Stephen Lawrence writes,
We've already blown 'way past that with the largest Linux clusters ... but then, it's turned out to take more raw power than anyone expected to do convincing AI. Either that, or we've just been falling down terribly on the algorithm end of things.
Or perhaps, once we get any coherent software package to the level where it can do its own learning, as Cyc is about to do, then we will all be very surprised to see how fast it grows-up on its own.
I don't think the problem is us falling down on the algorithm end, so much as not knowing what the minimum requirements are, and then just letting a minimal machine rip through the internet.
Cyc is way too complex IMHO. In a few years we will probably realize that we could have done it with much less.
Most experts I have talked say that speech recognition has been the hold-up... not so much the actual sound waves-to-words but the parsing and assignment of various levels of meaning and the framing of proper assumptions. This is all falling into place now.
The next few years should be extraordinarily interesting in the AI field... FINALLY.
Jones
