By Peter Hankins
The thesis of his massive 2003 work is summed up in the title: Being No One: The Self-Model Theory of Subjectivity. In that book, Metzinger made a commendable effort to balance philosophy and science; but the sheer size of the resulting text may have deterred some readers -- I confess to being somewhat daunted myself. Now he has come back with a slimmer volume The Ego Tunnel which is aimed at a wider public and raises wider issues which Metzinger suggests need public attention.
Metzinger’s theory -- the Self-model Theory of Subjectivity or SMT -- suggests that subjective experience is really a kind of trick the brain plays on itself. Our brain sets up a model of the world (actually based on fairly limited data) to which it then adds a model of us, ourselves. The coherence of the model and the fact that the processes supporting it are transparent -- i.e., invisible to us -- yield the vivid impression of a self in direct contact with reality, and that’s where subjectivity arises; although in fact the whole thing is simply an illusion.
Metzinger’s view of qualia is characteristically complex. He has a good argument against the existence of what he calls canonical qualia, qualia conceived as subjective universals. He points out that our ability to discriminate is far greater than our ability to recognise. So, if we are presented with examples of green 64 and green 66, we can readily tell the difference: but if at a later stage we are presented with one of the examples, we have no hope of telling which it is. So there is no single thing that consistently goes along with the experience of green 64.
Concluding that at any rate we need to distinguish between ‘qualia’ available to memory and qualia available to the faculty of recognition, Metzinger goes on to distinguish a series of possible conceptions of qualia, ending with ‘Metzinger qualia’ which are available attentionally but not cognitively. These are slippery customers for obvious reasons, impossible to report and broadly ineffable -- but then that’s how qualia are generally assumed to be.
Even as a summary, the foregoing is a bare and radically, probably over- simplified view of the theory, however. Metzinger actually presents ten constraints which need to be satisfied for the occurrence of subjective experience: they are:
- Globality; as in ‘global workspace’; conscious items are always integrated into an overall world-model,
- Presentationality; present in the now, temporal immediacy,
- Convolved holism; objects of experience are made up of collections of other objects in a nested hierarchy,
- Dynamicity; the perceived world flows through constant changes,
- Perspectivalness; we experience the world from a point of view,
- Transparency; we cannot see the works -- the neural processing which gives rise to our experience is excluded from conscious experience,
- Offline activation; subjective experience is not confined to the live inputs from our senses, a notable exception being dreams, where a whole non-existent world appears.
- Representation of intensities; besides distinguishing between qualia (whichever version we’re dealing with) we can distinguish their levels of intensity,
- Homogeneity; areas of pink, for example, are made up of smaller areas of pink, not red and white at once, and
- Adaptivity; the features of subjective experience have to be things that could reasonably have appeared in the course of evolution.
You may feel that there’s something a bit odd about this list, especially the appearance of ‘adaptivity’. One does not have to be a creationist to feel uneasy about the idea that that is really an essential feature of conscious experience in any deep sense. In a précis which he prepared for a discussion on the Psyche site -- sadly this no longer appears to be available -- Metzinger discussed only the first six constraints, which suggests that the last four are at least somewhat dispensable. This is a bit confusing -- is homogeneity essential to conscious experience, or was that just a kind of bonus, a description of a property of phenomenal experience which is important but not a defining requirement? I think one issue here may be that Metzinger seems to want to do two jobs at once; he wants to explain subjectivity in a philosophical sense, but he also wants to describe and categorise phenomenal experience in a way which can be related to scientific observation, clarifying the puzzling phenomenology of blindsight and other unusual conditions. These are not incompatible aims, exactly, and there’s no absolute reason why one account couldn’t do both, but there is a tension. A robust philosophical explanation would pare down the account to its essentials, while the description of subjectivity seems to be something we would want to do as fully as possible.
Illusions
Nice review Peter!
When you say "illusion", do you mean he's saying that what we are subjectively aware OF is an illusion or our subjective awareness itself is an illusion? I presume you/Metzinger must mean the former - our mental model of the world is undoubtedly an illusion. A pretty good one too, to the extent that most people don't even realize it and some flatly deny it. As you say, becoming aware of floaters, etc. or even major breakdowns in the model (like a schizophrenic's voices, hemispheric neglect or alien hand syndrome) don't really break our illusion that we're witnessing reality, although since that's all we've ever known and could know, it's not that surprising - just as it never seems to trouble us for a moment that the sky contains a huge glowing orb or that we willingly cease to exist every night as if it were perfectly normal, because it is to us.
The existence of a mental model, and the existence of a model of us inside it, along with the recursion this implies, does seem to be the crucial factor in generating an ego. Recursion is often the key trick that makes emergence happen - like the way a vortex on water wraps around on itself so that it becomes a self-maintaining entity. I hope he's not saying that this generation of a subjective self is an illusion. For a start, who would be there to be deceived? Just because something emerges from a snake eating its own tail, that doesn't mean it's not real - if that were so then matter wouldn't be real either, because matter is just a self-perpetuating distortion in space. If solid matter can just pop into existence then why not subjectivity?
Anyway, I look forward to part two.
The Self
My own view of this is that what we think of as "the self" is a model generated by the brain to explain its previous behavior, and it is no more special than models of many other things in the world. This model is used to predict the consequences of future actions, and this is used to select appropriate behavior, and this behavior becomes part of the past behavior which is analyzed and a model generated to explain it and so on.
