My research on a competence theory for intelligence has concentrated on three fields: inheritance, temporal reasoning, and counterfactual reasoning. My theory of inheritance reasoning unifies path-based and model-theoretic approaches, defining a single framework that subsumes most previous approaches and providing the first tractable algorithms for skeptical and credulous on- and off-path inheritance. An anonymous reviewer for Artificial Intelligence stated that my work ``significantly enhances the understanding of inheritance beyond the path-based approaches developed by earlier researchers.'' In the field of temporal reasoning, Motivated Action Theory---developed jointly with Leora Morgenstern---integrates causation as a structural primitive, solving the infamous Yale shooting problem and other benchmark problems and yielding perhaps the most thorough characterizations of the underlying issues in the field. Finally, my work on counterfactual reasoning explores the deeper issues that relate counterfactual, causal, and explanatory reasoning, exposing principles on which these forms of reasoning are based. The clear characterization of these principles allowed me to describe new and fundamentally different approaches to these problems, indicating misconceptions inherent in earlier work and presenting solutions to several open problems.
[References: Journals # 1-4; Conferences # 2-7; Other # 3, 4; Internal # 1, 2;
Student Publications: Conferences # 5 ]
My work on the semantics of sharing provided the first demonstration of the relationship between delegation (in Hewitt's actors languages) and inheritance (in traditional object-oriented languages). It was nominated for the Best Paper award at OOPSLA '87. I subsequently used this result to write what is widely regarded as a seminal paper on the semantics of sharing in object-oriented languages (together with Henry Lieberman of MIT and David Ungar of Stanford) and to build a new object-oriented idiom based on these insights (with Stanley Zdonik of Brown). At a recent OOPSLA, I was identified as "Mrs. 'Delegation is Inheritance'."
[References: Journals # 12; Conferences # 1, 2, 14; Other # 1, 2, 5; Internal # ;
Theses: Sc.M # 1 ]
Together with the students in my "AP" research group, I have designed, built, and analyzed a series of robots whose cognitive capabilities rest on non-traditional architectures. Each of these projects relies on non-modular-functional (or "postmodular") principles that violate traditional CS dogma.
This work ranges from engineering -- the construction of robots -- to philosophy -- implications of these architectures for theory of mind. The robotic achievements all involve bridging the gap between the low-level problems traditionally associated with robotics and the higher-level issues of cognition. More recently, I have turned to the question of how this alternate understanding of cognitive science informs the philosophical questions of intentionality and mind. Robotic work such as ours forces a reevaluation of such basic issues in that field.
In one project, a robot performed a cognitive task -- reading and understanding a map -- by directly reusing its underlying physically interactive capacities, without recourse to traditional AI techniques. This robot learned to interpret directions by imagining that it was acting in the world. In another, a society of robots learned a shared communication pattern -- a sort of proto-language -- through performance of a shared goal. Communication -- including simple compositionality -- was forced on the robots by their task environment. Yet other projects involved robots able to converse about their environments, a biologically inspired active vision system, and a developmental approach to visuo-motor coordination to learn reaching behaviors.
In each of these projects, the common theme is the rejection of traditional compositional architectures (as are prevalent in computer science) and the reliance on more interactive, body-based, and opportunistic strategies to bootstrap an unprecedented level of cognition directly from physically interactive subsystems.
[References: Journals # 6, 10; Conferences # 8, 11-13; Other # 8; Internal # 3, 4, 10;
Theses: Sc.M # 2, 4-6; Ph.D # 1;
Student Publications: Journals # 1; Conferences # 1-4, 6-14, 16, 17; Other # 1-3; Internal # 3 ]
Just as agents and computational systems are better understood as communal entities, so too information. The Haystack project (co-led with David Karger of MIT) brings to an individual's information workspace the power of modern information management tools. At the same time, the focus on a single individual allows a level of customization, adaptation and contextualization not possible for more global information spaces. At Haystack's heart is the notion of personal informational context: the network of associations and interactions within which a piece of information exists for its owner. Building on the individual information-space, Haystack is exploring the ways in which this system can be used to facilitate collaboration, expert identification, and the fulfilling of other information needs. Squeal (Spertus) provides infrastructure tools for exploring the basic structure of an information space and on which a variety of information management tools can be built.
[References: Conferences # 15-18; Internal # 7, 11;
Theses: Sc.M # 8, 9; Ph.D # 2;
Student Publications: Journals # 2; Conferences # 26, 30; Other # 4; Internal # 4 ]
I was involved in the founding and early leadership of two major projects at the MIT AI Lab: Cog and the Intelligent Room. Each of these projects played a significant role in the establishment of a new field of international research. In both cases, my students went on to assume leadership roles in these projects. My own formal affiliation with both projects ended when I went on sabbatical at the end of 1997.
Cog is an upper-body humanoid robot that is intended to mimic a child's development. It relies on architectural approaches that defy computer science's conventional wisdom. At the time of its conception, Science magazine said that "Cog's ambitious agenda is drawing applause mixed with guffaws" and called "Stein's next offspring...a leap ahead in robotics". (Science 264:1080-1082) The Cog vision is outlined in Brooks and Stein 1996, a paper that the editor of the journal Autonomous Robots called "the clearest statement of philosophy and goals for the next generation of autonomous robots that exists in print at this time." (G. Bekey, pers. corr.)
