Gerald Edelman

Picture: Edelman
Blandula Gerald Edelman’s theories are rooted in neurology. In fact, he insists that this is the only foundation for a successful theory of consciousness: the answers are not to be found in quantum physics, philosophical speculation, or computer programming.

The structure of the brain is accordingly a key factor. The neurons in the brain wire themselves up in complex and idiosyncratic patterns patterns during growth and then experience: no two people are wired the same way. The neurons do come to compose a number of structures, however. They form groups which tend to fire together, and for Edelman these groups are the basic operating unit of the brain. The other main structures are maps. An uncontroversial example here might be the way some sheets of neurons reproduce the pattern of activity on the retina at the back of the eye (with some stretching and squashing), but Edelman sees similar strucures as applying much more widely, and mapping not just sensory inputs, but each other and other kinds of neuronal activity. The whole system is bound together by re-entrant connections, sets of paths which provide parallel connections from group A to Group B and Group B back to Group A.

The principle which makes this structure work is Neuronal Group Selection, or Neural Darwinism. Some patterns are reinforced by experience, while many others are eliminated in a selective process which resembles evolution. Edelman draws an analogy with the immune system, which produces a huge variety of random antibodies: those which link successfully to a foreign substance reproduce rapidly. This explains how the body can quickly produce antibodies for substances it has never encountered before (and indeed for substances which never existed in the previous history of the planet): and in an analogous way the Theory of Neuronal Group Selection (TNGS) explains how the brain can recognise objects in the world without having a huge inherited catalogue of patterns, and without an homunculus to do the recognising for it.
The re-entrant connections between neuronal groups in different parts of the brain co-ordinate impressions from the different senses to provide a coherent, consistent, continuous experience; but re-entry is also the basic mechanism of recategorisation, the fundamental process by which the brain carves up the world into different things and recognises those it has encountered before. The word recategorisation is potentially confusing here for two reasons: first, it is not to be taken as implying the existence of a prior set of categories: in fact, every act of recognition modifies the category; nor is it meant to suggest any parallel with Kant’s categories, which limit how we can understand the world. Very much the reverse, in fact.
Edelman attaches great importance to higher-order processes – concepts are maps of maps, and arise from the brain’s recategorising its own activity. Concepts by themselves only constitute primary (first-order) consciousness: human consciousness also features secondary consciousness (concepts about concepts), language, and a concept of the self, all built on the foundation of first-order concepts.

The final key idea in the theory, another one with a slightly misleading name isvalue, a word used here to describe inbuilt tendencies towards particular behaviour. These forms of behaviour may be driven by what we value in a fairly straightforward sense – seeking food, for example, but they also include such inherent actions as the hand’s natural tendency to grasp. Edelman seems to think that, like a computer, if left to itself the brain might sit and do nothing. It’s the value systems which supply the basic drives. This sort of set-up has been modelled in a series of robots rather cheekily named Darwin I to IV.
Edelman is emphatically opposed to the idea that the brain is a computer , however.

Bitbucket Being anti-computationalist but using robots to support your theory seems a little strange. It needn’t be strictly contradictory, of course, but it does expose the curious fact that while Edelman insists the brain is not a computer, all the processes he describes seem perfectly capable of computerisation. He gives two reasons for not considering the brain a computer: one, that individual brains are wired up in very different ways; and two, that reality is not an orderly program feeding into the brain. Neither of these is convincing. Computers can differ enormously in physical detail while remaining essentially the same – how much similarity is there between a PC and a model Turing machine, for example – and wiring differences between brains might perhaps count as differences in pre-loaded software rather than anything more fundamental. Certainly reality does not structure itself like a program, but why should it? The analogy is with data, not with the program: you have to think of the brain as a computer which has its software loaded already and is dealing with the data coming down a wire from cameras (eyes), microphones (ears), and so on. I see no problem with that.

