Origins of the American Association for Artificial Intelligence

Home , David Luckham, David Waltz, Raj Reddy, Rodney Brooks, Roger Schank

AI Magazine Volume 26 Number 4 (2006)(2005) (© AAAI) 25th Anniversary Issue The Origins of the American Association for Artiﬁcial Intelligence (AAAI)

Raj Reddy

■ This article provides a historical background on how AAAI came into existence. It provides a ratio- nale for why we needed our own society. It provides a list of the founding members of the community that came together to establish AAAI. Starting a new society comes with a whole range of issues and problems: What will it be called? How will it be financed? Who will run the society? What kind of activities will it engage in? and so on. This article provides a brief description of the consider- ations that went into making the final choices. It also provides a description of the historic first AAAI conference and the people that made it happen.

The Background and the Context hile the 1950s and 1960s were an active period for research in AI, there Wwere no organized mechanisms for the members of the community to get together and share ideas and accomplishments. By the early 1960s there were several active research groups in AI, including those at Carnegie Mel- lon University (CMU), the Massachusetts Insti- tute of Technology (MIT), Stanford University, Stanford Research Institute (later SRI Interna- tional), and a little later the University of Southern California Information Sciences Insti- tute (USC-ISI). My own involvement in AI began in 1963, when I joined Stanford as a graduate student working with John McCarthy. After completing my Ph.D. in 1966, I joined the faculty at Stan- ford as an assistant professor and stayed there until 1969 when I left to join Allen Newell and Herb Simon at Carnegie Mellon University Raj Reddy. (CMU). The 1960s at Stanford AI Labs (SAIL)

A Sampling of AI Research in 1980, from the Proceedings of AAAI-80

Vision Badler, O’Rourke, Platt, and Morris, Uni- Problem Solving versity of Pennsylvania Early Vision Processing Using Contextual Information in Com- Cooperating Expert Systems Recovering Surface Orientation from Tex- puter Vision, Olivier D. Faugeras, USC HEARSAY-III: A Framework for Expert ture, Andrew P. Witkin, SRI International Systems, Balzer, Erman, London, and Program Synthesis Shape-from-Texture Paradigm, John R. Williams, USC ISI Kender and Takeo Kanade, Carnegie Mel- Question Ordering in Mixed Initiative Quantifying and Simulating the Behav- lon University Program Specification Dialogue, Louis ior of KBIS, Lesser, Reed, and Pavlin, Uni- Steinberg, Rutgers Low Level Vision Systems, William B. versity of Massachusetts Thompson and Albert Yonas, University of Some Algorithm Design Methods, Steve Representation of Task Knowledge in Minnesota Tappel, Systems Control, Inc. User Interfaces, Eugene Ball and Phil Interpreting Line Drawings as 3D Sur- Automatic Goal-Directed Program Hayes, Carnegie Mellon University Transformation, Stephen Fickas, USC ISI faces, Harry G. Barrow and Jay M. Tenen- Problem Solving and Control baum, SRI Incremental, Informal Program Acquisi- Representation of Control Knowledge