What Needs to be Shared to Enable the

What Needs to be Shared? • • bits/bytes/streams/network… alphabet, special characters, … words, morphological variants, … syntactic meta-level markups (HTML) semantic meta-level markups (SGML, XML) content (logical representation of doc/page/. . . ) context (common sense, recent utterances, …)

• content (logical representation of doc/page/. . . ) • context (common sense, recent utterances, …)

Help find relevant information Combine info. from n sources Back & forth clarification dialogue • content (logical representation of doc/page/. . . ) • context (common sense, recent utterances, …)

Help find relevant information • Query: “Someone smiling” • Caption: “A man watching his daughter take her first step”

Help find relevant information How it works in CYC® • Query: “Someone smiling” l ica (d n) tio c du e re e nf I e nc ( x) (feels. Emotion x Happiness Positive) g Lo • Caption: “A man watching his daughter take her first step” ( x, y) (and (father x y) (gender x Female) (sees x y) (walking

Help find relevant information How it works in CYC® ( x) (feels. Emotion x Happiness Positive) l ica (d re e nf I e nc n) tio c du e g Lo ( x, y) (and (father x y) (gender x Female) (sees x y) (walking

• Query: “Show me pictures of strong and adventurous people” • Caption: “A man climbing a rock face”

Find/clean (consistency-check) info.

Find/clean (consistency-check) info. Combine info. from n sources Back & forth clarification dialogue

Combine info. from n sources DB 1 DB 2 DB 3 DB 4 CYC HAL you! DB 5 DB 6 DB 7 DB 8

Combine info. from n sources “Which hit TV shows from the past ten years have starred two people with the same astrological sign? ”

• combine info. from n sources DB 1 DB 2 DB 3 DB 4 DB 7 DB 8 CYC DB 5 DB 6 DB 7 DB 8 DB 5 DB 6

• find info. by inference (+KB) • combine info. from n sources • back & forth clarification dialogue

ELIZA: Rogerian Psychiatrist (1965 Joe Weizenbaum, MIT) • Carl Rogers advocated reflection Patient: “I swear a lot. ” “How do you feel about the fact that you swear a lot? ” • Counted a lot on anthropomorphism • Count on patient-to-doctor respect/awe/… Patient: “My dog’s mother died recently. ” “Tell me more about your mother. ”

ELIZA: Rogerian Psychiatrist “blah mother blah” Patient: “My dog’s mother died recently. ” “Tell me more about your mother. ”

ALICE • JUDGE 11[9: 26: 0]I WILL ASK YOU SOME QUESTIONS • PROGRAM[9: 26: 0]Do you think your plan will succeed? • JUDGE 11[9: 26: 35]WHERE IS SUE’S NOSE WHEN SUE IS IN HER HOUSE • PROGRAM[9: 26: 35] Try searching the World Wide Web. • JUDGE 11[9: 27: 25]HOW IS THE FATHER OF ANDY S MOTHER RELATED TO ANDY • PROGRAM[9: 27: 25]Fine as far as I know.

• find info. by inference (+KB) • combine info. from n sources • back & forth clarification dialogue Why has NLU advanced so little in 35 years? • It requires having lots of knowledge • Full understanding or bust

Natural Language Understanding requires having lots of knowledge 1. The pen is in the box. The box is in the pen. 2. The police watched the demonstrators… …because they feared violence. …because they advocated violence.

Natural Language Understanding requires having lots of knowledge 3. Mary and Sue are sisters. Mary and Sue are mothers. 4. The White House announced today that. . .

Natural Language Understanding requires having lots of knowledge 5. Every American has a mother. Every American has a president. 6. John saw his brother skiing on TV. The fool didn’t have a coat on! The fool didn’t recognize him!

