It begins to look as if the elegant approach to artificial intelligence – using extremely complex software on not very powerful machines – will lose out to the brute force method of testing every conceivable alternative to find a match but doing it extremely fast with specialised circuitry, the Wall Street Journal suggests: that we would reach this point was almost inevitable given that hardware development is progressing so much faster than that of software – the first generation 32-bit iAPX-86 microprocessor, the 80386, is already obsolescent, but the 32-bit version of OS/2 isn’t even available yet – and when it comes to applications like speech recognition, where it is becoming possible to store vast amounts of continuous speech for comparison, the perception of a pretty dumb but very fast machine will be not different from that of a very smart but rather slow one.
