I-F (Interactive Fiction) is an attempt to write stories that are
interactive, and embody a great deal of knowledge about the world that
the story is about.
How many I-F stories are about worlds that don't exist?
IE: the 'assumptions' in the story wouldn't be incorporated in the common
sense database created by the CYC researchers?
It seems to me that a lot of the I-F stories would need augmentation
to a CYC type database.
There's an interview (Discover magazine?) where Lenat says CYC isn't
fazed by the 'conflicting' facts that 1) unicorns don't exist and
2) unicorns have a single horn.
I guess there's some kind of modality/aspect arrangement that allows
opposite facts to be true in different ways...
What would you get if you rewrote SimCity in TADS? (or "TADS++" ;^/
>CYC is an attempt by some researchers to record the hundreds and thousands
>of facts that we call "common sense".
>I-F (Interactive Fiction) is an attempt to write stories that are
>interactive, and embody a great deal of knowledge about the world that
>the story is about.
CYC attempts to model all the facts that make up late 20th century human
common sense, it seems to me. I think people writing sophisticated I-F
certainly *should* take notice of CYC, but I think a CYC / I-F comparison is
a little premature. Most of the code in most of the I-F that I've ever seen
is not there to model knowledge or common sense, but is in fact
instantiating the world in which the action takes place. I-F is usually a
simulation of a small world that doesn't exist. Knowledge representation in
I-F occurs infrequently, and in I-F without halfway reasonable actors it
doesn't occur at all.
I think there is an isomorphism that explains the importance CYC-like
projects have for I-F:
Real Life Reality I-F Reality
Knowledge base of people <====> Knowledge base of actors
(CYC tries to model this) (Rarely implemented to
a great extent)
People themselves <====> Actors, and the routines
that give them a semblance
of life and intelligence
The physical substrate, <====> The I-F world: objects,
ie. planets, rocks, trees rooms and their
marshmallows, houses. descriptions, links
Thus CYC is suggesting to us that to write better I-F, we should incorporate
a common knowledge base for all our actors (and then perhaps build dynamic,
actor-specific knowledge on top of that). Examples would be all actors
knowing that they need a light to go into a dark room; that if somebody
hits you, you should either hit back or run away; that you can only carry a
certain amount; that a locked door will stop someone who's chasing you; that
someone who attacked you once will probably do so again; that objects tend
to remain where you leave them, etc.
>It seems to me that a lot of the I-F stories would need augmentation
>to a CYC type database.
I agree with your point, David, that our common sense knowledge and the
common sense knowledge appropriate to a particular I-F world will not be the
same. In that sense, an I-F knowledge base would need augmentation, ie. it
would include things not mentioned in CYC.
However, I-F knowledge representation has the advantage that the number of
possible relationships between people, places and things is determined by
the nature of the code that represents the simulated world. In a default
TADS world, for instance, one object can be in the possession of an actor,
loose in a room, on something, or in something. That's about all (roughly).
In real life, which CYC has to come to grips with, there are many more
permutations for possible people-place-thing relationships. Consider "I've
got a frog in my throat". (Anyone care to write metaphor.t?) The point
being that I-F knowledge representation is dealing with a simpler world, and
thus should be a little easier to implement, and somewhat less unwieldy than
the mushrooming CYC database.
Jason Noble | jno...@bunyip.bhs.mq.edu.au
National Centre for HIV Social Research | jno...@laurel.ocs.mq.edu.au
Macquarie University, Sydney, Australia | ph. (61 2) 850 8667
I contacted MCC about 7 months ago and told them I wanted to use Cyc
for interactive fiction. They said they would mail our department the proper
forms to begin with, but never did.
> Doug Lenat has proposed using Cyc for interactive fiction or virtual
> reality, but the odds that you'll get to incorporate it into TADS are low.
> Cyc is free, but you have gto sign a nondisclosure agreement to get it.
> They don't mean to put it into the public domain. They won't even give it
> to someone whose security is not good enough for them.
As I know one of the long term plans with Cyc is to have one or many
of these online, available through a TCP connection, so you could issue
questions and get resolutions. I'm not sure of the payment system, though.
Personally I think this makes sense, because then you don't need to waste
a lot of CPU cycles, neither disk or RAM space, for a fairly complete
knowledge system. This all also implies that households and similar
environments where games using Cyc are distributed will have access
to one or many decent high-speed networks. I truly hope that this aspect
of possibilities -- querying existing bigger servers on a network -- would
be fully explored. There's a lot of intriguing cross-relations based
on such designs. Sort of open-ended services, knowledge available
as easily as electricity. And the basic games could be run on a Nintendo
with 1MB RAM!
Wintermute is nearer than we think.
KenT SandviK: san...@apple.com
-- Private activities on the net. ---
> I contacted MCC about 7 months ago and told them I wanted to use Cyc
> for interactive fiction. They said they would mail our department the proper
> forms to begin with, but never did.
What exactly is Cyc?
