As regards the wager:

First, do you really mean "1990 dollars"? This seems like an unnecessary complication.

Second, Condition 4 (ability to understand *any* English speaker, regardless of accent) is a brave assertion. There are many English-speakers - for example, certain foreigners - who I have a good deal of trouble understanding. Is the assertion really that computers will do better than humans in this regard?

Condition 6 (requests for clarification at most once per 100 sentences) is also dicey. I would frequently fail this test also, even listening to someone who speaks with the same accent as me. During an average day at work I *frequently* have to ask people to repeat things. Am I unusual in this regard?

These conditions suitably corrected, I would be willing to support John's side of the wager (any takers?).

Anecdotal ammunition: At DECworld last month, Digital demonstrated DIDL (pronounced "Didley", for Digital's Integrated Database Laboratory), a small network of computers which interact with the user via voice commands. DIDL complies at least partially with conditions 4 (accents are OK) and 5 (no training required). Condition 2 (price) should resolve itself quite nicely long before the end of the decade.