[comp.ai] Eric Mueller's Daydreamer

jmh@coyote.datalog.com (John Hughes) (06/03/91)

I recently finished Eric Mueller's book "Daydreaming in Humans and
Machines", and I was quite taken aback by the overall coherence of the
internal 'daydream' stream-of-thought generated by the program. I was
wondering why I haven't seen more about this book. Has anyone read it?
If so, what did you think of it?

(apologies if I mispelled the author's name)


-- 
|     John M. Hughes      | "...unfolding in consciousness at the            |
| datalog.com!moondog!jmh | deliberate speed of pondering."  - Daniel Dennet |
| jmh%coyote@noao.edu     |--------------------------------------------------|
| jmh%moondog@datalog.com | P.O.Box 43305, Tucson, AZ 85733    602-624-8008  |

csd10@seq1.keele.ac.uk (Y.O. Busia) (06/04/91)

In article <1991Jun3.075711.11333@coyote.datalog.com>, jmh@coyote.datalog.com (John Hughes) writes:
> I recently finished Eric Mueller's book "Daydreaming in Humans and
> Machines", and I was quite taken aback by the overall coherence of the
> internal 'daydream' stream-of-thought generated by the program. I was
> wondering why I haven't seen more about this book. Has anyone read it?
> If so, what did you think of it?

nagle@well.sf.ca.us (John Nagle) (06/05/91)

jmh@coyote.datalog.com (John Hughes) writes:

>I recently finished Eric Mueller's book "Daydreaming in Humans and
>Machines", and I was quite taken aback by the overall coherence of the
>internal 'daydream' stream-of-thought generated by the program. I was
>wondering why I haven't seen more about this book. Has anyone read it?
>If so, what did you think of it?

     The idea is interesting, but it seems that most of the potential
daydreams were essentially built into the program.  As Muller puts
it, under "Shortcomings of the Program", "DAYDERAMER cannot daydream
for a long time and cannot generate many novel sequences".  The
program has provided to it a number of possible script fragments
and some rules for assembling them, and the number of possible scripts
resulting is limited.  

     There was a system in the 1950s (!) which generated scripts for
TV Westerns by somewhat similar means.  Anyone have the reference?

     Mr. Mulller seems to have moved from daydream to speculation.
The preface ends "Preparation of the revision for publication as a book was
made possible by the Analytical Propretary Trading Unit of Morgan
Staley and Company in New York."

					John Nagle

hm02+@andrew.cmu.edu (Hans P. Moravec) (06/05/91)

> From: nagle@well.sf.ca.us (John Nagle)
> Subject: Re: Eric Mueller's Daydreamer
>     There was a system in the 1950s (!) which generated scripts for
> TV Westerns by somewhat similar means.  Anyone have the reference?

Knuth "The Art of Computer Programming" Volume 2 (p 174:175) mentions it,
  and gives sample output.
  It was written especially for a (very good) 1960 CBS TV program
  entitled "The Thinking Machine".  I (re)read more about it recently
  in a 1961 paperback based on that program excavated from my
  archives, but it's at home, and I'm not.
                                           -- Hans Moravec

G.Joly@cs.ucl.ac.uk (Gordon Joly) (06/06/91)

In a similar vein, a friend wrote a program to generate music, in the
style of Syd Barrett, (he of the early Pink Floyd) in BASIC. Circa 1980.

These days IBM have products (in house only?)  to check your English
language style.

Gordon Joly                                       +44 71 387 7050 ext 3716
Internet: G.Joly@cs.ucl.ac.uk          UUCP: ...!{uunet,ukc}!ucl-cs!G.Joly
Computer Science, University College London, Gower Street, LONDON WC1E 6BT

                        Drop a utensil.

srt@aero.org (Scott "TCB" Turner) (06/10/91)

John Nagle writes:
>The idea is interesting, but it seems that most of the potential
>daydreams were essentially built into the program.  As Muller [sic]
>puts it, under "Shortcomings of the Program", "DAYDREAMER cannot
>daydream for a long time and cannot generate many novel sequences".
>The program has provided to it a number of possible script fragments
>and some rules for assembling them, and the number of possible
>scripts resulting is limited.

DAYDREAMER is a finite program with no learning capability.  Naturally 
the number and kinds of daydreams it can produce is limited.

DAYDREAMER was intended as an exploration into the types of processes
involved in daydreaming.  The daydreams it produces demonstrate that
the processes and knowledge structures Mueller uses can lead to
plausible results.  DAYDREAMER is not intended as a "performance"
model of daydreaming, and it isn't meaningful to criticize it on that
basis.  I think there are some problems with DAYDREAMER, but the fact
that it can't produce an infinite number of daydreams isn't one of
them.

In AI, creativity is a much younger problem than natural language
understanding or planning.  (Because, I think, it is a much harder
problem.)  Criticizing an early creativity program because it can only
produce "limited" output is like criticizing SHRDLU because it could
only handle the blocks world.  It may be that someday we'll have 
creativity programs that, given a large knowledge base, can invent 
endlessly and fruitfully.  But at this stage of our understanding and 
research into creativity that isn't to be expected.

What you might question is whether DAYDREAMER produces enough
different daydreams to support Mueller's model of daydreaming.
Although you trivialize DAYDREAMER as a program "with some scripts and
rules to combine them" I think that reading Mueller's book will
convince most readers that DAYDREAMER does indeed perform well enough
to support the plausibility of Mueller's theories.

					-- Scott Turner