Sveinbjorn Thordarson's Website

The human brain is said to be the most complex object known to man. Be that as it may, a satisfactory account of how the brain actually works has consistently eluded scientists. The creation of a device which emulates the brain seems to be a promising path towards greater understanding in this field, or at least an avenue worth exploring. I shall therefore be concerned with two questions in the course of this paper: Is emulation of the human brain a reasonable possibility (i.e. is it possible in principle?), and if so, what light would a successful emulation shed on the philosophy of mind?

There are several conceptions of how emulation of the brain is to be realized. David Marr (1982)¹ presents a view which is now orthodox. Emulation is to be achieved through the implementation of algorithms which reflect the brain's functionality. Marr distinguishes between "three levels" in consideration of which a model of the brain's functionality is to be constructed: computational theory, representation and algorithm, and hardware implementation. It is a marked feature of the strong artifical intelligence (AI) proponents that they believe that the hardware used to execute the algorithm in question is merely a practical matter of implementation -- it is the algorithm itself which is a manifestation of "intelligence", or the specific function of the brain that we are trying to emulate. Implicit in this view is the idea that if we truly understand what is going on in the brain, we should be able to construct computer programs which do the same thing.

Another conception, which has proven more popular in recent years due to positive empirical results, is the idea of self-modifying, learning computer programs. It is argued that these programs, given sufficiently sophisticated and extensive initial learning algorithms (or, in some cases, genetically selective algorithms), and fed with vast amounts of appropriate input, could develop over time to exhibit intelligence. This approach, however, has hitherto not proven helpful in shedding light on the functionality of the human brain, and is hardly likely to do so in the future.

In the realm of pure speculation, we have ideas such as deciphering the blueprint of the brain from human DNA or emulating the brain spatially on an atomic level (which is in all likelihood impossible due to Heisenberg's Uncertainty Principle). These ideas remain the subject of science fiction and deserve scant attention for now.

While it is all very well to speculate concerning the practical methods by which an emulator of the human brain could be created, it seems more important to ask whether such a device could even be created in principle -- whether there are genuine logical and/or philosophical reasons to dismiss the very notion of such emulation.

In The Emperor's New Mind (1989)², and to a greater extent in Shadows of the Mind (1994)³, Roger Penrose launches a controversial, multi-pronged attack on strong artificial intelligence and thereby on the possibility of emulating the brain. Central to his attack is the claim that the human brain operates non-algorithmically. He backs this claim with an argument based on Kurt Gödels famous Incompleteness Theorem from 1931, which states that in a given formal mathematical system which is powerful enough to describe integer arithmetic, there will always be a certain correctly formed statement of the system which implies its consistency but cannot be proved within the system. As a consequence of this, Gödel maintained that mathematics has an independent reality which transcends any logical system -- we humans have a sort of 'intuition' which allows us to see the truth of this correctly formed statement. Roger Penrose takes this a step further, and argues that Gödel's conclusion has profound implications for the science and philosophy of mind. If a computer is merely a logical system operating on a specific piece of hardware, then our minds have a capacity to comprehend truths which will always surpass the ability of computers -- we have "external insights". The argument, in much abbreviated form, is the following:

A mathematician can make judgements about the truth of mathematical statements. If the functionality of a mathematician's brain is essentially no different from that of a computer, then in principle, one could write an extremely complicated computer program which duplicates his behavior. However, any program that infers mathematical statements can only infer that which can be proved within a formal system of mathematical axioms and rules of inference. Gödel's theorem shows that there is at least one true statement which such an axiomatic system cannot prove. However, mathematicians can 'see' the truth of this statement. Therefore, no program we can know to be axiomatically sound can be emulated by a computer.

This argument has some problems. To begin with, why must a computer's formal system be axiomatically sound? The human brain is fallible (despite Penrose's claims about the extreme accuracy of human mathematicians), and thus any device which emulates the brain must also implement this fallibility somehow. Also, does this argument work if the axioms in the formal system of which the mathematician's mind is comprised are too complex for a human to comprehend? In that case, we cannot know it to be axiomatically sound.

