Andy of Niall

On the soul of the machine Posted on 25 February 2008 | Tagged ai, philosophy | Comments (17)

Today marked my return to university for my final semester of undergraduate education. I'm taking three papers – Concurrent & Distributed Systems, Introduction to Artificial Intelligence, and Discrete Mathematics. I had my first lecture for each of these, but the one that has proved most interesting thus far (to be fair, 'thus far' is a single lecture), is Artificial Intelligence.

In particular, we were discussing a debate in the field of AI – strong AI versus weak AI, or in layman's terms, whether or not a machine can possess true intelligence, a mind. For starters, it is useful to establish exactly what constitutes a mind, a conciousness, or the soul. The concept of a soul is something that is universal across all human cultures – the part of a person that makes them distinct and separate, the part that thinks, that feels. The question is then whether the soul is a simply the manifestation of the physical make up of the brain, or something separate and non physical. In both cases this raises the question of how do you model such a thing with a machine, and can it even be modelled? Although it seems much more feasible if the soul is a result of the physical happenings in the brain, there is still little understanding of how the brain actually works. Also, if the brain is merely a machine, albeit a complex one, then how is it we have the feeling of choice – that given a set of inputs, we not only rationalise, but we also act on impulse, on feeling. The concept of free will does not fit well with the idea that the mind is a machine, mapping inputs to outputs.

A slightly less philosophical point (although only slightly less), is what it means to comprehend something. Sure, given the word 'tree', a picture of a tree, or a description of a tree, a machine may be able to correctly associate it with a tree – but does the machine truly have comprehension of what a tree actually is? It understands all of these symbols as pointing to the same object, but does it understand what a tree is? Another way to put this is to think of it in language – given the word 'horrifying', a machine could learn about its definition, where it is appropriate it should be used, when it does or doesn't make sense. But is knowing about 'horrifying' the same as knowing 'horrifying'? Can a machine know what it is to be horrified? John Searle points out this difference in his thought experiment, the Chinese room. Although I will probably fail at trying to explain this, and you are better reading it elsewhere, it basically says the following: given a question in Chinese, and a 'program' of sorts in the form of a book written in English, an English speaker like you or me could follow this program and produce the correct answer in Chinese. The point made here is this – although given the input, an 'intelligent' output is given, the Chinese characters are still just nonsensical squiggles to the uninformed. We have returned the correct output, without actually comprehending what either the input or the output actually mean. It is in this way we have exhibited what seems to be intelligence, but without actual comprehension.

For me, the idea of strong AI is something that is unachievable, but like many unachievable goals there is much to be gained in the process of trying. I'm looking forward to my first chance to dabble with AI in these coming months.