Joho the Blog
An Entry from the Archives

This is really a very good analogy from another perspective: When human beings are subject to brain trauma that impacts their communications abilities, they strive to rewrite their transmission systems to fit their meaning.

For example, if we change Deep Blue's output layer rules to be a go-moku table and map each response to a go move, suddenly the brilliant thinking machine plays a fools game of go. The human, on the other hand, will recognize that the chess-motivated go moves are not working and would observe the game, learn the new rules and adapt their fundamental strategies from that first principle, the concept of "winning some kind of game".

The human retains their own 'meaning' whereas the machine, as Searle points out, only has the meaning we give it. Now, this is not to say that at some point there could not exist a machine capable of it's own meaning. This is only to say that in the whole opus of all our current AI we haven't a shred of code to even suggest an avenue for further research, ie, there is something fundamentally wrong with our symbolic-manipulation approach.

there is something fundamentally wrong with our symbolic-manipulation approach.

Right, and Weinberger et all thing the problem is with the "symbolic-manipulation" part, and I think it is the approach. Put another way, when/if there is a real strong AI example, it will most likely be a binary digital computer at the heart of it, although the software will be nothing like anything tried to date (ok, maybe some stuff from genetic algorithms, but probably not).

Holy-Synchronicities Batman. I just missed an NPR piece with none other than David Weinberger while I was typing this comment. Just caught the identifications at the end. Not only are on-line sources are recognizing your punditry, but NPR, quite impressive.