tion, Brian McCune, AI and D Systems Shape Encoding and Subjective Con- in Expert Systems, Janice S. Aikins, Stan- tours, Brady, Grimson, MIT and Lan- A Basis for a Theory of Program Synthe- ford University gridge, CSIRO sis, P. A. Subrahmanyam, USC ISI DELTA-MIN: A Search-Control Method Scene Analysis A Program Model for Computer Aided for Information-Gathering Problems, Program Synthesis, Richard J. Wood, Carbonell, Carnegie Mellon University Information Needed to Label a Scene, University of Maryland Eugene C. Freuder, University of New On Waiting, Arthur M. Farley, University Hampshire Theorem Proving of Oregon Interpretive Vision and Restriction An Efﬁcient Relevance Criterion for Me- A Planner for Reasoning about Knowl- Graphs, Rodney Brooks and Thomas Bin- chanical Theorem Proving, David A. edge and Action, Douglas E. Appelt, SRI ford, Stanford Plaisted, University of Illinois Urbana- International Sticks, Plates, and Blobs: A 3D Object Champaign Making Judgments, Hans J. Berliner, Car- Representation for Scene Analysis, On Proving Laws of the Algebra, Jacek negie Mellon University Shapiro, Moriarty, Mulgaonkar, and Haral- Leszczylowski, Polish Academy of Sciences Multiple-Agent Planning Systems, Kurt ick, Virginia Polytechnic Institute and State Establishing Completeness Results in Konolige and Nils J. Nilsson, SRI University Theorem Proving, Peterson, University of A Simple Game-Searching Algorithm Motion Analysis Missouri at St. Louis with Proven Optimal Properties, Judea Automatic Generation of Semantic At- Pearl, UCLA Constraint-Based Inference from Image tachments in FOL, Luigla Aiello, Stanford Problem Solving in Using Interactive Motion, Daryl T. Lawton, University of Dialog, Harry C. Reinstein, IBM Palo Alto Massachusetts HCPRVR: An Interpreter for Logic Pro- grams, Daniel Chester, University of Texas Scientiﬁc Center Static Analysis of Moving Jointed Ob- at Austin Representing Knowledge in an Interac- jects, Jon A. Webb, University of Texas at tive Planner, Ann E. Robinson and David Austin First Experiments with Rue Automated Deduction, Vincent J. Digricoli, The Cou- E. Wilkins, SRI Bootstrap Stereo, Marsha Jo Hannah, rant Institute Inference with Recursive Rules, Stuart C. Lockheed Palo Alto Research Laboratory Shapiro and Donald P. McKay, SUNY Buf- Robotic Vision Mathematical falo and Theoretical Foundations Locating Partially Visible Objects: The Knowledge Representation Local Feature Focus Method, Robert C. What’s Wrong with Non-Monotonic Bolles, SRI International Logic? David J. Israel, Bolt Beranek and Advanced Knowledge Collision Avoidance Among 3D Ob- Newman, Inc. Representation jects, Ahuja, Chien, Yen, and Bridwell, Pathology on Game Trees: A Summary A Frame-Based Production System Ar- University of Illinois of Results, Dana S. Nau, University of chitecture, David E. Smith and Jan E. Automated Inspection Using Gray-Scale Maryland Clayton, Stanford Statistics, Stephen T. Barnard, SRI Interna- Max-Min Chaining of Weighted Causal Knowledge Embedding in the Descrip- tional Assertions Is Loop Free, Ng and Walker, tion System Omega, Hewitt, Attardi, and Human Movement Understanding, Rutgers University Simi, MIT

