PHIL/AI: Humongous Lookup Table

Hal Finney (hal@rain.org)
Mon, 10 Feb 1997 16:39:48 -0800


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.

I had a new (to me) idea about this old puzzle.

For people who have not run into it, the humongous (slang for huge) lookup
table is a device which should be able to pass the Turing test, while being
simple enough in design that it is hard to understand how it could be
conscious. It is often presented as a challenge to the validity of the
Turing test.

As I understand the HLT concept, it would be a table which is indexed by
"the conversation so far" and produces "the next response". You could
create an HLT by running all possible conversations through a conscious
computer program and recording its output for each conversation. (We will
ignore the practical impossibility of running so many simulations.)
The resulting table would be truly humongous, since the number of possible
conversations is astronomically large.

Once the table is created, though, it is easy to use, a simple matter of a
single lookup per response. You can sit down and converse with the table,
and have an intelligent conversation with this seemingly mindless device.

My idea is to suggest that you're not "really" conversing with the table.
I have an electronic device on my desk, smaller than a shoebox, with which
I can have an intelligent conversation. It converses in natural English
and easily passes the Turing test. It is my telephone.

The telephone isn't a challenge to the Turing test because we know that
we're not really talking with the telephone. We're talking with the
person on the other end of the line. The telephone is just a conduit,
a channel, to the person we're really talking to.

I think that the HLT can be thought of in the same way. It is a conduit
to the person or program which was used to create the HLT. When we talk
to the HLT, we're really talking to that program.

This is a little different from the telephone because the HLT is in
effect a channel into the past. The reactions which the program had to
our conversation, and which were recorded as its reactions in the HLT,
have already occured when we talk. Still, the mental picture I form of
the person I am talking to is real. It doesn't exist now, but it did
exist in the past, just in the form I am imagining it. So I think this
justifies thinking of the HLT as being just a channel between that mind
and my own.

In this light, the HLT is not really a challenge to the Turing test because
we're not really talking to the HLT at all. The HLT attracts our attention
so that we forget about the program which was used to create it. That is
where the mind actually is.

Hal