History of the Philosophic Institute for Artificial Intelligence
Collaborative research in artificial intelligence (AI) began at Notre Dame (ND) in 1962, peaked in the late 1960s, and continued in conjunction with other efforts until about 1980. The research was funded by a continuous series of grants from the National Science Foundation (NSF), which brought a total of $352,746 to the University. At one time or another, 26 ND faculty members were involved in this research, representing all five colleges of the University. Also involved were 9 PhD candidates and 9 professional people from other organizations. From 1965 onward, this work was centered within the Philosophic Institute for Artificial Intelligence. Following is a brief history of that group.
While completing his PhD at Harvard, Kenneth Sayre (KS) worked for M.I.T.’s Lincoln Laboratory on the SAGE (Semi-Automatic Ground Environment) program. SAGE was an air defense system based on M.I.T.’s Whirlwind I computer, in which many complex data-reduction tasks were performed automatically. Several pioneers of AI (Marvin Minsky, Oliver Selfridge, Edward Fredkin) were already working on this system when KS joined the effort. When he left M.I.T. for ND in 1958, KS brought with him an active interest in AI as a theoretical discipline.
By 1962, KS had secured a grant from NSF under the title 'Simulation of Mental Processes. Its objective was to compare in detail the status of mechanical pattern recognition projects with philosophic accounts of human recognition typical of the Aristotelian tradition and of modern empiricism. Early collaborators were ND Professors Joseph Bobik, Milton Fisk, and Frederick Crosson (FC). During the initial year of the grant, Bobik and Sayre co-authored "Pattern Recognition and St. Thomas' theory of Abstraction," Crosson and Sayre co-edited The Modeling of Mind, and KS began work on Recognition: A Study in the Philosophy of Artificial Intelligence.
When the grant was renewed at a higher funding level in 1963, negotiations began with the University to establish a center for the study of AI with its own budget and office space. In 1965, the Philosophic Institute for Artificial Intelligence (PIAI) was incorporated as a branch of the Center for the Study of Man in Contemporary Society (then headed by George Shuster, past President of Hunter College). KS and FC were appointed Director and Associate Director of the PIAI, respectively.
Work under this second grant was divided into two stages. First was an analysis of the concept of pattern, with specific attention to letter patterns, patterns of formal proof, and patterns of melodic line. The aim was to identify the factors that distinguish a letter from a meaningless squiggle, a proof from a series of disconnected symbols, and a melody from a series of incongruous sounds. The second stage was to begin formulating and testing programs to recognize such patterns mechanically on the University's UNIVAC 1107A computer.
In partial fulfillment of the first goal, KS completed Recognition (published 1965), and members of the Institute (James Massey, David Burrell, FC and KS) produced several papers on topics of pattern recognition and information theory. These papers were published in 1967 under the title Philosophy and Cybernetics. In addition, KS and FC began giving public lectures and making TV appearances on topics of cybernetics and automation.
In August 1965, trial runs were made of an automated hand-written-letter-recognition program. Results were promising enough to be presented as a demonstration of our work to the Advisory Council of ND's College of Arts and Letters in late 1965. By this time, ND’s sponsorship of the PIAI had broadened to include the Computing Center and the Department of Philosophy, in addition to the Center for the Study of Man.
Although initial results of the letter-recognition program were encouraging, considerable improvement was necessary before the project could be deemed a success. Additional funds were granted by the NSF in 1966 for continuation of the project under the title "Pattern Recognition and Formation by Machine." The express purpose of this grant was to enable further work on the letter-recognition program, and in addition to support (i) an attempt to explain certain pattern-formation phenomena of Gestalt psychology on a quantitative basis and (ii) an inquiry into the potential of information theory as a tool for resolving long-standing problems in the theory of perception. By way of preparation, FC and KS had previously completed formal courses in information theory (with J.L. Massey) and in recursion theory (with Thoralf Skolem).
Further preparation came with a two-week conference on information theory sponsored by the PIAI during August of 1967. Its purpose was to explore the application of information theory to outstanding problems of philosophy (e.g., the interaction between syntax and semantics) and to the immediate task of mechanical handwriting-recognition. Participants were FC, KS, David Burrell, Vaughn McKim, and Donald Mittleman (Director of the Computer Center) from ND, and invited speakers Milton Fisk (then from Indiana University) and Fred Attneave (psychology, from University of Oregon). Fred Atteave was the author of the ground-breaking Applications of Information Theory to Psychology: A Summary of Basic Concepts, Methods, and Results (Holt, 1959).