6 AI MAGAZINE 25th Anniversary Issue

A Representation Language, Russell Consultation System Behave Intelli- Trouble-Shooting by Plausible Infer- Greiner and Douglas B. Lenat, Stanford gently, René Reboh, SRI ence, Leonard Friedman, JPL, Caltech. Applied Knowledge Representa- An Approach to Acquiring and Apply- An Application of the Prospector Sys- tion ing Knowledge, Norman Haas and Gary tem to Uranium Resource Evaluation, G. Hendrix, SRI John Gaschnig, SRI Spatial and Qualitative Aspects of Rea- Self-Correcting Generalization, Stephen soning about Motion, Kenneth D. For- Some Requirements for a Computer- B. Whltehill, University of California, Ir- bus, MIT Based Legal Consultant, L. Thorne Mc- vine Computer Interpretation of Human Carty, Rutgers Stick Figures, Martin Herman, Carnegie Specialized Systems Mellon University Natural Language Intelligent Retrieval Planning, Jonathan Research on Expert Problem Solving in When Expectation Fails: A Self-Correct- J. King, Stanford University Physics, Novak and Araya, University of ing Inference System, Richard H. Gran- Texas at Austin A Theory of Metric Spatial Inference, ger, Jr., University of California Irvine Drew McDermott, Yale University Knowledge-Based Simulation, Philip Generating Relevant Explanations: Nat- Design Sketch for a Million-Element Klahr and William S. Faught, The Rand ural Language Responses to Questions NETL Machine, Scott E. Fahlman, Carne- Corporation about Database Structure, Kathleen R. gie Mellon Interactive Frame Instantiation, Carl McKeown, University of Pennsylvania Perceptual Reasoning in a Hostile Envi- Engelman, Ethan A. Scarl, and Charles H. The Semantic Interpretation of Nomi- Berg, MITRE ronment, Thomas D. Garvey and Martin A. Fischler, SRI nal Compounds, Timothy Wilking Finin, University of Illinois Specialized Issues in Knowledge Overview of an Example Generation Representation System, Edwina L. Rissland and Elliot M. Towards an AI Model of Argumentation, Descriptions for a Programming Envi- Soloway, University of Massachusetts Birnbaum, Flowers, and McGuire, Yale ronment, Ira Goldstein and Daniel Bo- Structure Comparison and Semantic In- University brow, Xerox PARC terpretation of Differences, Wellington Knowledge Representation for Syntactic Rule-Based Inference in Large Knowl- Yu Chiu, USC ISI / Semantic Processing, Bobrow and Web- edge Bases, William Mark, USC ISI Performing Inferences over Recursive ber, University of Pennsylvania A Process for Evaluating Tree-Consisten- Data Bases, Naqvi and Henschen, North- Language and Memory: Generalization cy, John L. Goodson, Rutgers University western University as a Part of Understanding, Michael Leb- Reasoning about Change in Knowledge- Piaget and Artificial Intelligence, Jarrett owitz, Yale able Office Systems, Gerald R. Barber, K. Rosenberg, University of California, Ber- Failures in Natural Language Systems: MIT keley Applications to Data Base Query Sys- On Supporting the Use of Procedures in tems, Mays, University of Pennsylvania Office Work, Fikes and Henderson, Jr., Xe- Applications rox PARC R1: An Expert in the Computer Systems Memory Models Metaphors and Models, Michael R. Gene- Domain, John McDermott, Carnegie Mel- Organizing Memory and Keeping It Or- sereth, Stanford University lon University ganized, Janet L. Kolodner, Yale University Rule-Based Models of Legal Expertise, D. About Authority Structures You Were Meta-Planning, Robert Wilensky, Univer- Unable to Represent, James R. Meehan, A. Waterman and Mark Peterson, Rand sity of California, Berkeley University of California, Irvine Corporation Narrative Text Summarization, Wendy Real Time Causal Monitors for Complex Exploiting a Domain Model in Expert G. Lehnert, Yale University Physical Sites, Rieger and Stanfill, Univer- Spectral Analysis, David Barstow, Sch- sity of Maryland lumberger Research Knowledge Acquisition A System for the Automatic Analysis of Business Correspondence, Lance Miller, Applying General Induction Methods IBM Research to the Card Game Eleusis, Tom Diet- A Knowledge Based Design System for terich, Stanford Digital Electronics, Milton R. Grinberg, Modeling Student Acquisition of Prob- University of Maryland lem-Solving Skills, Robert Smith, Rutgers Theory Directed Reading Diagnosis, University Christian Wagner and John Vinsonhaler, A Computer Model of Child Language Michigan State Learning, Mallory Selfridge, Yale University A Word-Finding Algorithm with a Dy- The Instructable Production System namic Lexical-Semantic Memory for Pa- Project, Michael D. Rychener, Carnegie tients with Anomia Using a Speech Mellon University Prosthesis, Colby, Christinaz, Graham, Using a Matcher to Make an Expert and Parkison, UCLA

WINTER 2005 7 25th Anniversary Issue

Photograph Courtesy Les Ezrnest. Stanford Artiﬁcial Intelligence Laboratory, 1968.

under the leadership of John McCarthy was indeed the golden age of AI at SAIL. Many of the areas of research at SAIL at that time continue to be active areas of research to this day. These include robotics, speech, vision, natural language processing, knowledge-based systems and heuristic programming, logic-based AI, and systems for playing games such as chess and checkers. In spite of much active research in AI at the many centers around the country, there were no organized mechanisms for fostering the field. The first organized effort was the holding of the International Joint Conferences on AI (IJCAI) in 1969. This effort was organized by Rebecca Prather and Don Walker of SRI. This biennial conference was the only mechanism that existed throughout the 1970s for AI researchers to get together. The 1977 IJCAI conference was held at MIT, and I was the program chairman that year. Or- ganizing a program committee, reviewing the papers, creating a coherent technical program, and producing and transporting the proceedings turned out to be a nontrivial task for a group of faculty members whose primary re- sponsibilities were teaching and research! The 1979 IJCAI conference was held in Don Walker. Tokyo, and I was the general chairman tasked to deal with many of the organizational issues. Site selection, financial management, interna-

8 AI MAGAZINE 25th Anniversary Issue tional coordination, communicating with the global community, and identifying and cajol- ing group leaders for different tasks needed continual attention.

