COMMENT ⊗   VALID 00017 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00003 00002	∂03-Nov-81  1520	Tom Dietterich <CSD.DIETTERICH at SU-SCORE> 	Meeting to talk about theseS  
C00006 00003	∂27 Nov 1981 0014-PST	Russell Greiner <CSD.GREINER>	Come and get it!
C00009 00004	∂TO STT 14:05 1-Dec
C00010 00005	∂TO BROWN@PARC (CC LENAT@PARC)  17:45 4-Dec
C00012 00006	∂TO RHAYES-ROTH@SRI-KL 18:23 11-Dec
C00015 00007	∂TO CSD.BENNETT@SCORE 18:13 18-Dec  [earlier to CSD.BACH & KEDES]
C00023 00008	∂TO CSD.KEDES@SCORE (CC CSD.BRUTLAG, CSD.BACH)  13:35 21-Dec
C00028 00009		Abarbanel
C00038 00010	∂TO WALDINGER@SRI-AI 16:29 30-Dec-81
C00043 00011	∂TO clancey@sumex (CC csd.clancey@score) 11:33 19-Jan-82
C00048 00012	∂TO stefik@parc 11:49 20-Jan-82
C00054 00013	∂23-Apr-82  0932	TW   
C00062 00014	∂TO darden@sumex, csd.dietterich@score, tom@kestrel, csd.gardner@score,
C00064 00015	∂25-Apr-82  1231	TW   
C00065 00016	∂TO sklein@isib 15:13 3-May
C00066 00017	∂04-May-82  2002	ML  	analogy   
C00071 ENDMK
C⊗;
∂03-Nov-81  1520	Tom Dietterich <CSD.DIETTERICH at SU-SCORE> 	Meeting to talk about theses  
To: rdg at SU-AI
cc: CSD.DIETTERICH at SU-SCORE

Russ,

I read your thesis stuff and it's clear that there is a lot of
potential for overlap with what I'm doing.  We really need to talk
about this so that we can cooperate rather than competing.  When can
you get together and talk?  How about Wednesday or Friday morning?

--Tom
-------

∂03-Nov-81  2124	Paul Cohen <CSD.PCOHEN at SU-SCORE> 	Regular meetings  
To: csd.dietterich at SU-SCORE, rdg at SU-AI, csd.lenat at SU-SCORE

Hi Russ and Tom, I met with Doug today and mentioned that we had been 
chatting rather a lot lately, and Doug suggested regular meetings between
the four of us. I stressed that we are all in the development stage.
If you want to do this, maybe you could send me times when you are
UNABLE to meet. Thanks, --P
-------

∂TO CSD.DIETTERICH@SCORE 22:23 16-Nov
Theses
Tom -
	Can we get together sometime in the near future to discuss theses?
My proposal has undergone some modification in style (in particular it
now works through an example), which may render it more readable.

	And yours?
Russ

∂17-Nov-81  1156	Tom Dietterich <CSD.DIETTERICH at SU-SCORE> 	Re: Theses     
To: RDG at SU-AI
cc: CSD.DIETTERICH at SU-SCORE

Russ,

Sure, let's get together.  I have made zero progress on my stuff since
I last talked to you.  I'll be glad to read your new stuff, though,
and comment on it.  Send me hardcopy (preferred) or a pointer.

--Tom
-------

11AM Friday
∂27 Nov 1981 0014-PST	Russell Greiner <CSD.GREINER>	Come and get it!
To: CSD.GENESERETH
cc: CSD.GREINER

The thesis proposal is awaiting you in the G box of Dover output stack.
Read, and remember -- "the quality of mercy is never strained".
(Just in case, I'm buying stock in the "red ink" market.)

A not-so-pretty-but-still-readable version is <CSD.GREINER>THESIS.DOC.

Yawn,
	Russ
-------

∂29 Nov 1981 2006-PST	Mike Genesereth <CSD.GENESERETH>	Re: Come and get it!
To: CSD.GREINER

Having taken a quick look at yoiur refernces and not finding the
metaphors and models paper listed there, I can only conclude that
the proposal isn;t too interesting.  However, I'm willing to take a look
at it.
mrg
-------

∂TO CSD.GENESERETH@SCORE 12:49 30-Nov
Yea, but count those fonts!
Huh?  If you check the bibliographies associated with books I referenced
you'll find several occurances of (AND Models Metaphors).
(You do believe in indirect pointers, don't you?)
I just didn't want to be too obvious a party to this clear example of 
plagarization -- see the 1962 book by Black.
Besides, I thought it was generally conceeded that that articular was devoid
of semantic content!  (See referee's comment.)
Finally, I, like the rest of us, tend to forget where specific
ideas may have come from.  (Your ref has been included...)

