I have read Hans Moravec's thoughts on the "Humongous Look Up Table"
in the past and so I thought I would step in and offer my interpretation
in the hopes of forestalling misunderstandings.
The HLUT is a way of passing the Turing test which many people consider
unlikely to be conscious. Hans disagrees that the HLUT is any less
conscious than an ordinary Turing test passing program.
The HLUT consists of a table which has a response to all possible
conversations. In database terms it is indexed by a conversation-so-far,
and its value is the response which would be given at that point in the
conversation.
For example, one entry in the table consists of:
Index:
"Hello, my name is Martha."
"Hello, Martha, I am wondering if you are conscious."
"Well, I wonder that myself sometimes!"
"So, are you?"
Response:
"As far as I can tell, I am!"
Every possible conversation is included in the HLUT, which is what makes
it "humongous" (slang for very large). Once the HLUT is constructed, it
can pass the Turing test. Every time you ask it a question, it gets
added to the record of the conversation-so-far. For its response, it
simply looks through the table to find a matching index value for the
current conversation-so-far, then issues the corresponding response.
Since all possible conversations-so-far are included in the table,
it can always come up with a response.
One way to construct the HLUT is by constructing a conscious program and
then querying it with all possible conversations. This would take a
"humongous" amount of time, but that's OK. In practice you can restrict
yourself to conversations which last no more than 100 years, so the size
of the table is finite.
It seems clear that you need a conscious program (or perhaps person) to
construct the HLUT, but the question arises as to whether the HLUT is
conscious after it has been created, at the time it is used to pass the
Turing test. The workings of the HLUT program are incredibly simple,
just appending questions and comments to a memory buffer, and looking
through the table to find a match. It is hard to understand how such a
simple program could be conscious.
Hans Moravec argues that the HLUT program is as conscious as the original
computer program which was used to create it. In fact, he suggests that
the two programs are isomorphic, in the sense that there is a mapping
from one to the other. Obviously there is a mapping from the original
program to the HLUT, since that is how I proposed to create the HLUT.
Hans suggests that there is a mapping in the other direction as well,
that is that given the HLUT, you could reconstruct the original program
which created it. There is so much information in the HLUT, it is so
humongous, that it should be possible to deduce all the details of the
conscious program which could have created it. Hence there are mappings
in both directions, what is called in mathematics an isomorphism.
Despite the fact that both programs look very different, Hans suggests
that they are, in essence, the same. One has a complex program and
simple data structures, the other has a complex data structure and a
simple program. But the essence of the calculation done by each is the
same, as demonstrated by the existence of the isomorphism.
My view is that it is possible that the isomorphism exists, but I am
not convinced that it is guaranteed to exist. Much information is not
recorded in the HLUT - emotional states, alternate answers which were
considered and then rejected, etc. People have been known to keep
secrets their entire lives. Is it guaranteed that every private thought
of the original conscious program can be deduced by looking at its
responses to all possible conversations? Maybe there are some programs
so closemouthed that no conversation could cause them to reveal their
secrets. In that case I don't see how any amount of study of the HLUT
could reveal the full structure of the original program.
Hal
Received on Tue Jul 27 1999 - 18:46:25 PDT
This archive was generated by hypermail 2.3.0
: Fri Feb 16 2018 - 13:20:06 PST