Blandula The argument is a bit more specific than you make out. Edelman points out that the selective processes he has in mind have an unusual feature he calls ‘degeneracy’ (I’m not quite sure why). Degeneracy means that the same output can be reached in a whole range of different ways. This is a feature of the immune system as well as mental processes, but it doesn’t look much like an algorithm. Of course there are other arguments against considering the brain a computer, but I think Edelman’s main point is that to deal with reality, you have to be able to arrange the streams of mixed-up and ever-changing data from the senses into coherent objects. Your computer with a camera attached finds this impossible except in cases where the ‘reality’ has been made artificially simple – a ‘toy world’ – and the computer has been set up in advance with lots of information about how to recognise the objects in the ‘toy world’. I know you’re going to tell me that great strides have been made, and that you only need another couple of decades and it’ll all be sorted.

Bitbucket I wasn’t, though it’s true . I was just going to point out again that, however difficult it may be to digest reality, Edelman gives us no definite reasons to think computers couldn’t do it; his robots even demonstrates some aspects of the methods he thinks most likely. But never mind.You expect me to attack Edelman just because he and Searle have spoken favourably of each other: but actually I’ve got nothing much against him except that I think he’s misunderstood the nature of computationalism. Just because we haven’t got USB ports in the back of our heads it doesn’t mean brain activity isn’t computable.

As for that bit about ‘degeneracy’, I don’t see it at all. Imagine we had a job we wanted done by computer – we call in a hundred consultants to tender for the project. They’ll find a hundred different ways to do it. Even if we set aside most of the possible variation – whether to use PCs, Macs, Unix boxes or what, Java, C++, visual Basic or whatever. Even if we assume the required outputs are narrowly defined and all the tenderers have to code in bog-standard C, there’ll be thousands of variations. So I reckon computers can be degenerate too…

Blandula I don’t expect you to attack Edelman at all. As a matter of fact, I’m not an unqualified admirer myself. Take his views on qualia. The temptation for a scientist is always to miss the point about qualia and end up explaining the mechanics of perception instead (a different issue) Edelman, in spite of his scientist bias, is not philosophically naive and a lot of the time he seems to understand the point perfectly. But in ‘A Universe of Consciousness’ he swerves at the last minute and ends up talking about how the neurons could map out a colour space – which might be interesting, but it ain’t qualia. Perhaps his co-author is to blame.

However, I’m with him on the computer issue. Edelman’s views about selection illustrate exactly why computers can’t do what the brain does. I think his ideas on this are really important and have possibly been undersold a bit. The thing about programming a computer to deal with real situations is that you have to anticipate every possible kind of problem it might come up against – but there are an infiinite number of different kinds of problem. Now this is exactly the kind of issue the immune system faced: it had to be ready to deal with any molecule whatever, no matter how novel. The solution is analogous: the immune system fills your body with a really vast number of variant antibodies; your brain is full of an astronomical number of different neuronal patterns. When the problem comes along, even a completely novel one, you’re going to have the correct response waiting somewhere: and the one that matches gets reinforced and reused. Edelman called this a Darwinian process: it isn’t really (hence Crick’s joke about it really being ‘neural Edelmanism’): the remarkable thing is, it might be better than Darwinian in this context!

Bitbucket Anything’s better than Darwin to you, up to and including spontaneous generation and Divine Creation.

Blandula Nonsense! But, honestly. It’s not particularly original to suggest that the mind might use selective or Darwinian mechanisms, (or be infected with memes evolved in the memosphere) but normal Darwinian selection is just obviously not the answer. When we confront a sabre-toothed tiger or think what to say to a question in an interview, we don’t start by copying some earlier response, try it out repeatedly and gradually refine it by random mutation. We don’t even do that in our heads, normally. 99% of the time, the response is instant, and appropriate, with nothing random about it at all. It’s a bit easier to understand how this could be so on the Edelman theory, because some reasonably appropriate responses could already be sloshing around in the brain and the best one could be reinforced very quickly.

Bitbucket I think you’re going further on that than Edelman himself would be inclined to do. In fact, I’ll give you a prediction. Eventually, Edelman himself will come round to the view that there is nothing unique about all these processes, and that while the brain may not be literally a computer, its processes are computable.

Blandula I think not. You ought to remember what the man said himself about changes of heart – the unit of selection in successful theory creation is usually a dead scientist…

7 thoughts on “Gerald Edelman

  1. Hi there,

    What a wonderful website! I’m sure you’re already aware, but I just wanted to hopefully explain the etymology of the term “degeneracy”.