Why We Needed Our Own Society The experiences of IJCAI-77 and IJCAI-79 con- vinced me that we need an enduring administrative mechanism and institutional memory such as those that are traditionally available at organizations like the Institute of Electrical and Electronic Engineers (IEEE) and the Association for Computing Machinery (ACM). For our community to thrive, it was clear that we needed more than just a biennial conference. Taking a cue from other scientific societies, it seemed desirable for AI to have its own society. We could have been part of one of the existing societies, but the time and the effort needed to get them to accept AI was too daunting. To quote Allen Newell from his “AAAI Presi- dent’s Message” published in the first issue of AI Magazine (and reprinted in this issue): Why did we come into existence? Note, I did not ask why we exist. Wherever, in our culture, a science is explored, a scientific society arises to nurture that exploration. AAAI exists because I remember a memorable lunch at an unmemorable Chinese the science of artificial intelligence is being ac- restaurant, where we sketched out the organizational design on tively and vigorously explored in the United a napkin, and Raj somehow talked me into agreeing to serve per- States. manently as conference chair. I agreed, but only on the condition … The one part of the birthday tale I haven’t that I could hire Lou Robinson to do most of the work. So many told is about the people. Societies are sui generis. other innovations came from Raj talking various people into They pass from nothing to something by an act things. of social will, in which particular people take to I also have some recollection of the first trade show we hosted themselves the prerogative of forming them- at Stanford’s Tressider Union. There were about 16 companies in selves to be a scientific society for whatever a room that maybe was 100 by 100 feet at most: robots from (here, for artificial intelligence). Some set of companies like Machine Intelligence, natural language querying people have to feel the calling and to make the systems from companies like Artificial Intelligence Corp, and inner decision at a particular point in history. For the AAAI the time was the recent IJCAI, perhaps some early expert system shells. Within a few years, at held in Tokyo in August 1979. The people were our peak, we had taken over Cobo Hall, the main conference almost entirely US participants on the IJCAI venue in Detroit! program and conference committees (including —Marty Tenenbaum some who served prior IJCAIs)—a fact of mild (AAAI Conference Chairman, 1980–1987) significance, as I’ll note in a moment. Their names appear as the founding council, whose lized. The story of AAAI’s origins comes safely formation by an act of self selection is required to rest at this point, and need be spun out no to get from nothing to something, society-wise. further. I was not one of them, so I can point out our Founding members of the community that (AAAI’s) collective debt to them for being the came together to establish the AAAI are given founders. in table 1. It is necessary to take one more step in the nar- rowing social microdynamics of how AAAI got started. For it finally comes down to Raj Reddy, What Is the Name? chairman of the IJCAI Board of Trustees and After many months of debates and discussion, general chairman of the IJCAI in Tokyo, who took the initial personal act of decision that re- we finally settled on the name of American As- ally started precipitation, and who carried all of sociation for Artificial Intelligence and the us before him, until the AAAI was safely crystal- acronym AAAI. Again to quote Allen Newell:

