home *** CD-ROM | disk | FTP | other *** search
- Xref: sparky comp.ai:4684 comp.ai.philosophy:7139
- Path: sparky!uunet!dtix!darwin.sura.net!udel!louie!mercury.cis.udel.edu!hughes
- From: hughes@mercury.cis.udel.edu (John Hughes)
- Newsgroups: comp.ai,comp.ai.philosophy
- Subject: Re: discussion with Penrose
- Message-ID: <1992Dec22.221613.3198@udel.edu>
- Date: 22 Dec 92 22:16:13 GMT
- References: <1992Dec22.021109.4401@oracorp.com>
- Sender: usenet@udel.edu (USENET News Service)
- Organization: University of Delaware, Newark
- Lines: 46
- Nntp-Posting-Host: mercury.cis.udel.edu
-
- In article <1992Dec22.021109.4401@oracorp.com> daryl@oracorp.com (Daryl McCullough) writes:
- >The problem with the Turing Test is that there is no precise
- >definition of what it means to pass it. However, if one gives a time
- >limit for how long the test may go on (say, somewhere between 15
- >minutes and 100 years), and one limits the test to purely textual
- >exchanges between the judge and the contestant, then it follows that
- >whatever the test involves, there exist Turing Machines that can pass
- >it.
- >
- >This is a somewhat controversial point. However, it seems to me that
- >it has to be true. There are only finitely many sequences of questions
- >that could possibly be asked within the time limit, so all it would
- >take in principle to pass it would be a finite lookup table that gave
- >an appropriate answer for each question. If you want to include such
- >questions as "What time is it?", or "Was it raining yesterday?", then
- >you would have to have inputs corresponding to the current time, the
- >weather report, etc. As long as the total information needed to answer
- >each question is finite, then a lookup table could in principle answer
- >it.
-
- This is not so. Though in a finite amount of time there are indeed only a
- finite number of questions I can ask, those questions can be chosen from
- an infinite set. Your program would have to be able to anticipate any
- legal utterance from the language (and probably several illegal ones), which
- is infinitely composable ("Hey computer, do you know Mary? Bill and Joe?
- Bob and Ted and Carol and Alice?" etc.). Your program will have to be able
- to have a response for an unbounded set of utterances while it has only
- finite resources. Your table will have to work by decomposing the utterance
- into pieces, and figuring out how the pieces fit together and why. That is,
- it will have to understand language. It is not known that there is a Turing
- Machine which can do that.
-
- If what you mean is that there will be exchanges where a properly programmed
- TM passes the test by pure luck (i.e. because it just happened to be
- programmed with snappy answers to the exact questions the interrogator asked),
- you are quite right; there are many Turing Machines (in fact, many finite state
- machines) which will pass many timed Turing Tests. However, I don't think any
- of us can be very happy with that result. If you artificially bound your
- domain, there are TMs (and FSAs) that can do *anything*. They can't learn new
- words the interrogator idiosyncratically uses or respond to misconceptions
- or even be used in another Turing Test, however, so they would basically be
- worthless.
-
-
- John
- hughes@cis.udel.edu
-