Erik Max Francis, &tSftDotIotE ...!uuwest!alcyone!max m...@alcyone.darkside.com
USMail: 1070 Oakmont Dr. #1 San Jose, CA 95117 ICBM: 37 20 N 121 53 W _
H.3`S,3,P,3$S,#$Q,C`Q,3,P,3$S,#$Q,3`Q,3,P,C$Q,#(Q.#`-"C`- ftmfbs kmmfa / \
Omnia quia sunt, lumina sunt. ("All things that are, are lights.") -><- \_/
It's an attempt to encapsulate "common" knowledge (the knowledge
of the world a high school teacher might expect of their students)
in an efficient, useful database, allowing AI researchers to move
beyond the toy domain, and develop techniques that work in the real
world (the theory being that many current AI techniques seem to
fail when brought beyond their limited test domains into areas
where a significant amount of real-world knowledge is required).
There is an good article in the July (I believe) issue of Communications
of the ACM, in which Doug Lenat (the project leader) discusses CYC's
history and current status.
"We're aimed the wrong way to be going home, Gumby."
"Home...? We're on an express elevator to HECK!"
boy is that an open ended question.
D. Lenat is head of a group in Austin Texas that is attempting to
record the 'common sense' knowledge of the average 20th Century
American citizen in a computer.
He estimates (at one writing) that he is talking about 10^8 facts.
This is not including the relationships between the facts, but just the
different fact-ideas that are 'commonly known'
This includes everything from horses have four legs to the fact that twins
are the same age.
Who knows if he can do it?
On a related note, the mailing list wis...@mcs.com is trying to lay the
ground work to create a mini-CYC that handles the subdomain of I-F....
Doug Lenat's CYC (from enCYClopedia) Project at MCC in Austin, a ten-
year $35 million project begun in 1984, uses logic, too, but emphasizes
the Aristotle-problem, employing a dozen 'ontological engineers' to
enumerate tens of millions of common-sense facts that will ultimately
'add up' to logical intelligence. Lenat's goal is a system that can
understand and speak ordinary language, and detect violations of common
sense as readily as humans can.
As of 1994, CYC's sponsors are: Apple, Bellcore, DEC, the DOD, Interval,
Kodak and Microsoft. A version of CYC for Macintoshes is supposed to
be released this year.
Lenat has revised his estimate of the total number of 'rules' required
for this, upward by a factor of ten (to 20-40 million), and extended the
time needed by another ten years. It bothers me a lot that the sort of
thing being added apparently includes rules like, "A creature with two
arms probably has two legs." This seems out-of-control to me. CYC's
ontology includes abstractions like:
The Text Adventure Development System (TADS), by contrast, offers the
following, much more pragmatic, partial object hierarchy:
Item: vehicle, surface, lightsource, key, food, container, clothing
FixedItem: switch, dial, button, decoration, actor, chair
CYC's data-objects offer such slots as: instanceOf, inverse,
makesSenseFor, entryIsA, specSlots, slotConstraints, becameTrueIn,
qualitativeValue, sufficientCondition. One must expect that some of
these slots will ultimately have thousands of fillers, requiring hash-
tables and slowing processing proportionately, and that some objects
will have thousands of slots, causing similar problems. Interestingly,
in a recent interview Lenat claimed that the number of *facts* has been
at a plateau lately, even as the amount of useful knowledge continues to
grow, because massive redundancies in the representation are also being
detected and repaired.
CYC/MCC's presence on the Net is unfortunately very low-profile. There
was an excellent Nova (?) program about it, in a series on the brain.
Here's some periodical references, and the only book:
"CYC" AI Magazine 7(1), 1986
"When will machines learn?" Machine Learning, Dec 1989
"Cyc: Toward Programs With Common Sense" CACM, Aug 1990
"Knowledge and Natural Language Processing" CACM, Aug 1990
"Common Sense and the Computer" Discover magazine, Aug 1990
"Cyc: A Mid-Term Report" AI Magazine, Fall 1990
"The commonsense reviews" Artificial Intelligence, 61(1), 1993
"CYC-O" Wired magazine, Apr 1994
"Enabling agents to work together" CACM, 37(7), 1994
Douglas B. Lenat and R.V. Guha, "Building Large Knowledge-Based
Systems" Addison-Wesley 1991
> Lenat has revised his estimate of the total number of 'rules' required
> for this, upward by a factor of ten (to 20-40 million), and extended the
> time needed by another ten years. It bothers me a lot that the sort of
> thing being added apparently includes rules like, "A creature with two
> arms probably has two legs." This seems out-of-control to me. CYC's
> ontology includes abstractions like...
I have to disagree with the assessment that knowledge of the form `a
creature with two arms probably has two legs' is in any way out of
control viz-a-viz Cyc's stated objective of representing enough `common
sense' to adequately engage in discourse on the level of, say, the
average high-school student. After all, my personal `wetware'
knowledge base certainly contains the rule `a creature with two arms
probably has two legs,' and also contains concepts to the effect that a
human exception to this rule is known as a `paraplegic.' Etc.
Granted, when you consider how big Cyc must be, it seems impractical
for implementation and delivery of a work of Interactive Fiction on,
say, the average (or even muscle-machine) PC of today. However, if you
contemplate the creation of a large, multi-player virtual world
Interactive Fiction project, suddenly Cyc starts to sound very
attractive (at least to me, and I have read `Building Large
Knowledge-Based Systems,' among other heavy AI works, both those
related to Cyc and many that are not).