WINTER 2005 9 25th Anniversary Issue

The society has taken the name Artificial Intel- ligence. As all good AI’ers know, this name, in- Officers and Chairmen troduced by John McCarthy in the fifties, has been controversial for quite awhile. It is often President: Allen Newell,Carnegie-Mellon University remarked that some of the controversy that President-Elect: EdwardA.Feigenbaum, StanfordUniversity swirls around our field is due to our name—the Secretary-Treasure: DonaldE.Walker,SRIInternational suspicion of people outside AI, including some Membership Chair: Bruce G. Buchanan, StanfordUniversity in computer science, that our enterprise is not Publications Chair: Lee D. Erman, USC/Information Sciences Institute a legitimate scientific enterprise. …I believe that AAAI Mag. Editor: Alan M.Thompson, Jet Propulsion Laboratory the controversies have their natural cause in the Conference Chair: Jay M.Tenenbaum, SRI International type of knowledge our science reveals. They Program Chair: Robert M.Balzer,USC/Information Sciences Institute must be dealt with on the basis of substance Conf. Treasurer: Lester D. Earnest, StanfordUniversity and truth. So cherish the name Artificial Intel- Tutorial Chair: Frederick Hayes-Roth,RandCorporation ligence. It is a good name. Like all names of scientific fields, it will grow to become exactly Executive Council what its field comes to mean. Woody Bledsoe, University of Texas Bruce G. Buchanan, StanfordUniversity Lee D. Erman, USC/Information Sciences Institute Sustainable AAAI H. Penny Nii,StanfordUniversity NilsJ.Nilsson, SRI International Having decided that it may be desirable to have D. Raj Reddy, Carnegie-Mellon University a society of our own, we needed to establish Earl D. Sacerdoti,SRIInternational guiding principles for our society. The major Roger C. Schank,YaleUniversity concern was whether such a society could sur- DavidL.Waltz, University of Illinois Patrick H. Winston, Massachusetts Institute of Technology vive and become sustainable in the long term. It was agreed that each activity of the society would be self-sustaining. For example, we set the registration fees for the first AAAI conference around $60 because the budgeting Table 1: Founding Organizers of AAAI. showed that we would break even at that price by holding the conferences at universities. Sim- ilarly the membership fee was supposed to just cover the cost of AI Magazine and one or two support staff members. Any surpluses from year to year would go to an endowment. Annual income from the en- AAAI-80ProgramCommittee dowment would be used for outreach activities Chairman: RobertBalzer, UniversityofSouthern California and to cover losses if any. We were fortunate in Information Sciences Institute that, during the early years of our existence, Saul Amarel, Rutgers University AAAI had substantial surpluses primarily com- Jon Bentley, Carnegie-Mellon University ing from the highly successful tutorial pro- Woody Bledsoe, UniversityofTexas grams leading to an endowment of more than Ron Brachman, Bolt Beranek &Newman $5 million. In the more recent lean years we Bruce Buchanan, Stanford University have not been as successful. But by following EdwardFeigenbaum, Stanford University the basic founding principles we’ve set forth, it IraGoldstein, Xerox PARC is believed that we can be self-sustaining and Peter Hart, SRI International stay alive for the foreseeable future. Fredrick Hayes-Roth, Rand Corporation Carl Hewitt, MassachusettsInstituteofTechnology DougLenat,Stanford University Governance DavidLuckham, Stanford University We augmented the traditional structure of a DavidNitzan, SRI International president and council to enhance continuity Charles Rieger, University of Maryland and provide for smooth transition. The office Roger Schank, Yale University of the president had three members: past pres- Jay M. Tenenbaum, SRI International ident, president, and president-elect playing DavidWaltz, University of Illinois roles analogous to the roles of chairman, CEO, and COO. Staggered membership of the council also provided for preservation of continuity. Table 2. AAAI-80 Program Committee. All other appointments and terms were the re- sponsibility of the president.