Russ

∂TO CSD.GENESERETH@SCORE 17:50 4-Dec
Now with NO phosphates!
Mike -
	I made some changes to the proposal, mostly cosmetic -- 
things like padding out the bibliography.
If you have yet to begin reading it, feel free to dover yourself a copy of the 
new improved version -- see THESIS.PRE[1,rdg].
	Russ
∂TO STT 14:05 1-Dec
This and that
Steve -
	Greetings!  How was the Channel 5 series, by the way?
Did it present anything of use?

	Have you had a chance to peruse my proposal?  Any comments?
Russ
∂TO BROWN@PARC (CC LENAT@PARC)  17:45 4-Dec
Thesis Work
John -
	I'm just now beginning serious thesis work on the topic of analogy;
and would greatly appreciate the benefit of your insights and experiences.
Would it be possible to get together in the near future, to discuss and
hopefully improve some of my ideas?  
A PRESS version of my current thesis proposal is on <GREINER>THESIS.PRE;
feel free to examine it.  (Don't be alarmed by its bulk -- much of
it is "appendix"-ish.  Also, not all of it is in the form of a partial statement,
followed by some needless punctuation and another partial clause; much of
it reads much better than this message.)

Context:  We've met a few times before, both over at Parc and
here at Stanford, after a few of your lectures.
My previous work has been on RLL, with Doug and Mike.

Thanks,
	Russ

∂TO BROWN (CC KEANE) @PACR  11:47 17-Dec
Thesis proposal
The current version of my thesis proposal is now stored on (in?, at?)
the press file [Parc]<Greiner>Thesis.Pre;1.  
Its protection is now 777777, which should permit you to DOVER it.

Feel free to read whatever subset of it seems relevant -- much of it
is ancillary.

I look forward to discussing it, whenever you've time.
	Russ

∂Called 15:40 22-Jan-82
∂TO RHAYES-ROTH@SRI-KL 18:23 11-Dec
Greetings
Rick -
	Welcome!  I just learned you too couldn't resist beautiful Palo Alto.
What sort of schedule do you have now?  Could we get together and chat
some time in the not-too-distant future?
	Context: I finally assembled that thesis proposal I've been
mentioned for the last several months.  As promised, it's on the topic
of analogy.  Do you time/interest/desire to read it, and provide
comments/improvements/...?
	Thanks,
Russ

∂14-Dec-81  0910	RHAYES-ROTH at SRI-KL 	Re: Greetings    
To: RDG at SU-AI

Russ:
	It's true, I'm moving back to the country.  Ah goats! Ahh pigs!

Ellie Engelmore at Tek keeps my schedule which is generally weds & thursdays
in palo alto, except the two holiday weeks in december when it'll be tues &
weds.  Why don't you drop a copy of the proposal of at Teknowledge and
then I'll set up ameeting to discuss it with you.

Cheers,

	Rick
-------

∂TO RHAYES-ROTH at SRI-KL 15:34 14-Dec
En route
Rick -
	"back" to the country?  Anyway, I sent a copy of that proposal with
a courier (Denny).  It is rather long -- and some parts will be yet even
less relevant than others.  Do feel free to read only certain sections, if you
wish.

Tally-ho,
	Russ

∂Called TeKnowledge 15:44 22-Jan-82
∂TO CSD.BENNETT@SCORE 18:13 18-Dec  [earlier to CSD.BACH & KEDES]
 ∂TO CSD.BACH@SCORE 17:47 17-Dec
 ∂TO KEDES@SUMEX (CC BGB) 14:45 18-Dec
Potential Molgen/Thesis Interaction
Larry -
	I'm currently considering using MOLGEN as an application domain
for my thesis work.  
In a nutshell, I'm exploring various ways of using analogy to enter new data
into an existing KB.  I've some preliminary ideas, and want to see how 
often this type of mechanism would really be used.
Let me give an example of the sort of thing this program should be able to do.
[Note - the chemistry is probably wrong.  Ignore that, if possible.]

Dendral currently has a bunch of rules which deal with certain chemical
configurations, including, (for the sake of argument,) hydrates.
[R-C(OH)2-H].
Now someone comes along, and wants to analyze ammoniates [R-C(NH2)2-H].
The challenge is how to incorporate this into Dendral's Knowledge Base.

