991%njit-eies@sri-unix.UUCP (06/14/84)
From: "Martin R. Lyons" <991@njit-eies>
An excerpt from the June 1984 issue of the Communications of the ACM,
"Talking to UNIX in English: An Overview of UC", by R. Wilensky,
Y. Arens, and D. Chin on page 576 contains the following dialogue
between a UNIX user and the AI program, here denoted by NI:
User: I'm trying to get some more disk space.
NI: Type 'rm *'
--------------------------------------------------------------------
MAILNET: Marty@NJIT-EIES.Mailnet
ARPA: @MIT-MULTICS.ARPA:Marty@NJIT-EIES.Mailnet
USPS: Marty Lyons, CCCC @ New Jersey Institute of Technology,
323 High St., Newark, NJ 07102 (201) 596-2929ARENS@USC-ECLC.ARPA (06/16/84)
From: Yigal Arens <ARENS@USC-ECLC.ARPA> In his message about the June CACM article about the Unix Consultant system, Marty Lyons demonstrates that he either misunderstood or simply did not read the article in its entirety. Below are the paragraphs preceding and following the two lines he quoted. I believe this will clarify any misunderstanding. "A useful natural language interface must also incorporate some processes that may be collectively referred to as common-sense reasoning. To demonstrate the importance of such extra-linguistic mechanisms, consider the following hypothetical dialogue with a naive interface (NI). We assume here that NI possesses knowledge about a language's syntactic and semantic conventions, but is not otherwise an intelligent system: User: I'm trying to get some more disk space. NI: Type 'rm *' "NI's suggestion, if executed by the user, would destroy all the user's files. This rather disturbing response might be generated by a naive interface because the response fulfills the user's literal request for more disk space. However, the answer lacks a certain cooperative spirit. A more felicitous answer might be "Delete all the files you don't need" or "Ask the system manager for some more storage." However, in order to prefer these responses over the above, the interface must be able to infer that the user possesses some goals other than the one stated in the request, and that these background goals interact with the request to constrain the beneficial courses of action." Yigal Arens USC -------
BLARSON@USC-ECLB.ARPA (06/16/84)
From: Bob Larson <BLARSON@USC-ECLB.ARPA> Gee... and I thought the only way to get more disk space was to buy another disk drive. (On all systems I know of, deleting files makes more UNUSED disk space, but does not "get more disk space") Sure, asking the system administrator is a valid answer IF the user doing the asking is not the system administrator. This reminds me of convincing tops-10 6.03 that I was using a negative amount of disk storage. Bob Larson <Blarson@Usc-Eclb> -------
gwyn@BRL-VLD.ARPA (06/17/84)
From: Doug Gwyn (VLD/VMB) <gwyn@BRL-VLD.ARPA> The dialogue about "rm *" that you cited was in fact used as an example of problems with naive interfaces of this type. The discussion in the original CACM article was reasonable. Quoting such a thing out of context is not, since it gives the impression that the article's authors were pretty stupid. P.S. I am posting this rather than mailing to the author since the MIT mailer appears to be totally brain-damaged; at least, I can't find a reasonable return Internet address in his message.
chip@t4test.UUCP (06/18/84)
=== REFERENCED ARTICLE =============================================
From: "Martin R. Lyons" <991@njit-eies>
User: I'm trying to get some more disk space.
NI: Type 'rm *'
====================================================================
User: I'm trying to get more disk space.
NI: Type 'rm *'
User: Okay, 'rm *'
NI: Whoa! You don't want to do that, turkey!!
--
Chip Rosenthal, Intel/Santa Clara
{idi|intelca|icalqa|imcgpe|kremvax|qubix|ucscc}!t4test!{chip|news}
Any resemblance between the author and persons
living or dead is entirely coincidental.991%njit-eies@sri-unix.UUCP (06/19/84)
From: "Martin R. Lyons" <991@njit-eies>
It appears that my message of Thursday, 14 June, did not have the impact I had intended.
Let me explain further. We received the CACM article here at NJIT a few days ago. I read with deep interest the article "Talking to UNIX in English: An
Overview of UC" (pp.574-593) since I have been doing some AI work, along the same lines as the article. When I had encountered the few lines between NI
(Naive interface) and User, I thought they were quite funny. After showing them to my collegues there was a general feeling that it was very funny and should
be shared. Thus, the posting to the net. Unfortunately, I failed to make the distinction between NI (Naive Interface) and UC (Unix Consultant) in the
message. This caused a great deal of message traffic over the weekend whilst I was away from my terminal.
To reply to some of the messages I received:
To David Chin and Yigal Arens (authors with Robert Wilensky): My sincerest apologies for the misunderstanding. I should have noted the NI (the worst
case scenario) was exactly opposite of what you are working on. I very much enjoyed the article, it has been one of the best pieces I have read on AI
applications in quite some time. My best wishes for success. I would also like to thank Yigal Arens particularly for posting paragraphs preceeding and
following the statements I extracted, for clarification. Yes, I *did* read the entire article. My sole purpose was to share what I thought a rather funny
man-machine dialogue. It seemed humorous in its location in the article, I am suprised that everyone took my extract as an attack, rather than just enjoying
its humorous content.
To Doug Gwyn: I was not aware that my original message was out of context completely, it was not my intention to make the authors appear 'pretty
stupid'. On the contrary, as I've previously stated, the paper, in my opinion, is an excellent work. I had stated "between a UNIX user and the AI program,
here denoted by NI:...". Anyone who reads the article in its entirety will surely realize that NI is NOT the AI program the authors were working on, nor did
I state that. The 'the' preceeding AI program can be taken to infer that it was the authors program, and in that respect I am admittedly wrong in my wording.
To J. Eric Roskos: I enjoyed reading your comment, but please realize that NI is the Naive User Interface, NOT what the authors intended UC to do.
In conculsion: Again my apologies for any misunderstandings. But please, can't you guys laugh once in awhile? I got the feeling by the tone of some of
your replies that I had proved the entire paper in error. If you read the article (I hope you do) you will see that those few lines are awfully funny, used
in the way the were. It is also an excellent example of why natural language processing is do difficult to deal with -- there are many ways in which humans
state ideas that are not clear.
Please send your opinions, arguments, or other material relevant to this issue to me as private messages, so that we don't clutter up the mailing list.
Thanks...
-- Marty
--------------------------------------------------------------------
MAILNET: Marty@NJIT-EIES.Mailnet
ARPA: @MIT-MULTICS.ARPA:Marty@NJIT-EIES.Mailnet
USPS: Marty Lyons, CCCC @ New Jersey Institute of Technology,
323 High St., Newark, NJ 07102 (201) 596-EIESdave@qtlon.UUCP (Dave Lukes) (06/21/84)
BE FAIR: I've read that paper and I think UC is a turkey too (but for COMPLETELY different reasons). The quote given was TOTALLY out of context.