[References: Journals # 5; Conferences # 9; Internal # 6;
Theses: Sc.M # 5, 6 ;
Student Publications: Conferences # 17, 22, 23 ]
The Intelligent Room is a project that integrates research in human-computer interaction in much the same way that Cog attempts to pull together the whole of research in human cognition. Like Cog, it relies on interactive and communal architectures rather than conventional CS approaches. Its goal is to create an environment capable of interacting intelligently with its occupants. My students and I worked on software agents and information infrastructure, including Michael Coen's SodaBot platform, which recently spawned MetaGlue. The Intelligent Room Project was one of the seminal works spawning the Intelligent Environments research area.
[References: Journals # 5; Conferences # 9; Internal # 6;
Theses: Sc.M # 3, 7 ;
Student Publications: Conferences # 18-21, 25, 27-29, Internal # 2 ]
This project is a critical analysis of the historical, sociological, and scientific framework of computer science. Bringing to bear some of the techniques of critical studies on mainstream computer science, it calls into question the very foundational metaphors of our field.
Why should a computer scientist care? Because we as a discipline are hamstrung by our history. The current flight from academia -- the so-called "brain drain" -- is only the most dramatic symptom of the radical disconnect between our ostensible conceptual framework and our everyday practice.
The central thesis of this project is that traditional computational thinking is ill-suited for a wide range of current -- and apparently computational -- problems. This work poses a profound philosophical challenge to our disciplinary roots. I have enacted this foundational analysis by following through its ramifications in the introductory classroom. That project is important both as a curricular contribution (see below) and as a concrete realization of the underlying claims about computational metaphors.
[References: Journals # 14; Conferences # 19; Internal # 13, 14]
In the past several years, I have developed an innovative curriculum and textbook for introductory computer programming. The focus of this curriculum is the kind of interactive, concurrent, dynamic, embedded, distributed computation that increasingly typifies modern systems. This work builds directly on much of my earlier research, especially in Cognitive Architectures and Computational Metaphors, but has received recognition independent of these efforts. A recent international working group on innovative CS Education says, "Any discussion...would be incomplete without citing...Stein's radical departure from the traditional CS1 curriculum." [n1]
In the fall of 1999, over a thousand students at 18 colleges and universities in the US and internationally will use the beta version of my textbook in their introductory CS courses. My curricular materials are available on the world-wide web from http://www-cs101.ai.mit.edu/; the textbook is currently being provided in pdf by Morgan Kaufmann (who will publish it next year) at ftp://ftp.mkp.com/IPIJ/manuscript/
This project has garnered significant attention in recent years. The Rethinking CS101 Project has been awarded a Educational Innovation grant by the National Science Foundation (CISE EIA). Microsoft donated a teaching laboratory. Sun Microsystems named me a Java Technology Hero. Most recently, I was invited to give a keynote address at ITiCSE '00 (the International Conference on Innovation and Technology in Computer Science Education), ACM SIGCSE/SIGCUE's European conference.
[References: Books # 1; Journals # 8, 11, 13; Conferences # 14; Other # 6, 9; Internal # 5, 8, 9 12;
Theses: AUP # 2, Sc.M # 11]
July 1997, "Preaching What We Practice: How AI is Changing the Concept of Computation," invited talk at AAAI '97, Providence, Rhode Island.
April 1998, "Reconceptualizing Computation", Keynote Address, Consortium for Computing in Small Colleges Third Annual Northeastern Conference, Fairfield, Connecticut.
April 1998, "Challenging the Computational Metaphor: Implications for How We Think", W. Ross Ashby Memorial Lecture of the International Federation for Systems Research, Plenary Address at the Fourteenth European Meeting for Systems and Cybernetics Research, Vienna, Austria.
April 1999, "Cultural Revolutions in Computer Science", Conference on Diversifying the Culture and Curriculum of Science and Women's Studies, Kingston, Rhode Island.
July 2000, Keynote Address, Fifth International Conference on Innovation and Technology in Computer Science Education, Helsinki, Finland.
A full list of invited lectures appears in my CV.
The American Association for Artificial Intelligence (AAAI) is the premiere American scientific society in the field of Artificial Intelligence. I have served in a variety of leadership roles within the AAAI. From 1992 to 1996, I co-chaired and then chaired the standing committee responsible for the AAAI Symposium series, its incubator program for workshops in emerging research areas. In 1995, I was elected to AAAI's twelve-member Executive Council. Most recently, I represented AAAI at a CRA Summit of Women in Computing Leaders.
In 1993, I founded and ran the first Robot Building Laboratory at AAAI's national conference (together with David P. Miller and a team of MIT students). This effort, which has become an annual conference event, was the first to widely disseminate simple robotic technology to the national and international AI community. As a result of this program, many institutions have begun to apply their AI research using simple robots or to integrate these robots into their curricula.
I have served on almost two dozen program committees in the last decade, spanning a wide range of CS subdisciplines--from logic to embodied AI to programming languages--and am currently an Associate Editor of the journal Computer Science and Information Management.
Other leadership activities are documented in my full CV.