    The term is commonly used in e.g. quantum mechanics to describe energy levels corresponding to multiple different states; that is, in essentially the same way as used above. I suspect the biologists have borrowed this usage.

    Ok, well why do physicists use this word? Well, we took it from the mathematicians. A mathematical object is said to be “degenerate” if it is a special, simpler case of a wider class – the “generic” case. The “degenerate” object typically has special properties that do not hold “generally”, and which it would lose under a tiny perturbation. For example, a point can be viewed as a degenerate circle; one with 0 radius. Points have many special properties that circles do not, and which depend upon their radius being *exactly* zero.

    In QM energy levels are determined by taking the “Hamiltonian” under investigation, which defines the system, and solving what is called the “eigenvalue problem”, which essentially involves solving a particular polynomial equation. If multiple states have the same energy level, then the polynomial equation has multiple identical roots, which is a special, “degenerate”, case.

  2. Of course the brain is a computer. It inputs data, processes the data, stores results and has outputs dependent on the results. That’s pretty much the definition of a computer. It isn’t a digital computer, an analog computer or even a quantum computer, it’s a neural computer. Just because it uses a different technology and because it is orders of magnitude more complex than our simple digital computers doesn’t make it a non-computer.

    We may or may not be able to emulate our brain function on our current or future computers. The power of trying to do it is in the process of describing the algorithm, however complex, to do the segment of brain function under investigation. An algorithm may contain degeneracy if called for. If the algorithm is correctly described it will be a good predictor of the actual brain activity and we can have a reasonable confidence that we understand that part of the process. It’s certainly not the only investigative path, but it seems a reasonable path if that is your interest.

  3. I think when people say a brain is not a computer they mean it’s not a Turing Machine (or arguably a state machine).

    Even on “processing the data” it’s not clear that “information processing” is what a brain does in the sense we think of computers do so – see Peter’s critique of Graziano’s article in the Atlantic.

  4. Well there I go thinking like an engineer again. Philosophers seem to prefer fluid definitions that allow them to argue their case more efficiently. (just joking, kind of)

    Yes, the brain has states, and yes they may not be exactly the same as the way our current digital computers maintain many of their states. But after all, what is a representation but a state of the machine. The brain is full of representations. States are just one aspect of computing though. There are lots of stateless computer functions.

    Peter’s response to Graziano seems to compare the abilities of our simple digital computers to the power of our complex neural computer brains, so of course their capabilities are widely different. The complex machine has the ability to develop consciousness, morals and all the rest humanity has to offer. This is an emotional response to the word “computer” that is usually tied to an adjective like “just” or “merely” rather than appreciating the amazing capabilities of our biological computers. Saying it isn’t a computer is a kind of distraction that allows all sorts of odd theories that attempt to explain consciousness through unlikely and empirically unsupported concepts. I stand by my comments that the brain is a complex neural computer.

  5. “Well there I go thinking like an engineer again. Philosophers seem to prefer fluid definitions that allow them to argue their case more efficiently. (just joking, kind of)”

    AFAIK the argument has always centered around Turing machines? Or, at the least, something in the same “family” of machines?

    Check out Peter’s book, he gets into this.

    “But after all, what is a representation but a state of the machine. The brain is full of representations.”

    I think that’s one of the questions though – can a Turing machine actually hold a representation? For example it seems to me that’s what Jochen’s paper* is about, finding a noncomputational yet still naturalized explanation?

    *Peter covers it here:

    http://www.consciousentities.com/?p=2179

  6. Thanks for the references and comments. I’ve got something to ponder.
    I think what is bothering me about this is the different responses between what a typical modern digital computer can do and what a theoretical Turing machine is capable of. I don’t see any reason a large complex adaptive neural network could not be implemented on a Turing machine, but we clearly can’t currently do that (and likely will never implement it) on a digital computer. The difference between capabilities seems to affect our attitudes and beliefs about the direction of research.

Leave a Reply

Your email address will not be published. Required fields are marked *