The Shadows of the Mind
by Roger Penrose
I have concentrated in above discussion, on the quality of 'understanding' as something essential that is missing from any purely computational system. That particular quality was after all,what featured in the Gödel argument of 2.5- and whose absence,within the mindlessness of computational action, revealed the essential limitations of computation,thus spurring us on to try to find something better.Yet understanding is but one of the qualities for which conscious awareness is of value to us. More generally, we conscious beings benefit from any circumstance where we can directly 'feel' things; and this,I am arguing, is just what a purely computational system can never achieve.
We may well ask : in what way is a computer-controlled robot disadvantaged by its inability to feel,so that it could not appreciate, say,the beauty of a starlit sky,or the magnificent splendour of the Taj Mahal in a still evening, or the magical complexities of a Bach fugue-or even the stark beauty of the Pythagorean theorem? We could simply say that it is the robot's loss that it cannot feel what we are capable of feeling when confronted by such manifestations of quality. Yet there is more to the matter than this. We might ask a different question. Accepting that the robot is not actually capable of feeling anything, might not a cleverly programmed computer be nevertheless capable of producing great works of art?
This is a delicate question, it seems to me. The short answer, I believe, is simply "no" -if only because the computer cannot possess the sensual qualities that are necessary in order to judge the good from the bad,or the superb from the merely competent. But, we may ask : why is it necessary for the computer actually to 'feel', in order for it to develop its own 'aesthetic criteria' and to form its own judgements? One might imagine that such judgements could simply "emerge" after a long period of (bottom-up) training.However, as with the quality of understanding, I feel that it is much more probable that the criteria would have to be part of the computer's deliberate input, these criteria having been carefully distilled from a detailed top down analysis (very possibly computer aided) that has been carried out by aesthetically sensitive human beings.Indeed, schemes of this very kind have been put into action by a number of AI researchers. For example, Christopher Longuet - Higgins,in work performed at the University of Sussex, has implemented various computer systems that compose music according to criteria that he has provided. Even in the eighteenth century, Mozart and his contemporaries showed how to construct 'musical dice' that could be used to combine known aesthetically pleasing ingredients with random elements in order to produce vaguely creditable compositions. Similar devices have been adopted in the visual arts , such as the , AARON system programmed by Harold Cohen, which can produce numerous "original" line drawings by invoking random elements to combine fixed-input ingredients according to certain rules.(See Margaret Boden's (1990) book The Creative Computer for many examples of this kind of 'computer creativity'; also Michie and Johnston (1984).)
I think that it would be generally accepted that the product of this sort of activity has not, as yet, been anything that could stand comparison with what can be achieved by a moderately competent human artist.I feel that it would not be inappropriate to say that what is lacking, when the computer's input reaches any significant level,is any "soul" in the resulting work! That is to say, the work expresses nothing because the computer itself feels nothing.
Of course, from time to time, such a randomly generated computer - produced work might,simply by chance,have genuine artistic merit.(This is related to the old matter of generating the play Hamlet by typing letters entirely at random.) Indeed, it must be admitted, in this context, that Nature herself is capable of producing many works of art by random means,as in the beauty of rock formations or the stars in the sky.But without the ability to feel that beauty, there is no means of distinguishing what is beautiful from what is ugly. It is in this selection process that an entirely computational system would show its fundamental limitations.
Again,one could envisage that computational criteria could be fed in to the computer by a human being, and these might work fairly well,so long as it is just a matter of generating large numbers of examples of the same sort of thing (as one might imagine could be done with run-of the-mill popular art)-until the products of this activity become boring, and something new is needed.At that point, some genuine aesthetic judgements would be required,in order to perceive which 'new idea' has artistic merit and which has not.
Thus in addition to the quality of understanding,there are other qualities that will always be lacking in any entirely computational system such as aesthetic qualities. To these must be added, it seems to,other kinds of things that require our awareness, such as moral judgements.We have seen in Part 1 that the judgement of what is or is not true cannot be reduced to pure computation. The same (perhaps more obviously) applies to the beautiful or to the good. These are matters which require awareness and are thus inaccessible to entirely computer - controlled robots. There must always be a continuing controlling input from a sensitive, outside, conscious- presumably human-presence.
Irrespective of their non-computational nature we may ask : are the qualities of 'beauty' and 'goodness' absolute ones, [Ref: R.Pirsig "Lila"] in the Platonic sense in which the term 'absolute' is applied to truth-especially to mathematical truth? Plato himself argued in favour of such a standpoint. Might it be the case that our awareness is somehow able to make contact with such absolutes, and it is this that gives consciousness its essential strength? Perhaps there might be some clue, here, as to what our consciousness actually 'is' and what it is 'for' .Does awareness play some kind of role as a 'bridge' to a world of Platonic absolutes?
These issues will be touched upon again in the final section of this book.
The question of the absolute nature of morality is relevant to the legal issues of 1.1 1.There is relevance,also,to the question of "freewill",as was raised at the end of 1.11 : might there be something that is beyond our inheritance,beyond environmental factors, and beyond chance influences-a separate "self" that has a profound role in controlling our actions? I believe we are very far from an answer to this question. As far as the arguments of this book go,all that I could claim with any confidence would be that whatever is indeed involved must lie in principle beyond the capabilities of those devices that we presently call 'computers'.