How would you describe it to me?  Probably by telling me that such these ammoniate
groups are just like hydrates, except ... where that exception clause
is very similar to the way hydrozyl compounds differ from amine compounds, in
general.

From this information I could deduce many of the properties of these ammoniates.
in particular, I (might) be able to generate the rules Dendral should use
for analyzing these compounds, but taking the rules for hydrate and twiddling
them.

Reiterating, we could pose this inquiry as:
for each rule for hydrate, R, find the rule R' which satisfies the analogy
relation
	Hydrate : R :: Ammoniate : R'.

(This R -> R' transformation may be fairly straightforward, if we
make the assumption that R' will probably differ from R in much
the same way a rule for, say, hydrates differs from one for ammoniates.)

End of strained example; now for my question:
How often does this sort of thing happen?  That is, how often are you
updating Molgen's KB with such new cases?  Can you think of examples of
hand?

New I have some fairly rigid requirements for these new cases, X:
Each must be a domain concept for which
(1) There is some domain concept Y s.t.
  (i)  X is like Y
  (ii) much is known about Y
(2) Little is known about X, except
(3) How X is {like	    } Y is known.
 	     {different from}

----
Of the top of his head, Bruce suggested that Molgen would be in this situation
if some new class of (non-restrictive) enzyme were found; and had to be added.

I also spoke briefly with Rene today.  He suggested that geneticists use this
type of reasoning more when dealing with ideas and plans,
than with "domain" level objects (like nucleotide sequences or enzymes).  
The example he mentioned involved taking the plans developed for modifying a
vector, and using them, mutatis mutandis, to twiddle lambda-phages.

Comments?  Could we perhaps meet sometime soon to pursue this idea.
Thanks,
	Russ

∂17-Dec-81  1803	Rene Bach <CSD.BACH at SU-SCORE> 	Re: Potential Molgen/Thesis Interaction       

Yes,
	Lets talk tomorrow. Of the top of my head I would say that we have not 
yet run into that problem. The analogy stuff is more used to plan new experiments
than to update the KB. If you think of problems more than of technical informations,
then it may be discussed. You may also want to talk to LArry and Peter.
See you tomorrow.
Rene
-------

∂18-Dec-81  1606	Buchanan at SUMEX-AIM 	Re: Potential Molgen/Thesis Interaction        

I happen to know that Larry is out of town over the holidays.  Probably
will be back in January, so don't be disappointed about a lack of response.

good thoughts.
bgb
-------

∂18-Dec-81  1755	Larry Kedes <CSD.KEDES at SU-SCORE> 	Re: Potential Molgen/Thesis Interaction    
To: RDG at SU-AI
cc: brutlag at SU-SCORE

Russ,
your approach seems interesting.  I would say that molecular biologists
and geneticists do use that form of resoning to analyze by analogy
quite often.  Isolating a new compound (e.g. gene or enzyme or phage or
bacterium etc.) and knowing how it differs from a known compound is
common.  It at least is the analogy that helps in experiment planning:
the hypothesis is that for the new compound X will exhibit trait y since
compound Y (the known) exhibited trait Y.

One problem I forsee is that KB building has gotten short shrift by
the molecular biologists in MOLGEN lately in terms of creating
new information about "compounds".

Isn't the strictest type of structural analogy the simple concept in
a DNA structure that if we clone one gene and work out its restriction map,
a second, unknown, gene will have a restriction map too?  Or is this
too elementary a reduction of your idea.

I forwarded a copy of your msg to Doug Brutlag who may respond as well.

-Larry

∂TO CSD.KEDES@SCORE (CC CSD.BRUTLAG, CSD.BACH)  13:35 21-Dec
Analogy is a many splendered thing
Thank you for so quick a response.  Yes, analogy does occur at many different
levels, and is present in numerous situations.
Your first example (of experiment design) is right along the lines
I'm beginning to examine; and I would profit immensely from a 
few particular examples of when this was (or could have been) applied.
(That is, can you relay an experiment whose design was based on an earlier,
similar experiment, mutatis mutandis?  Are there many such instances?)

I don't think I quite understood your next example: it sounded like you were
inferring that one gene will have a restriction map because another gene
has such a map.  A more interesting result would involve finding that a pair
of (perhaps somehow related) genes were (unexpectedly) found to have
similar (or even identical) respective restriction sites
-- indicative of similar overall nucleotide sequence, and hence perhaps
pointing to a similar function, on either a DNA-structure level, or on the
level of the derived protein.

In the case of cloning (if I understand that concept)
the actual cloned sequence would be identical to the original.  
Finding that one clone is indeed identical to another
is indeed an instance of analogy; but the fact that they were designed to
correspond exactly makes this analogy seem rather degenerate, and uninteresting.