While we're on knowledge representation, has anyone besides me bothered
to look at LOOM, from USC's Information Sciences Institute? And does
anyone know how I can license AT&T's CLASSIC system?
Paul F. Snively "Just because you're paranoid, it doesn't mean
ch...@shell.portal.com that there's no one out to get you."
Mine doesn't. I had to make an inference when I tried to assess the
validity of that statement. I seem to have quickly narrowed
"creature" to "human" when I first tried to find a plausible-true
interpretation of that statement, but even now I have problems
asserting the statement's general truth. Part of the problem is
"creature". Is a centaur a creature? Too much fantasy in my life..
My knowledge base does have a statement to the effect that normal
vertebrate lifeforms have four limbs of some sort. But nothing
explicitly about two arms -> two legs. I had to run through a few
examples before I could decide it was reasonable.
I guess I have problems engaging in normal discourse..
Right, but the point is, why is this sort of statement being
explicitly represented in CYC if it's something that can be inferred
by relating other background knowledge?
Are they also going to include the rules "two arms -> two hands", "two
arms -> two ears", "two arms -> one nose", "two arms -> two nostrils",
"two arms -> two lungs", "two arms -> one heart", etc.?
> Right, but the point is, why is this sort of statement being
> explicitly represented in CYC if it's something that can be inferred
> by relating other background knowledge?
> Are they also going to include the rules "two arms -> two hands", "two
> arms -> two ears", "two arms -> one nose", "two arms -> two nostrils",
> "two arms -> two lungs", "two arms -> one heart", etc.?
You make it sound as though facts are either a) force-fed to Cyc, or b)
arrived at by inference and then immediately forgotten. Neither of
those is true. It's not clear to me that anyone `told' Cyc that having
two arms implies two legs. It wouldn't surprise me to discover that it
was a rule that Cyc had arrived at on its own.
> Mine doesn't. I had to make an inference when I tried to assess the
> validity of that statement. I seem to have quickly narrowed
> "creature" to "human" when I first tried to find a plausible-true
> interpretation of that statement, but even now I have problems
> asserting the statement's general truth. Part of the problem is
> "creature". Is a centaur a creature? Too much fantasy in my life..
Perhaps. I guess my question would be `If you run through a list of
creatures with two arms that you've actually seen, do most of them also
have two legs?' I think the giveaway is that having two `arms' is a
precondition that immediately triggers my mind to narrow the search
basically to primates, and the overwhelming majority of primates I've
seen also have two legs.
> My knowledge base does have a statement to the effect that normal
> vertebrate lifeforms have four limbs of some sort. But nothing
> explicitly about two arms -> two legs. I had to run through a few
> examples before I could decide it was reasonable.
> I guess I have problems engaging in normal discourse..
I wouldn't say that follows. ;-) What I might conclude is that we drew
our inferences at different points in time... but the likelihood that
we're both caching them is quite high (in fact, I'd go so far as to say
that the brain's great strength is the amount of caching of
already-performed inferences that it does).
My source for this *paraphrase* was someone who saw the Nova (?) special
on Cyc (I think part of a series on the brain). The way I heard it,
the on-camera humans were trying to fix an inferencing problem and
proposed adding this rule as the solution...
This project was discussed in one episode of "The Machine that
Changed the World" documentary series that aired a year or two ago. They
discussed the types of relationships the computer "figures out".
That is, they feed it a whole bunch of data during the day, and at
night, it cranks through the data and outputs inconsistencies or the types of
relationships it makes. The only one I remember with even a tiny bit of
information was that at first, the people they gave it information about was
the people who were on the project, and then it made the 'logical' conclusion
that all other people were like the people on the project.. (geeky in various
unk...@apple.com Apple II Forever
These opinions are mine, not Apple's.
The subject of patterning, which is what this CYC project is really about,
or so it sounds to me, is of great interest to educators, who call
difficulty with patterning dyslexia. In journalism we used to say two facts
and a theory made a story possible, and three facts required one. Is pattern
recognition (identified as such) a technique that would be useful in IF
> Re the object that has two arms: I think this is logically interesting
> because putting it in suggests that the reverse (inverse? converse?
> obverse? I can never keep those straight) isn't true--that creatures with two
> legs usually have two arms. And it isn't true--viz, birds.
Right. It's important to remember in most knowledge representation
systems that when you assert something, the converse isn't asserted
automagically (nor should it be).
> The subject of patterning, which is what this CYC project is really about,
> or so it sounds to me, is of great interest to educators, who call
> difficulty with patterning dyslexia. In journalism we used to say two facts
> and a theory made a story possible, and three facts required one. Is pattern
> recognition (identified as such) a technique that would be useful in IF
I think so. I'm particularly thinking of Schank and Riesbeck's work on
Conceptual Dependency Theory and Scripts. If a piece of software can
accurately identify a script into which a series of player actions fit,
then presumably it can respond in a much more meaningful way than would
be otherwise possible.