From this time onward, work on AI in the Institute proceeded along two parallel paths. During the summers, a group of programmers (graduate students Kerry Koller, Charlie Davis, Michael Crowley, and James Heffernan, and after 1967 Professor Vaughn McKim), under supervision by KS, developed the letter-recognition program through several versions of increasing sophistication. During the school year, KS and FC spend most of their available research time working on book-length studies that satisfied the theoretical mandate of the grant.
Work on KS's Consciousness: A Philosophic Study of Minds and Machines, published in 1969, provided guidance in designing the computer program then undergoing development. FC currently was editing Human and Artificial Intelligence, which came out in 1970. During this period, KS also published "Intelligence, Bodies, and Digital Computers" and "Information Processing and Mind-Brain Identity," while FC published "Phenomenology and Computer Simulation of Mind-Brain Behavior."
FC was appointed Dean of ND's College of Arts and Letters in 1968. Although he continued to publish on related topics, this marked the end of his direct involvement in the PIAI. Fr. David Burrell (DB), C.S.C., who had been with the Institute as Faculty Associate since 1965, became its new Associate Director.
By 1968, work on the letter-recognition program had progressed to a point where it was able to identify individual letters from previously segmented handwritten lines with about 90% accuracy. The process of isolating individual characters within the cursive line, however, remained problematic. We were faced with what became known in later handwriting-recognition literature as "Sayre's paradox" (e.g., www.idiap.ch/publications/vincia03c.bib.abs. html): a letter cannot be segmented before having been recognized and cannot be recognized before having been segmented. Another limitation of our results at this point was that the successfully identified letters all came from the handwriting of a single individual. More work was clearly needed to complete the project satisfactorily.
A new grant was issued by the NSF in 1969 under the title "Completion of Program for the Mechanical Recognition of Handwritten sentences." Celso Souza (Department of Electrical Engineering) meanwhile had devised a statistical technique for bringing contextual information to bear in the segmentation process. Part of the technique was to classify segments of the cursive line according to their most probable letter interpretations. This was done without prior specification of individual stroke sequences making up the letters, a key component of other handwriting-recognition programs under development at the time. The other part of the technique was a statistical filtering routine that eliminates all possible letter combinations that are infrequent or nonoccurrent in English text. This technique was ready for testing in late 1969.
Before realistic testing could begin, however, our sample set of handwritten lines had to be expanded. Although examples of handwriting are easy to come by, samples had to be encoded for presentation to an automated system in order to be usable (automated scanners were not then available). ND's Computing Center contributed the significant amount of computer time necessary to prepare the samples. Our resulting sample set consisted of phrases written by 10 different persons, amounting to 100 words containing all letters of the alphabet. The system was "tuned up" on samples from three of these penmen, and then fed the remaining samples for the initial test.
The initial test resulted in correct identification of 42% of the input samples. Other handwriting-recognition programs then under development had reported higher success rates (60% for one, 72% for another). As already noted, however, these other programs relied on sequences of stroke inscriptions as part of their input. Since ours did not, it was difficult to compare results in a useful fashion.
Further refinement of our program took place in 1970 and 1971. Improvements in our statistical filter (dealing with probabilities of letter sequences) suggested that it would be more meaningful to take words rather than letters as cursive units to be identified. What turned out to be the final test was conducted in late 1971. The same 100 word inscriptions served as our sample set (the program had no "memory” of past results). Sixteen words provided "tune up," and the remaining 84 were presented to the system for identification.
In this final test, the program correctly identified 66 of those 84 words, for a success rate of 79%. This not only exceeded the success rates of "competing" programs, but did so under more demanding circumstances (as stroke sequences were not provided as input). By the end of the project, we had produced the most successful handwriting-recognition program then available.
The final report of the project was written during the summer of 1972, and published in the technical journal Pattern Recognition in 1973 under the title "Machine Recognition of Handwritten Words: A Project Report." The project that began with the analysis presented in KS's Recognition was now complete.
Other ventures in AI had been contemplated in the course of this project. FC and KS often talked about working on a program for playing GO (an oriental game considerably more difficult than chess), of which both were enamored. There were excursions into automated logical-theorem-proving techniques, and a brief flirtation with automated music-composition. There was also some thought about the applicability of the analytic approach we had been following to philosophy of mind generally, the idea being that you don't really understand a mental process until you can either duplicate it mechanically or explain why it can't be so duplicated.
With the successful completion of the handwriting project, however, members of the Institute felt that a time of transition was at hand. Although conceptually challenging, computerized recognition systems then appeared to have little social significance. And this was a period of considerable social turmoil.