Or is this view of cloning totally inaccurate (ie I assumed the "replication"
process was fault-free)?
Are there systematic "errors", producing non-homologous regions?
Were you proposing that these systematic errors could be discovered by
some analogizer?
Or that these errors could be anticipated in a new cloning instance,
based on the errors found in an earlier experiment?

The latter proposal is reasonable -- a case of finding the ? which satisfies
	Orig1 : Clone1 :: Orig2 : ?
where Orig1 is the original sequence, Clone1 the derived cloned sequence, and
Orig2 is the new sequence.
(Perhaps after many of these analyses the first question
(of discovering these errors) could be achieved as well
-- by noting the common properties of the mis-transcriptions.
As you can tell by now, I'm still at that grandious stage of thesis construction;
and am considering all sorts of far-fetched ideas...)

Anyway, I would enjoy a higher-bandwidth discussion.  
If you are interested as well, when can we get together?
I will be in the area from now until 1-Jan, and then after 6-Jan.
(By the way, Bruce implied you currently were on vacation.  
Was he totally mistaken?)

Russ
	Abarbanel
∂22-Dec-81  0910	Rene Bach <CSD.BACH at SU-SCORE> 	Re: Analogy is a many splendered thing   

Russ, the address is ABARBANEL @ sumex, for the protein stuff.
Rene
-------

∂TO ABARBANEL@SUMEX (CC CSD.BACH at SU-SCORE)  11:42 22-Dec
Uses of an Analogizing Program
Dr Abarbanel,
	I'm just starting serious thesis work, and am looking for
an application domain.
The research question is how to use analogy to enter new data
into an existing KB.  I've some preliminary ideas, and want to see how 
often this type of mechanism would really be used.
Rene Bach suggested I look into the protein stuff you're doing; hence
this message.

Let me give an example of the sort of thing this program should be able to do.
[Note - the chemistry is probably wrong.  Ignore that, if possible.]

Dendral currently has a bunch of rules which deal with certain chemical
configurations, including, (for the sake of argument,) hydrates.
[R-C(OH)2-H].
Now someone comes along, and wants to analyze ammoniates [R-C(NH2)2-H].
The challenge is how to incorporate this into Dendral's Knowledge Base.

How would you describe it to me?  Probably by telling me that such these ammoniate
groups are just like hydrates, except ... where that exception clause
is very similar to the way hydrozyl compounds differ from amine compounds, in
general.

From this information I could deduce many of the properties of these ammoniates.
in particular, I (might) be able to generate the rules Dendral should use
for analyzing these compounds, but taking the rules for hydrate and twiddling
them.

Reiterating, we could pose this inquiry as:
for each rule for hydrate, R, find the rule R' which satisfies the analogy
relation
	Hydrate : R :: Ammoniate : R'.

(This R -> R' transformation may be fairly straightforward, if we
make the assumption that R' will probably differ from R in much
the same way a rule for, say, hydrates differs from one for ammoniates.)

End of strained example; now for my question:
How often does this sort of thing happen?  That is, how often do you imagine
a KB will be updated with such new cases?  Can you think of any real crisp
examples of this, off hand?

I do have some fairly rigid requirements for these new cases, X:
Each X must be a domain concept for which
(1) There is some domain concept Y s.t.
  (i)  X is like Y
  (ii) much is known about Y
(2) Little is known about X (at least in the KB), except
(3) How X is {like	    } Y is known.
 	     {different from}

----
Off the top of his head, Bruce Buchanan suggested that Molgen would be in
this situation if some new class of (non-restrictive) enzyme were found;
and had to be added.

I also spoke briefly with Rene Bach the other day.
He suggested that geneticists use this type of reasoning more when dealing 
with ideas and plans, over applications at the "domain" level objects 
(like nucleotide sequences or enzymes).  
The example he mentioned involved taking the plans developed for modifying a
vector, and using them, mutatis mutandis, to twiddle lambda-phages.

Larry Kedes had a few vague examples, involving both experiment planning
and the results of a cloning experiment.

Let me know if you'd like to discuss this.  Thanks,
	Russ

∂22-Dec-81  1215	Rene Bach <CSD.BACH at SU-SCORE> 	Re: Uses of an Analogizing Program       
To: RDG at SU-AI, abarbanel at SUMEX-AIM

Bob,
	One thing Russ didn't mention is the protein structure stuff, wHere, I 
think, a lot of "facts" are deduced from comparing structures. It is along that
line that I suggeted him to talk to you. But thi should not restrain you from
any other suggestions. One thing which is also limiting is that Russ needs
a fairly large database. But about more constraints, you'll have to talk with Russ.