AI: Dodo or Phoenix ? 1982 2001 Lenat Pessimistic Optimistic The Rest of the World Optimistic Pessimistic

What Changed ? 1982 2001 No CYC Codifying and formalizing (in machine-manipulable form) the millions of the things that we assume everyone knows

CYC Bridges the “Knowledge Gap” Water is wet upper ontology intermediate theories Vehicles slow down in bad weather task-specific knowledge HUMMV’s lose 18% traction in 4 -inch-deep mud

rate of learning 2001 1984 How will we get that knowledge? g via learnin nguage tural la na e ov ry c ng rni b is yd lea codify & enter each piece of knowledge, by hand CYC amount known 400 person-years 15 realtime years $40 million Fro nti er of hu ma nk no wle dg e

Lessons Learned 1. Use an expressive language, to represent knowledge. (nthorder predicate calculus, rather than frames or C++ etc. ) 2. Separate the epistemological problem (what it should know) from the heuristic problem (how to reason very efficiently). 3. Sacrifice completeness; do resource-limited best-first search. 4. Tolerate and even embrace redundancy as a useful heuristic. 5. Beware of “made-up” probabilities for everyday assertions. 6. Reasoning is default; gather and weigh pro/con arguments. 7. Give up on global consistency; only insist on local consistency. Reason with facts nearby in an n-dimensional “context space. ”

“It’s just common sense” • • CYC effort: “automating the white space” CYC is a power source, not an application CYC applications Announcement

CYC Applications • NLU/Speech (incl. prosody) – conversable appliances – games and training simulations • Semantic integration of n SIS’s in linear time – databases, spreadsheets, web pages, email forms • Security : scan a network, model it in the KB – reason about it: discover viable multi-step attacks, including cyber- and macro-level vulnerabilities, and answer “what-if” questions about any changes

CYC Applications CYCORP : WWW. CYC. COM

ANNOUNCEMENT • content (logical representation of doc/page/. . . ) • context (common sense, recent utterances, …)

OPENCYC: Dimension 1 • topics • depth • contexts Water is wet upper ontology Vehicles slow down in bad weather intermediate theories (weather, paths, emotions, …) task-specific knowledge HUMMV’s lose 18% traction in 4 -inch-deep mud

OPENCYC: Dimension 2 • • the names of the terms (constants) explanatory (NL) comments for each term taxonomic relations (element-of, supersets, . . . ) constraints (argument-types, disjointness, …) most fundamental facts about each term most fundamental rules about each term lesser-known non-proprietary facts and rules

OPENCYC: Dimension 3 • • inspect online at www. cyc. com download flat file of terms, assertions, etc. download DAML/XML file of that content download a version with a browser/navigator download a version with an editor/transcriptor download a version with collaboration toolbox download a version with full RKF tool support

OPENCYC: Dimension 4 • free for internal R&D only • free internal deployment • free use in external products/services – public license (such as Apache or GNU GPL)

OPENCYC: Quarterly Releases • Each release will move content forward one “click” in one or more dimensions • Each existing piece of Cyc will eventually move through this process, all the way to public (GPL) • Each (sub)context will be eventually be released • Each topic will be released in more detail • Each OE tool will eventually be released – incl. NL dialogue tools; incl. each covered language

OPENCYC: Release #1 (and 0. 9) • Release 1. 0: JULY 1, 2001 • Various offerings throughout the 4 dimensions • Pre-alpha v 0. 9 is available now www. cyc. com – 3, 000 most fundamental terms. 30, 000 assertions – very broad, not very detailed, “NY Times” context – terms, taxonomy, constraints (not rules) – browse or download a flat file; no OE tools released – public use (free commercial use)

OPENCYC: Administration • Managed by Open. Cyc. org – coordination point for new terms/constraints/rules – determination of release content, each quarter, based on maturity and stability of that content • Cycorp will continue to develop Cyc, and make ever more content available to Open. Cyc. org • Cycorp will develop and aggressively market applications of Cyc based on Open. Cyc. org

OPENCYC: More Information • check out the latest news on www. cyc. com • similarly, for more information about CYC in general, and using it as a semantic interlingua • press conference here on Tuesday 10: 30 am Level 3, Room 4 B. Text of the press release will be available on our website www. cyc. com