As Hilary Putnam⁴ has pointed out, Penrose makes the mistake of assuming that in order to create a device which successfully emulates the brain, we must "know" its programming, i.e. it must be "simple enough to appreciate in a perfectly conscious way". I too regard that line of thought as entirely fallacious. There is really no good reason to suppose this to be the case. Even today there are computer programs consisting of hundreds of thousands (or even millions) of lines of code, far beyond the "perfectly conscious" appreciation of any single human being. There seems to be nothing inherently illogical about the idea of an arbitrarily large number of different people contributing to the creation of a strong artificial intelligence or brain emulator, each of them working on a seperate, specialized component, none of them capable of completely appreciating of the workings of the whole. While the implementation of a successful human brain emulator might very well require millions or even billions of lines of extremely complicated code, and thus prove to be impossible in practice, this point is completely irrelevant to the question at hand -- that is, whether such an emulation is possible in principle.

Penrose puts forth another argument against the "algorithmic brain" based on natural selection. He find it implausible that an algorithm sufficiently sophisticated to encompass human thought could have developed by a process of natural selection. He says:

"Any selection process of this kind could act only on the ouput of the algorithms and not directly on the ideas underlying the actions of the algorithms. This is [...] totally unworkable. [....] the slightest 'mutation' of an algorithm would tend to render it totally useless, and it is hard to see how actual improvements in algorithms could ever arise in this random way."⁵

Penrose could not, of course, be expected to anticipate the explosion in self-modifying genetic algorithms in recent years, or their remarkable success in solving a variety of problems. While a reasonably simple algorithm may very well be rendered useless by minor mutations, this is not neccesarily so in more advanced cases. As an algorithm grows to enormous complexity, minor changes may tend to have less effect on the overall output. Apart from all this, I consider it arrogant to overlook the possibilities involved in literally hundreds of millions of years of evolution, when the capacity for development is so great in much shorter time spans.

If we accept Penrose's arguments and concede that the human brain is not algorithmic and not a formal system, we seem to be left with a difficult question. How does it work? Penrose (unlike a fellow Oxonian AI-skeptic John Lucas⁶, who inspired his Gödelian argument) is hesitant to draw any mystical conclusions. Instead he speculates that there is a close and possibly unknowable relation between the laws of quantum mechanics and human thought processes. Along with many other prominent physicists, he believes that quantum theory is incomplete, and suggests that this gap in physics might be related to shortcomings in the science and philosophy of mind. If the brain exploits unknown quantum physical effects, a machine which exploits these effects is required in order to successfully emulate it -- thus, a satisfactory emulation of the brain will not be possible until we discover the unknown quantum principles.

Penrose himself admits that this idea is highly speculative, for there is precious little evidence indicating that any physical operations of the brain take place on a quantum level. I gather that his ideas on this subject have received little credence amongst the scientific community⁷. While this is not a satisfactory reason to dismiss them out of hand, it seems to me that claims about the brain operating to some extent on unknown quantum principles are really just another (perhaps more credible) "cop-out" to the problem. Instead of attributing the purportedly non-algorithmic brain with mystical qualities such as a "soul from God" to explain the problem away, Penrose chooses to link it with unproven conjectures in physics -- conjectures that have as of yet no more evidence to support them than speculations or dogmas about a "soul" or a "divine spirit".

Putting aside the aforementioned arguments against strong AI, what would a successful emulation of the brain entail? Apart from its obvious relevance to cognitive science, it raises and provides answers to some profound philosophical questions. Assuming that the emulated brain passes a rigorous Turing test, i.e. exhibits interactive human behaviour to such a degree of accuracy that it is indistinguishable from actual human beings, the views of functionalists would seem to be strongly confirmed.