Rene
-------

∂05-Jan-82  2150	Abarbanel at SUMEX-AIM 	analogy    
To:   RDG at SU-AI

Be happy to discuss things... don't see any "crisp" examples in my work
but perhaps you will when I describe the problems/solutions.
I'm around PA most Thursday's except only for short time on the 7th.
Call me at home 415-8490921 if you'd like... or can we plan on something
like 1 pm on the 14th of Jan ?
bob abarbanel
-------

∂TO Abarbanel at SUMEX-AIM  11:33 9-Jan-82 
Meeting
Yes, let's meet this coming Thursday, the 14th.
My only constraint is a still underspecified rehearsal, beginning at
either 2 or 4 PM (I'll not know which until Tuesday).
Meeting at 1PM is OK by me, unless you think there'll be more than one hour
of stuff to discuss.  Alternatively, meeting at noon for lunch would leave open
the possibility of two hours of interaction
-- as well as give me an excuse to eat something reasonable.

Which sounds better for you?  Or is some earlier/later time better?

Russ

∂12-Jan-82  2227	Abarbanel at SUMEX-AIM 	Re: Meeting     

I have a meeting re protein structure at Noon  (M361 Med School) then
we can meet ? 1:30 at CS desk on MJH 2nd floor...
I'm 5'9" bearded with briefcase that day I expect.  Short talk may be
good then re-schedule for future interest, OK ? bob
-------

∂TO abarbanel@sumex 11:37 13-Jan
Logistics
Yes, let's meet tomorrow.  Arrangements:
(i) I could meet you outside M361 -- and we could yak en route back to MJH,
	or wherever.  [Indicate when your meeting ends.]
(ii) Alternatively I could wait patiently in my office - MJH 251. 
	(Across from the conference room 252.)
(iii) ____________ (write in)

Please vote for one of the above, supplying the additional information as needed.

[Stats: I'm beardless, non-mustashed, tall (6'ish) and lanky (?scrawny?), and
perennially en-backpacked.]

Russ

∂TO abarbanel@sumex 15:50 2-Feb-82
Oops...
Bob -
	Looks like I'll be the one who has to cancel out this time:
various HPPers are speaking at the computer forum at 11AM this Thursday.
Perhaps at noon?  (I do, however, want to attend the afternoon session,
beginning at 2PM.)
	Or would you prefer some other time?
Russ
∂TO WALDINGER@SRI-AI 16:29 30-Dec-81
Refs to things like analogy  (?analogy analogues?)
Richard:
	Thank you for mentioning those analogy-related references the other
day.  Nachum's thesis does seem, at first glance, potentially quite relevant --
it has been placed high on my to-read-soon stack (queue?).
I couldn't find you other suggestion -- a paper by Moll & Uhrlich.
Am I mispelling it, or is it just obscure, or ...  Any leads would
be much appreciated.
	Thanks,
Russ

∂01-Jan-82  1733	Waldinger at SRI-AI 	Re: Refs to things like analogy  (?analogy analogues?)     

moll and ulrich (robert and john, respectively)  there is a paper in
the acm conference on ai/programming languages in rochester, august,
1975_(?  same year as boston ijcai)
there may be a later one in the japan ijcai.

moll is visiting stanford startif;ing in january,  not doing 
analogue stuff anymore.   

there is some good french work on deriving programs from examples
kodratoff guiho jouannaud.  they get very hard programs but their
methods are algorithmic and perhaps limited

cheers
happy new year
richard
-------

∂TO WALDINGER@SRI-AI 12:05 9-Jan-82
Re: Analogy
Thanks for your quick response.  I was able to find both M&U refs, and glanced
through both.  Their approach to analogy is considerably  different from mine
in significant ways -- it, for example, includes information from the user
in the loop; and they are only dealing with a single type of analogy 
(i.e. first generating, and then extending, a mapping between programs
based on found corresponding aspects).
Nevertheless, I may still be able to use some of their ideas, and perhaps
their examples.  Thanks again for the refs.

The part of your message on Moll's upcoming visit was a bit garbled.  
Who (or what sub-department) will he be visiting?

You mentioned a French crew, (?kodratoff guiho jouannaud?,) who derive programs
from examples.  How are they incorporating analogy into this process?
Of course, sending a pointer to their work is an amply sufficient answer.

