in the context of consciousness. To illustrate, consider a device that can receive and send electrical pulses (a neuron stand-in) and that it gets inputs from two other ‘neurons’, either which can be firing or not, and in response to these inputs, itself gives one response, to fire or be silent. If a ‘neuron’ functions like a logic AND gate, it’s output would be to fire if both inputs were firing or if both inputs were silent. In other words, if both the one AND the other have the same value, the AND gate gives a ‘firing’ output. But if one input is on and the other off, it stays silent. You can do the same type of logic math with OR, NOT, and XOR logic gates and build a complex system of such gates connected to each other. Such logic-gate, ersatz-neuronal systems are what Tononi uses to explain his idea. They are the information bearers in his explanatory models. One gets absorbed in the details, but the crucial step by which IIT departs from predecessors is that the quantity of information that comprises consciousness involves time, not just the instantaneous disposition of firing neurons, but the immediately previous disposition of firing and the immediately future disposition of firing. This is crucial. It took me a long time to realize just how crucial.
Take a deep breath and think about this. The neural correlates of consciousness are not just what our neurons are doing now; it is instead an inextricable mix of what they are doing now together with what they could have been doing one time step ago and what they could be doing be doing at the next time step.
Dan: Very interesting. It sounds like Tononi has rediscovered something philosophers have been discussing since St. Augustine, namely, the conscious ‘feel’ of time. For your interest, on the history of the philosophy of temporal consciousness, Barry Dainton at the University of Liverpool has an excellent article in the online Stanford Encyclopedia of Philosophy. Edmund Husserl is primarily responsible for the modern development of time consciousness in the early years of the 20th century. He had a three-part view of consciousness, which he labeled primal impressions, retentions, and protentions. Primal impressions are what we think of as our now. Our primal impression continuously slips into the past, but it is briefly retained in felt consciousness, which is why Husserl calls it retention. I think of retention as our ability to retain the beginning of a spoken sentence until it is finished and understood as a thought. Or as our ability to remember where we are in singing a tune. Husserl’s retentions are different than ordinary memories as you can tell by my examples. They are part of one’s ongoing, felt consciousness. As for protentions, these are our neural anticipations. Our brains continuously predict the next thing that should happen based on retentions and primal impressions. If we had no protentions, we could never be surprised, never feel the satisfaction of a musical phrase ending on the tonic. From what you said, it seems Tononi has a mathematical version of Husserl’s retention and protention.
Julia: Yes, and earlier this year, Marek Pokropski, on the faculty of the Philosophy Department at the University of Warsaw, Poland, also noted the correspondence between Husserl and Tononi in a commentary in Frontiers of Psychology. So, indeed, you philosophers were there first.
But let’s get back to Tononi and IIT. As we said, to find the path from philosophy to science one must discover a way to represent consciousness mathematically. Tononi hit on the idea of using the theory of information as formulated by Claude Shannon in 1948. As things in the world go, information and consciousness are alike in that they seem to be inherently non-mathematical. But as explained by Warren Weaver in his section of the book he coauthored with Claude Shannon in 1949 “The Mathematical Theory of Communication”, there are three levels to the problem of communicating information: the technical problem (how much and quality), the semantic problem (accuracy of meaning), and the effectiveness problem (achieving desired effect). The first of these can be mathematized, and it is the treatment of the technical how-much problem that Tononi uses. Warren explains further:
“the word information in communication theory is not related to what you do say [i.e.,
semantics], but to what you could say [i.e., number of choices]. That is, information
is a measure of one’s freedom of choice when one selects a message.”
Tononi, in his 2008 “Manifesto” article mentioned earlier, says the same thing this way
“Information is classically deﬁned as reduction of uncertainty: the more numerous the
alternatives that are ruled out, the greater the reduction of uncertainty, and thus the
greater the information.”