10 AI MAGAZINE 25th Anniversary Issue

Activities of AAAI The primary activities of AAAI are similar to those of other societies: conferences and work- shops, journals and publications, membership and election of fellows. Much of the continuity from year to year was provided by the estab- lishment of an AAAI administrative office in Menlo Park, California, which prospered under the able management and successive leadership of three executive directors: Lou Robin- son, Claudia Mazzetti, and Carol Hamilton. AI Magazine was created to help the community to stay abreast of significant new research and literature across the entire field of artificial intelligence. It was modeled after Spectrum, the IEEE magazine. Under the able successive leadership of Alan Thompson, Bob Englemore, Ramesh Patil, Elaine Rich, Jude Shavlik, and David Leake, serving as the editors in chief, AI Magazine has surpassed many of our initial expectations and has provided the all-important links to the community. A key ingredient for success was Mike Hamilton, who serves as the managing editor providing the all important corporate memory. Perhaps the most satisfying aspect of em- barking on this enterprise was the enthusiasm and excitement and camaraderie of the AI researchers at the first annual conference of AAAI at Stanford. Even after 25 years, it still stands I remember having a kind of distant amazement that the USA out as one of the memorable conferences that could find so much money in such a short time. AI societies in I have ever attended. Most importantly it also Europe were run on shoestrings, almost entirely by volunteer aca- provides a snapshot of AI research in the Unit- demics, but AAAI had piles of cash from day one (well, actually maybe something like day three) thanks to the AI explosion and ed States in 1980. the trade fair at the early meetings. Nothing like that had hap- pened outside the USA. The First AAAI Conference —Pat Hayes (AAAI President, 1991–1993) The broad spectrum of the research topics cov- ered at the AAAI-80 conference was indicative tive research programs in various centers, there of the vitality and vibrancy of our field at that were not many papers in the areas of learning, time. There were multiple sessions in the areas speech, and robotics. of vision, problem solving, knowledge repre- The institutions represented at this confer- sentation. In vision, we had topics such as early ence were also interesting. Besides the usual AI vision processing, scene analysis, motion powerhouses of CMU, MIT, Stanford, and SRI, analysis, and robotics vision. In problem solv- we had multiple contributions from the Uni- ing, we had topics such as cooperating expert versity of Southern California Institute for Sci- systems, problem-solving end control, and in- entific Information, the University of Illinois at teractive problem solving. In knowledge repre- Urbana-Champaign, the University of Massa- sentation, we had topics of advanced knowl- chusetts, Yale University, the University of edge representation, applied knowledge Pennsylvania, and Rutgers, reinforcing the representation, and special issues in knowledge premise that AAAI can be a forum to bring all representation. In addition we also had ses- members of the community together, avoiding sions on program synthesis, theorem proving, the balkanization that can occur without a so- mathematical and theoretical foundations, ciety of our own. knowledge acquisition, specialized systems, The authors of the papers at the first confer- natural language processing, memory models, ence were also remarkable. They represented a and applications of AI. Individual topics cov- Who’s Who of AI, many of whom continue to ered within these sessions are provided in the be active to this day. What was unique about accompanying box. Surprisingly, in spite of ac- the attendees of the first conference was their

WINTER 2005 11 25th Anniversary Issue

Photograph Courtesy Carnegie Mellon University. Allen Newell.

genuine interest in the advancement of AI. gratifying. It shows that our community has Subsequently, during the AI bubble in the mid- matured and has a widespread following and 1980s, a large number of people adopted AI as membership who care about the advancement a quick way to become rich and famous. Dur- of AI. Starting from our modest beginnings we ing the AI winter, many of these visitors disap- have come a long way. A visit to our website peared, and now we seem to be back to the shows a wide range of activities and functions core group of researchers dedicated to the ad- we now provide for our membership vancement of our field. As we look to the future, we have a large un- There were other activities at AAAI-80 that finished agenda. The goal of creating artificial were equally exciting. There were a number of intelligences that reach or exceed human intel- invited talks and panels from the leaders of our ligence is still a long way away. We do not seem field on topics of immense interest to many of to have a coherent and coordinated vision and us. Unfortunately, we do not have a written plan to create a human-level AI. I hope the record of their comments. Perhaps in the fu- next generation of AI researchers, using AAAI ture, with advanced transcription and summa- as a launching pad, can formulate a plan of ac- rization technologies in hand, we will be able tion to get to our cherished goal. to keep a complete record of our views, agree- ments and disagreements, and expectations for the future. The program committee for the Raj Reddy is the Mozah Bint Nasser University Pro- first conference, given in table 2, started with fessor of Computer Science and Robotics in the a clean slate and had to make up the rules as School of Computer Science at Carnegie Mellon Uni- they went along. They deserve our respect and versity. He is a member of the National Academy of gratitude for the countless hours spent crafting Engineering and the American Academy of Arts and an outstanding program. Sciences. He was president of the American Associa- tion for Artificial Intelligence from 1987 to 1989. Reddy was awarded the Legion of Honor by President Conclusion Mitterand of France in 1984. He was awarded the ACM Turing Award in 1994. He served as cochair of The fact that we are celebrating the twenty- the President’s Information Technology Advisory fifth year of the founding of AAAI is indeed Committee (PITAC) from 1999 to 2001.

12 AI MAGAZINE