The U.S. incursion into neutral Cambodia in 1970, intended to impede the flow of supplies to the Viet Cong in their offensive against South Vietnam, had provoked wide-spread protests on the nation's college campuses. Although student response to the war was muted at ND by administrative action, there were many opportunities for discussion of the moral issues involved. In his dual role of priest and faculty member, DB was actively involved in these discussions. By the time the Vietnam War was winding down in 1972, there was consensus within the PIAI that our joint efforts should be redirected to issues involving ethical and social values.
In 1973, the PIAI submitted a large-scale proposal to RANN (Research Applied to National Needs), a different branch of the NSF from the one we had dealt with previously. The premise of this proposal was that value-conflicts in technological society tend to take shape either around tensions between commercial goals and environmental values or around tensions between individual values and values of society. Our test case for exploring the former was to be value-conflicts affecting decision-making in the power industry. As for the latter, we would look at value-issues regarding the use of medical technology in the treatment of disabled persons.
A group of 15 ND Faculty members had been assembled to undertake the research. Direct contacts with the power companies involved would be handled by Neil Schilmoeller (Nuclear Engineering), William Biles (Industrial Engineering), Michael McIntire (Environmental Law), and Robert McIntosh (Ecology). Medical technology and applications would be the responsibility of Stanley Hauerwas (Theology), Harold Moore (General Program), Charles Murdock (Law), and Armand Rigaux (University MD). Decision-making would be the responsibility of Vaughn McKim (Philosophy), David Burrell (Theology), Kenneth Jameson (Economics) and John Kennedy (Marketing). And matters of value theory would be engaged by Kenneth Sayre, Kenneth Goodpaster, and David Solomon (all Philosophy).
When notified that this proposal would not be funded, we began work immediately on two more modest proposals to yet a different branch of the NSF (the Office of Exploratory Research and Problem Assessment). One was entitled "Exploratory Study in Medical Ethics," addressing moral problems pertaining to the medical treatment of dying patients. This project was to involve Frederick Crosson (Dean, Arts and Letters) and Harvey Bender (Biology), along with Moore, Hauerwas, Murdock, Rigaux, Solomon, and Sayre (as listed above). After consultation with the NSF, the proposal was scaled back even further to a few months exploratory work (in 1973) involving only Moore and Sayre. This proposal received funding at a modest level. Following a summer of exploration, however, Moore left ND to become a corporation finance lawyer in New York. No further work in medical ethics was undertaken by the Institute.
Our second proposal to the NSF office of Exploratory Research was more fruitful. This proposal was entitled "Decision-Making in the Power Industry." Its main objectives were (1) to analyze conflicts between commercial, environmental, and ethical values in power industry decision-making, (2) to develop normative guidelines for resolving these conflicts, and (3) to work with the decision-makers involved in applying these guidelines. Organizations cooperating with this study were Commonwealth Edison of Chicago, Northern Indiana Public Service Company (NIPSCO), and the Environmental Statement Project at Argonne National Laboratory (near Chicago).
Principal Director was Kenneth Sayre (KS), who also directed the Value Theory Section of the study. Co-Principal Director was Neil Schilmoeller (NS), a nuclear engineer also responsible for coordinating the Corporate Decision Section of the study. Other ND participants were William Biles, David Burrell, Kenneth Goodpaster, Kenneth Jameson, John Kennedy, Michael McIntire, Robert McIntosh, and Vaughn McKim (disciplines listed above). Ellen Maher (Sociology, Indiana University at South Bend) joined the staff after the project began. Other outside participants were Mary Trainor (Valparaiso University), and Alasdair MacIntyre (Boston University).
The Corporate Decision Section of the study was charged with (a) providing a description of factual circumstances that influence policy decisions within the industry, and (b) analyzing the value presuppositions underlying these decisions. Data serving both purposes were acquired through questionnaires administered to upper level decision-makers at Commonwealth Edison, and through personal interviews with these executives and their counterparts at NIPSCO. The questionnaires were designed primarily by Ellen Maher (EM). Most members of the study-group were involved at some stage of the interviews.
The Value Theory Section participated in the analysis of value presuppositions indicated by the questionnaires and interviews. As its main task, however, this group was concerned to articulate potentially conflicting values behind company decision-making in ways enabling them to be weighed against each other, and to explore possible ways of resolving such conflicts. To this end, we drew extensively upon the major normative traditions informing our current culture, particularly those of Utilitarianism, Kantianism, and Judeo-Christian morality. The efforts of Kenneth Goodpaster (KG) and Vaughn McKim (VM) were central to this undertaking.