Note the words “reduction” and “numerous”. Here we have quantification – how numerous are the choices and how big the reduction? The quantity of information is defined as the probability that a choice will not be selected. This might sound paradoxical, but it makes sense. A highly probable choice means it has a low probability of not being selected, hence it carries low information – “We expected it, so the fact that it happened isn’t very informative.” Whereas very unlikely means high information – “Wow! it really happened? This is new.” To quantify information, technocrats use the logarithm to base 2 of the reciprocal of the probability of a choice. Thus, if the probability is ½, the amount of information is 1, and the unit is the bit, 1 bit. If the probability is 1/16, the amount of information is 4 bits.
Dan: Great, thanks for the tutorial. But as a dweller on the humanities, the ‘what matters’ side of campus, I feel unthreatened by quantified information with no message – no meaning. What is the poetry, the music, the art in 4 bits? And sticking to technical issues, I don’t see how Tononi can get probabilities out of neuron firings. If you know which neurons are firing now, then you also know what must have led to the present state of affairs and what happens next. It’s all physics with time-dependent differential equations to tell you what came before now and what will come after now. No probabilities.
Julia: You are touching on the free will problem, which has always been and still is controversial. Fortunately, we need not go there to answer your question: Where do the probabilities come from? In IIT they come from the fact that not all neurons contribute to felt consciousness, just those that comprise what Tononi unhelpfully calls the maximally irreducible conceptual structure, which he abbreviates MICS.I am not going to try to explain what this means. It is enough for you to know that there are probably more non-MICS neurons interacting with the MICS neurons than make up the MICS itself. Therefore, given all the neuronal activity going on outside the MICS and interacting with it, you can believe that you can get to the present neuronal MICS state – that is, felt consciousness – in more ways than one. This is where probability enters. Similarly, for the post-present MICS state; if you can get to the now MICS state from multiple pre-now states involving non-MICS neurons, then each of these options takes you to a different state in the next time step – or something like that.
Dan: Maybe one really needs to understand what maximally irreducible conceptual structure means to give a better explanation. Does Tononi actually connect his logic-element model of the MICS to the neuronal situation in the real brain, that is, to neurons interacting within, without, and between this MICS? That would be helpful.
Julia: No. You must do that yourself. But he is very clear about something else that others in consciousness research find hard to accept. This is his requirement that non-active neurons, despite being inactive, contribute to the phenomenal experience of consciousness. Recall that the quantity of consciousness you have is measured by how many other possible conscious states you could have had. For this to be the case – that the conscious states you could have but are not having nonetheless affect the quality of your conscious experience – means that these states must somehow subconsciously affect consciousness. Tononi calls this effect of non-active states on felt consciousness their ‘potential’ or their ‘disposition’. In his “Manifesto” article he justifies the effect-without-action aspect of IIT as follows:
“While this may sound strange, fundamental quantities associated with physical systems
can also be characterized as dispositions or potentialities, yet have actual effects. For
example, mass can be characterized as a potentiality—say the resistance that a body
would offer to acceleration by a force—yet it exerts undeniably actual effects, such as
actually attracting other masses if these turn out to be there. Similarly, a mechanism’s
potential for integrated information becomes actual by virtue of the fact that the
mechanism is actually in a particular state.”
In a more recent article that he with others published in a July 2016 issue of Nature “Integrated Information Theory: From Consciousness to Its Physical Substrate” he speculates that the property of inactive neurons to affect the quality of conscious experience might account for the contentless, pure consciousness reached by experienced meditators:
“IIT predicts that the cerebral cortex as a whole may support experience even if it is
almost silent, a state which may perhaps be reached through meditative practices
designed to achieve ‘naked awareness’ without content. This contrasts with the common
assumption that neurons only contribute to consciousness if they are active and
‘broadcast’ the information they represent.”
Dan: I’m sure this makes New Age science popularizers like Deepak Chopra happy. I must study all that we have discussed here today for next time. Then I’ll review my effort so that you can judge how well I understand IIT, if at all.