Thanks again,
cHag Samay-acH,
Russ

∂TO Waldinger at SRI-AI 15:01 17-Jan-82
Follow up questions
∂11-Jan-82  1225	Waldinger at SRI-AI 	Re: Analogy   

moll is visiting someone in linguistics i forget who.  he is not
working on analogy any more (unless it's in a linguistic context).

i was viewing any derivation of a program from examples as analogical
reasoning.  i think this is fair.  their work is in the bonas precedings
(i told you? the are nato workshops and such)

best regards
richard
-------

[RDG] I truddled over to the linguistics department, but cound find no one
who knew anything about Moll.  Any other leads?  (You did mean Stanford's
lingustics department, yes?)

I will seek out that "bonas precedings" stuff.  (No, you hadn't mentioned
it before -- thanks for this reference.  
While it's not clear how directly relevant it will be for me, it couldn't
hurt to know of this work... Besides there are a few other people here
(notably Tom Dietterich) whose work may-be/is quite related,
to whom I'll forward this pointer.

Thanks again for your continuing help,
	Russ
∂TO clancey@sumex (CC csd.clancey@score) 11:33 19-Jan-82
Uses of an Analogizing Program
Bill,
	I'm just starting serious thesis work, and am looking for
an application domain.
The research question is how to use analogy to enter new data
into an existing KB.  I've some preliminary ideas, and want to bounce them
off various knowledgable people.  Are you game?

	If so, when would be a good time to get together?
My schedule is quite open -- lots of free time chunks.

	By the way, I've concocted a contrived example to illustrate the
sort of thing I'm considering.  Let me know if I should forward it to you...

Thanks,
	Russ

[By the way, what is your preferred mailing address?]

∂Called TeKnowledge 15:44 22-Jan-82

∂26-Jan-82  0820	Clancey at SUMEX-AIM 	Re: Uses of an Analogizing Program    

Russ,
	I just got back from a trip.   I'm very interested in your
topic, so let's get together.  I'll send a message suggesting a time
after I've scanned all of my back messages.  
	Also, please send other info (maybe that's in your next msg?).
Everything gets forwarded to CLANCEY@sumex, so use that address.

	Bill
-------

∂26-Jan-82  0946	Clancey at SUMEX-AIM 	Re: Uses of an Analogizing Program    

Russ,
	Please suggest a time when you'd like to meet.  I am
setting aside most of this week for meetings, and it will be easiest
if everyone else picks a time.

	Bill
-------

∂26-Jan-82  2243	Clancey at SUMEX-AIM 	Re: Uses of an Analogizing Program    

How about Friday morning before Siglunch, say 10:30?

	Bill
-------

∂TO clancey@sumex-aim 15:36 27-Jan
Contrived Scenario
Bill -

Below is a sketchy example of the type of scenario I think 
my eventual program should be able to perform.

-----
Consider the task of inputting rules to Dendral.
There will certainly be many times when the rules used for
one class of chemicals were essentially identical to those for another class,
but not quite.
To illustrate, imagine you had just enterred the dozen rules specific to hydrates.
Now comes the time to enter the corresponding rules to be used for
ammoniates.  (To a first approximation,) these new rules should be deducable
from that first set -- making only the changes needed to distinguish
amine compounds from hydrozyl compounds.  
That is, each new ammoniate rule, R, should be analogous to a hydrate rule, R'.

It would be nice if a program could generate this R using a relation like
	hydrates : ammoniates :: R' : ?.
(This relation, in turn, would be based on a body of facts known about
hydroxyl and  amine compounds).
However, lacking such an analogizing routine, you the user would be forced
to tediously enter a new rule R for each such R'.
-----

Anyway, I'm currently searching for a crisp example of when this actually happened,
to replace this imprecise contrived one.  Can you recall any such instance?
(when "educating" an expert system, when man or machine...)

Thanks,
	Russ
∂TO stefik@parc 11:49 20-Jan-82
Uses of an Analogizing Program
Mark,
	I'm just starting serious thesis work, and am looking for
an application domain.
The research question is how to use analogy to enter new data
into an existing KB.  I've some preliminary ideas, and want to bounce them
off various knowledgable people.  Are you game?

	If so, when would be a good time to get together?
My schedule is quite open -- lots of free time chunks.

	By the way, I've concocted a contrived example to illustrate the
sort of thing I'm considering.  Let me know if I should forward it to you...

