Neuroscience and the Soul
Edmund Furse
Dear Sir

John Cornell (Letters, 27 November) makes a number of points in opposing the Strong Artificial Intelligence (AI) Hypothesis that one day machines will be as intelligent as humans. I would like to briefly respond to his points.

First, Cornell rightly says that there have been many predictions about AI that have subsequently had to be revised. One of the main outcomes of AI research has been to show that tasks such as understanding human language which were thought to be relatively straightforward are now seen to be very complex. Similarly there was hype about expert systems in the 1980s suggesting they could solve any problem, researchers now realise that not all human expertise can easily be represented in the form of IF-THEN rules.

Now, although most people working in AI believe the strong AI hypothesis, few would predict when they expect machines to be praying and discussing politics, etc. Estimates differ, from my own 30 years, to over 500 years. I may well be mistaken and over optimistic with my estimate of 30 years, but it needs to be remembered that the strong AI hypothesis does not state a time limit; it is merely saying that, in principle, it is possible for a machine some time in the future to do all the intelligent activities that humans currently perform; and I argue that could also include religious thoughts and actions.

Second, Cornell cites Penrose's book The Emperor's New Mind, and states that he has not seen a successful repudiation of his arguments against strong AI. Penrose has been refuted in Aaron Sloman's review of his book in the journal Artificial Intelligence 56 (1992), and also in over 30 articles in the journalBehavioural and Brain Sciences 13,4 (1990).

Third, Cornell argues that consciousness and other human qualities are 'intrinsic qualities in human beings as "living", evolved, creatures'. The simple response is to point to research in the allied field of Artificial Life where organisms have been built in silicon which move, sense, and reproduce. Certainly, current research in Artificial Life has not resulted in machines which can understand human language but has shown that the notion of living beings is not intrinsically confined to biological mechanisms.

Over the last five years there has been increasing research in AI into the subject of agents, and it is clear that from an evolutionary perspective there were autonomous agents on the planet long before there was intelligence. It may be the case that for a machine to be as intelligent as humans, it too will have to be a full autonomous agent capable of exploring the world and interacting with it. But in principle, if this were the case, it does not seem to be an insurmountable obstacle.

Fourth, Cornell argues that an essential quality of being human is that we are individual and unique, but that machines do not have this variation. This is a partly a question of scale and partly a question of the level of modelling. Many of the proteins and other building blocks in humans are the same. Likewise modern computers all rely upon instructions in some programming language. However, it is perfectly possible to design computer programs which as a result of their interaction with the environment, result in each one being different.

Fifth, Cornell argues rightly that our emotions are an essential component of being human. It may be the case that for a machine to be an intelligent entity that it too will have to have some form of emotions. Some people working in AI believe this to be the case, for example Aaron Sloman has a group at Birmingham University with quite a complex computational model of emotion.

Sixth, Cornell argues that strong AI "trivialises the character of our ordinary, human, mental life, our beliefs, desires, emotions, intentions, consciousness and subjectivity". In contrast I would argue that every one of these notions he lists are the subject matter of hundreds of AI papers, and that AI research has enabled us to come to a more precise understanding of what these notions mean.

Seventh, Cornell appears to argue that even if we could create such an intelligent machine we would not be able to communicate with it. It may be the case that, as Fredkin argues the intelligent machines of the future would not be able to communicate their own higher thoughts with us, but there seems to be no reason in principle why they could not discuss matters of interest to us. We assume in science fiction that we would be able to communicate with aliens, and in theology with angels, so why not also with machines?

In conclusion, I believe it is possible to refute all of Cornell's points, even if this brief letter does not do the arguments justice. The subject of strong AI is a well known topic in philosophy, but it is less well known in theology, neuroscience and physics. I believe that AI has an important contribution to make to theology. Just as AI has helped philosophers and psychologists to formulate their inquiries more precisely, so also I believe this to be the case in theology. For example, what exactly do we mean when we talk of humans being made in the image of God, or what is the nature of moral reasoning and how can it be learned?

Yours sincerely

Dr Edmund Furse
Department of Computer Studies
University of Glamorgan

Furse E., (1993), "Neuroscience and the Soul", Letter to the Tablet, 4 December, 1993, page 1591.

List of Papers