In a way, this means I think that the "self" is a model of someone who does not exist; however, once the model exists it gives reality to the self. To look at another way:
You are self prime.
Self prime is a model of a non-existent self which seems to be there due to the history of behavior.
However, given this we may as well label self prime as self.
I wonder if that makes sense?
I agreed with a lot of what Steve Grand said.
Is there really nothing in the Ego Tunnel?
Thanks - perceptive comments as always!
Yes, I think Metzinger's main point is that what we are aware OF is an illusion - he talks about colour not really being there in the world ( I wouldn't go quite that far myself, but you see what he means). He is very emphatic about denying that there is any real self in the Ego Tunnel, however. He concedes, rather grudgingly, that another valid interpretation would be that the process he describes constitutes a self (which seems the more natural conclusion to me). but he clearly finds that an unsatisfactory interpretation. I think he would be nodding enthusiastically to Paul's formulation 'the "self" is a model of someone who does not exist', but he would be frowning over the rest of the sentence. Quite why he is so radically sceptical is an interesting question - more on that in part 2.
Personally, I understand why people want to deny the existence of an immaterial soul, and also be quite clear about there being no homunculus in the brain, but denying the self altogether seems a step too far. After all, here pretty well undeniably is a large animal, and it seems clear that it produces thoughts occasionally. Whatever it is that produces those thoughts - and it may well be very complex, evanescent, and intermittent - I can't see why you wouldn't call that the self.
The Self as a Model
If we are going to deny the self, why not also deny trees and coffee tables? I mean that seriously. What is a tree but a pattern of atoms? If a pattern of atoms interacting in a "selfy" way does not confer selfness on me, why should a pattern of atoms interacting in a "tree-ey" way confer treeness on a tree?
In fact, going more deeply down, why should even atoms be awarded existence?
I would say that emergence answers these questions and we define things as existing if the appropriate pattern exists. I actually think declaring the self not to exist would be being irrational - it would be making a special case of it.
Some further comments which I added in an edit:
I will try to go a bit further here, and look at how the entire thing works.
If we just consider the "mechanics" rather then the philosophy, I very much agree with the idea that the self is just an object in a model. I don't even think there is a special "self-modeling" process. I think that the brain puts a model together that will predict future inputs based on previous inputs and outputs. The self is merely an object in this model, just as trees or cats are. To plan an action, all that the brain has to do is to hypothetically assume it made a particular output, put that into the model, and tyhen get a prediction of the future value of a special input which serves a bit like the evaluation function score in a chess algorithm: Planning is almost non-existent as a discrete, special process.
That brings us to the question of the model. It is clearly hierarchical, because we perceive things in terms of hierarchies of objects, and that is the only way the problem could become tractable. It is clearly probabilistic. So the best way of thinking of the brain as containing this history of inputs and outputs, representing everything that has happened. It looks for simple "objects" - patterns relating these inputs and outputs and builds objects, probabilistically on top. It builds further objects on top of these, and so on. The process is fairly simple. It is merely based on looking for repeating patterns and counting frequencies with which particular arrangements occur, so that partial patterns can be used to generate probabilities of futire objects. However, there is any extra piece of sophistication to this: Meta-patterns are also used. By this, I mean that the system tries to find generalizations of patterns - patterns expressed so generally that they apply in different contexts. For example, a particular pattern might correspond to "circle" and this would be much more general than a pattern relating two kinds of specific objects. It could relate any types of objects, at different levels of the hierarchy. This abstraction in pattern finding is critical to the system working.
All the time, it is running the simple "planning" process I mentioned. Whenever it has a choice of outputting 1 or 0 (simplifying this a lot) it first pretends it output 0 and sees what happened, then it tries pretending it outputted 1 - then it sees what gave the best predictions for its future inputs. This means that when it makes predictions it is taking account of its future behavior, and "the self" is just an element in the model that happens to be there because it relates inputs and outputs. There is nothing special about it.
This brings us now to the issue of memory. I have just described a system that is unwieldy. Each second, thousands of inputs and outputs would be added to the stored information (you can think of it as a series of graphs running back in time). The idea that the brain stores every sensory input or motor output you ever received or made would be absurd: It would be too unwieldy to do anything with that data. Instead, what happens is that the very low level input/output value data is only kept going back a certain amount of time- maybe even just a few seconds or minutes. However, before this data was discarded it was used to construct a second, probabilistic layer of objects. Objects on this level are kept longer before they are thrown out. Before this data is thrown out it is used to construct the third level, which is kept longer and so on. This means that as you go back in time, the data stored gets less detailed - the higher a level in the hierarchy, the longer it persists. This is why we forget things. It is necessary in us and will be necessary in a machine. We don't really "forget" as people think - the lower levels just get taken out so our recall becomes more vague. One thing about this – memory is not quite what people think it is.
It also suggests that the most successful methods of remembering things might be those that create objects higher in the hierarchy - abstractions of the original data.
An added thing that we might imagine is that some of the outputs may feed back into the hierarchy itself, and control its working. For example, when the system sends 1 to a particular output, it may control some aspect of the pattern finding mechanism, or it may control some aspect of the “forgetting” system. If this happens (and I am not sure it does) the system could learn to adjust itself in the same way as it learns to manipulate any aspect of the world.