Thanks,
	Russ

∂20-Jan-82  1613	Stefik at PARC-MAXC 	Re: Uses of an Analogizing Program
To: Russell Greiner <RDG at SU-AI>
cc: Bobrow, stefik

Russ,
	Let's get together for lunch soon.  I'll cc Dan Bobrow in case he's
interested as well.  We'll all be gone next week to MIT.  Pick 2 or 3 lunch dates
in the week of Feb 1, tell us, and we'll coordinate when we get back.  Please
send your preliminary examples along.  Cheers Mark

∂TO %parc stefik, bobrow 16:37 22-Jan-82
Discussion time, overview, etc.
Mark, Dan:
Thanks for your (hopefully not misplaced) enthusiasm!
Potential lunch times:  The best day for me is Wednesday, 3-Feb,
followed by Monday, and then Tuesday of that week.

Let me give a sketchy example of the type of scenario I think 
my eventual program should be able to perform.

-----
Consider the task of inputting rules to Dendral.
There will certainly be many times when the rules used for
one class of chemicals were essentially identical to those for another class,
but not quite.
To illustrate, imagine you had just enterred the dozen rules specific to hydrates.
Now comes the time to enter the corresponding rules to be used for
ammoniates.  (To a first approximation,) these new rules should be deducable
from that first set -- making only the changes needed to distinguish
amine compounds from hydrozyl compounds.  
That is, each new ammoniate rule, R, should be analogous to a hydrate rule, R'.

It would be nice if a program could generate this R using a relation like
	hydrates : ammoniates :: R' : ?.
(This relation, in turn, would be based on a body of facts known about
hydroxyl and  amine compounds).
However, lacking such an analogizing routine, you the user would be forced
to tediously enter a new rule R for each such R'.
-----

Anyway, I'm currently searching for a crisp example of when this actually happened,
to replace this imprecise contrived one.  Certainly something like this
must have happened as you were "educating" Molgen, yes?

There is one other, possibly relevent paper: my overgrown thesis proposal.
An updated version will soon be placed on [MAXC]<GREINER>THESIS.PRESS --
feel free to read as much of it as you like.  (Only the first few pages
are really relevant, to provide a feeling for some of my current objectives.)

You'll find the proposal has relatively little in common with this example
scenario -- that disparity should provide some insight into my current
state of confusion and aimlessness...

Finally, now that you've whetted my curiousity, can you relay what is
happening at MIT?

	Thanks,
Russ

∂02-Feb-82  1359	Stefik at PARC-MAXC 	Re: Discussion time, overview, etc.    
To: Russell Greiner <RDG at SU-AI>
cc: stefik, bobrow

Russ -- It seems that we have missed your optimal days this week -- since we
have a hard conflict on Wednesday.  Next week there are conflicts here on Mon
and Wed.  What is your schedule?  Mark

[rdg] - 12:15PM 5/Feb/82
∂23-Apr-82  0932	TW   
Russ,

I am finally at a point where (modulo Murphy's Law) where 
my book will be finished this week, so I have started excavating
my desk and in the mesozoic layer came across your proposal, which
I read (skimming all but the first few pages).  Since it has been
so long, I'm not sure what you are thinking now, so I'll make a few
coomments based on the written version, and we can get together
for further discussion.

My overall impression is that it is not a good thesis topic.  Analogy
is of course a fascinating problem, but not one that lends itself to
the kind of approach you propose.  My main objection is that in order
for something to be tacklable as a thesis project (or a substantive
scientific endeavor of any kind) there must be a more precise question
or problem.  My model of AI research is that it works well when there
is a domain in which there is already a clearly successful body of
expertise at least partially understood.  Of course every human has
expertise in analogy (and language, and vision, and thinking, ...)
but this is not at all parallel to expertise in chess or spectrometry.
If there were ten good programs that did useful and interesting things
with analogy, it would be time to come up with generalizations.  Since
there aren't, the only way to get anywhere is to try and create one.  My
fear is that by working at such a highly general level you will end
up with mush.   I predict that the thesis will consist of: a) a lot
of amusing speculation and handwaving; b) a program that does some
trivial task, for which it was hand tailored, along with a lot of
excuses about why this is not to be taken as a serious application
but as an evocative example; c) more speculation on what it would
take to apply the ideas in more realistic domains.

My advice (if you want to stay in the same general area) is to look
for a specific domain where: a) analogical techniques would be of
demonstrable use and effectiveness; b) you (not the user to whom
you pass along great flexibility) could make them actually work
and accomplish something.  It may not be easy to find one, but
without it I just don't see getting anything solid enough to be
interesting.