The question of whether an emulated brain would have 'consciousness' (a subject on which Penrose dwells at great length) immediately suggests itself, but it is in my opinion a philosophical red herring. If the emulated brain is capable of exhibiting all the signs commonly agreed to be indicators of intelligence, it seems pointless to discuss whether it is 'conscious'. The external behavioural evidence (which happens to be the only evidence available to us in this matter) would be the same as that from which we conclude that human beings other than ourselves are conscious. If it is reasonable to assume that other human beings are 'conscious', then consciousness must surely be attributed to phenomena which exhibit the same external signs of intelligence. Admittedly, an emulated brain would not possess a human body, or consist of the same form of physical matter as humans, but this does not detract greatly from my argument. To elaborate, let us imagine that an alien lifeform whose physical manifestation happened to be, say, crystalline, were to land on Earth and somehow communicate to us their extensive knowledge of mathematics, physics, etc. It would be absurd to write off these aliens as 'unconscious' based on the fact that their physical form is different from ours -- unless, of course, we choose to arbitrarily restrict our notion of 'consciousness' to anthropoid lifeforms.

If the orthodox view is correct, and an actual understanding of how the brain works is neccesary in order to successfully emulate it, then a successful emulation would not in and as of itself provide any further understanding. It would merely confirm the materialist view that human beings are machines, albeit machines of an extremely complex sort. As a consequence, this would bring about the long awaited death knell of dualism. Since those who constructed the emulator only assembled and programmed a physical device, it would be hard to maintain that mind and body are two seperate substances. Of course, it would always be possible for dualists to maintain that the "mind substance" mysteriously came into existence as soon as the device was activated, or that behavioural evidence was not sufficient to establish whether thought actually takes place. This is, of course, not only highly speculative and completely unverifiable, but also has a superstitious air to it. I regard it as extremely implausible.

We must, however, consider the possibility that even if the brain were to be successfully emulated on a physical level, it would fail to meet human degrees of ability. Functionalists would no doubt argue that in such a case the brain had not actually been successfully emulated. But if we could know beyond any doubt that the emulator in question was physically accurate, and it still failed to meet human standards, it would almost certainly lead us to the conclusion that the mind is more than machinery. In such a case, we would have to give up a philosophy of ontological monism and conclude that there are elements to human thought which lie beyond the realm of possible experience.

In conclusion, while a successful emulation of the brain may seem an almost impossibly complicated (or, at best, an extremely remote) achievement, Roger Penrose's arguments against this being feasible in principle are not sufficiently convincing to dissuade me from believing it possible. Furthermore, while we could not possibly know whether such an emulator was 'conscious', it would provide powerful, but not conclusive, evidence for the truth of the functionalist philosophy.

Reykjavík, March 24th 2005

Sveinbjorn Thordarson

1. Marr, David, "Philosophy and the approach" Ch. 1 of "Vision" (1982)

2. Penrose, Roger, "The Emperor's New Mind: Concerning Computers, Minds and The Laws of Physics", Oxford University Press, 1989, Oxford, Great Britain.

3. Penrose, Roger, "Shadows of the Mind: A Search for the Missing Science of Consciousness", Vintage, 1995, London, Great Britain.

4. Putnam, Hilary "Review of 'Shadows of the Mind' by Roger Penrose", Bulletin of the American Mathematical Society vol. 32 nr. 3, July 1995, p. 370-373.

5. See "The Emperor's New Mind" p. 415

6. Lucas, J.R. "Minds, Machines and Gödel", available at the following Uniform Resource Locator: http://users.ox.ac.uk/~jrlucas/mmg.html.

7. Source was private conversation with Thordur Jonsson, Professor of Theoretical Physics at the University of Iceland.

Hanson, Robin, "Has Penrose Disproved A.I.?" 1991, first published in Foresight issue 12, p. 4-5. Available at the following Uniform Resource Locator: http://hanson.gmu.edu/penrose.html

McDermott, Drew, "Penrose is wrong", Psyche: an interdisciplinary journal on research of consciousness, 2(17), October, 1995. Available at the following Uniform Resource Locator: http://psyche.cs.monash.edu.au/v2/psyche-2-17-mcdermott.html

Download PDF Version