> From: Eliezer Yudkowsky <sentience@pobox.com>
> > [...]
> > I can PROVE that a giant
> > hashtable-thermostat can maximize anything a computational mind can.
> > I.e. REALLY BIIG (but less than 3^^^3) lookup table, duplicates inputs
> > and outputs, no mind, but works as well.
The universal conversational hash table is like the infinite library (a
library with all possible books). If it could exist, it would be a weird
thing, but it's just not possible. There are too many phrases to encode. Do
you realize how long it would take to program 3^^^3 intelligent
conversational responses?
> Given present search engines on the net, it should be rather a simple
>step to add in a contextual/grammatical comparator. Using a database the
>size of the net, maybe even just recording all email or usenet traffic
>for a few months to create the database, such an engine could enter in
>text read from a person, look up contextually and grammatically relevant
>material, as well as with a thesaurus to vary the use of words to
>prevent redundancy, and pass as an AI rather handily.
This is the most plausible way to create a hash table conversationalist and I
believe similar methods are used to create certain computerized support
systems. However, it fails the Turing test quickly as once you say a
sentence nobody has spoken before, it will fail. It's very easy to devise a
sentence with a vanishing probability of having been spoken before and
trivial to devise such a conversation. Almost any innovative analogy will
do.