The article by David Salt (THES, April 28) is highly critical of the connectionist approach to artificial intelligence. Some of his criticism of is well aimed, but connectionism is not a monolithic unitary organisation and some of his criticisms address a particular wing of the field rather than the general approach. Others are, in my opinion, just wide of the mark.
Mr Salt argues that "there is no basis for attributing intelligence or consciousness to neural networks", well whatever else the human brain is, it is certainly a network of neurons. This is not, as Mr Salt suggests, a "mystical belief" but an observation, it does not require a "priesthood" to justify it, a microscope will do. The human brain is, at least, a neural network which seems a reasonable basis for contending that a neural network can be intelligent.
From lesion and drug studies we know that some mental states are direct products of the physical and chemical condition of the brain. There may be some states which involve other factors, but there is very little evidence to support such an assumption. We should start with what we know and see how far we can take it.
No artificial neural network comes close to anything like human-level intelligence and I would not expect to see one in the near future.
But then no artificial network is anything like as complex or as large a system as the human brain and much productive work can be done at a much less esoteric level.
Humans evolved from less intelligent animals, and it is as valid for connectionism to try and model the brain of an ant as it is to attempt to model the entirety of human consciousness, perhaps more so.
Some of us would argue that it is preferable to learn to crawl before we attempt to walk. After all, that was how life did it last time.
PATRICK BRADY
Brunel University