As I said, I'm not sure what your thoughts are now and would
be glad to talk more.
--t

---
∂TO tw 23-Apr-82
Proposal, and related things
Terry,

Thanks for the comments.  
Yes, I would value the opportunity to hear your views on my thesis related ideas.
I've forgotten which iteration of my thesis proposal you read,
but, based on your comments, I assume it was a fairly early (pre)release.

I've spent the last few months working on a "What's in an Analogy" paper,
trying to pin down some thoughts on analogies in general,
towards understanding what type of task I should address.
(A draft is now being circulated among some peers 
-- after incorporating their comments, perhaps I could get your response
as well.)

The particular research task I'm now considering 
(which I don't think was mentioned in that earlier proposal,)
deals with the use of analogies in Knowledge Acquisition.
The domain expert will often use an analogy to introduce or further
specify some new term;
the goal of this work is a module which can "understand" those analogies.
For example, consider the task of teaching an expert system how to use EMACS.
Once it knows about C-F, C-B, C-<rubout> and C-D, 
it would be nice to simply state that 
	"Word commands are like character commands,
	 except they use Meta rather than Control".
Assuming the system already knew some core editor principles, 
(eg what type of thing a word is,
or the fact that common commands have been designed to require few keystrokes, etc)
this instruction would communicate a great deal --
if not everything needed, at least enough that this KA program would know
what types of refining questions to ask to (more completely) understand these
what these new commands do, and how (and when) to use them.

There are still a great number of issues, even for this trivial example.
As noted above, this program will have to have access to a
large body of facts about the domain (here editors); but not too much.
E.g., if the system has a complete working knowledge of TECO, 
and had the definition of each EMACS command in terms of its TECO code,
it would be wasteful to bother the system with this superfluous analogy.
Another issue is the interface itself --
how does one describe an analogy?
Then, for each description, what types of inferences may (or must) follow?
What are the appropriate refining questions? etc etc etc.

Some first thoughts on these questions appear in that "What's ..." paper;
other ideas are in drafts of other documents.

Anyway, if you are still interested, when could we meet?  Perhaps sometime
next week?  Let me know if you would prefer glancing over the "WIaA" paper
beforehand.

Russ

-----
PS Did you get the AI Qual comments I left on your desk a few weeks ago?
∂TO darden@sumex, csd.dietterich@score, tom@kestrel, csd.gardner@score,
	csd.crangle@score, csd.pcohen@score -- 16:15 23-Apr-82
Get 'em while they're hot!
I've spend the last few weeks (well, months) working on a
"What's in an Analogy" paper.  
This 45 page monstrosity is now (just about) ready for perusal.  
I'd much value comments from members of this esteemed crowd.
Please send me a message if you are interested;
or just snarf the file -- NAIVE.PRE[rdg,dbl] at SAIL
(the source is NAIVE.MSS[rdg,dbl]).

Thanks.
	Russ

∂27 Apr 1982 1241-PDT	<CSD.CRANGLE at SU-SCORE>	analogies
To: csd.greiner at SU-SCORE

Russ,  I would very much like to read your paper.  Can you tell me how to
"snarf" a file.  (I'm serious about the request - what's the easiest way to
get a copy ? )    Colleen
-------

∂25-Apr-82  1231	TW   
It would be helpful to see the paper before meeting, so why don't you
drop it by.  Late next week might be OK, but let's check midweek to see.
Yes, I did get the qual comments, and have tried to improve things basded
on them. --t

Dropped it by on 3-May
∂TO sklein@isib 15:13 3-May
Ta Daaaa!!
Steve -
	I deposited copies of NAIVE.PRE and NAIVE.DOC on [ISIB]<MANCOM.RLL>.
These are two versions of the analogy file I've been mentioned for
oh-so-many-months.  Let me know if these are readable 
(typographically and/or semantically).  I'd of course enjoy hearing
your comments.
	Thanks,
Russ
∂04-May-82  2002	ML  	analogy   
Hi Russ,
	I just got back from MIT last week where I was doing some work with
Pat Winston on his analogy program that learns from precedents. We were applying
it to reasoning about the function of an object given a shape description of
an object. I might be developing this into a thesis topic, and I'd like to talk
with you about analogy, particularily as a method of learning. Is there a 
good time for you to meet about 1 hr. ?
				Mike Lowry

∂05-May-82  1224	RDG  	Function from form...   
Sure, whenever.  Thursday after 4 and Friday noon are bad, but almost any
other time.  How about tommorrow at 11AM?

By the way, I have a long "what's in an analogy" draft, already for comments.
Feel free to DOVER NAIVE.PRE[rdg,dbl].
	Russ