Major results of this study were (1) that growth is equated with progress in company thinking, (2) that its "bigger is better" ethos stems in part from pressure to make room for more executives on the corporate ladder, (3) that in cost-benefit analyses of new plant siting it is always assumed that new construction is needed (no growth in generating capacity is not an option), (4) that the industry's decision-making is dominated by an implicit utilitarian moral outlook, leading to a disregard for groups outside its customer base (e.g., future generations), and (5) that for practical purposes the relevant regulatory agency served as its conscience ("if the regulators let us get by with it, it must be ok").
These results are elaborated and documented in the volume Values in the Electric Power Industry (1977), edited by KS and containing essays by VM, EM, KG, and KS, as well as by Charles Murdock, Kenneth Jameson, and Alasdair MacIntyre. As the results may suggest, this volume was not uniformly well received by industry executives. Another volume resulting from this project is Ethics and Problems of the 21st Century (1979), edited by KG and KS, containing essays by KG and Alasdair MacIntyre along with others by moral philosophers not associated with the present study.
In view of these results (esp. (5)), a natural next step was to look at the value orientation of the regulators. In 1977 we receive a grant from the EVIST (Ethical and Human Values Implications of Science and Technology) branch of the NSF, under the title "Values and Electric Power Industry Regulation." Members of the Illinois Commerce Commission (ICC) had agreed in advance to cooperate with the study. Its key purpose was to determine the value orientations of these regulators with appropriately designed instruments and then to assess the extent to which the values in question serve the public interest.
The study was staffed by KS, KG, Peri Arnold (Government), James Stewart (Economics), John Lucey (Engineering), and Robert Rodes (Law), all from ND, and by EM (IUSB), Mark Lipsey (Claremont Graduate School), and Alasdair MacIntyre (Boston University). EM was primarily responsible for questionnaires to be used in the project (Allport's Study of Values, Rokeach's Value Survey, and one of her own devising). Rodes and Lucey handled various aspects of interaction with the ICC. Arnold and Stewart dealt with organizational aspects of the study. And KS assisted KG in matters of moral value assessment. All ND members of the group participated in interviews.
Value orientations of individual regulators were tested by questionnaire and interview, in much the manner employed in the previous study. As the present study progressed, it became clear that the main moral issue had to do with the extent to which regulatory commissions become "captured" by (aligned with) the industries they regulate. To test a common conclusion of academic literature on regulation that "capture" is widespread, we decided to evaluate the following hypotheses: (1) that most regulators with the ICC are not aligned with the industries they regulate, (2) that aligned and nonaligned regulators with the ICC hold significantly different personal values, and (3) that aligned and nonaligned regulators are guided by significantly different ethical principles.
Results of the study are reported in Regulation, Values, and the Public Interest (1980), coauthored by Sayre, Maher, Arnold, Goodpaster, Rodes, and Stewart. Regarding the three hypotheses, we found the first to be strongly confirmed. Most members of the ICC in the late 1970s were aware of the dangers of being "captured," and had succeeded in avoiding it. The second hypothesis was disconfirmed. Both aligned and nonaligned members were similar in value orientation to the populace at large (in the 1970s), as measured by the Allport and Rokeach tests. The third hypothesis also was disconfirmed, with both classes of regulators showing general agreement in their conceptions of the public interest and in the utilitarian principles that governed their decision-making. Also presented in this volume are examples of the questionnaires employed and a comprehensive explanation of the methodology followed in the study.
Finding these results in keeping with their own self-conceptions, several members of the ICC suggested that the study be extended to regulatory commissions in other parts of the country. By way of preparation, we submitted a proposal for a more general study under the title "An Ethical Analysis of State Regulatory Agencies." In the meanwhile, however, the National Association of Regulatory Utility Commissioners had officially declined to sponsor the project. The proposal for continuation was subsequently rejected.
By this time, the research group had begun to break up. KG (who was to have succeeded KS as Director of the Institute) was not granted tenure and moved on to Harvard. Kenneth Jameson and James Stewart also left for other institutions. DB decided to devote most of his research time to Islamic studies. And KS became more deeply engrossed in his work on Plato (his Plato's Late Ontology came out in 1983). No further applications for support of interdisciplinary research were forthcoming from the Institute after 1980.
Between 1973 and 1980, the group had used both 'Philosophic Institute for Artificial Intelligence' and simply 'Philosophic Institute' as titles in correspondence. From 1980 onward, reference to our original interest (AI) was officially dropped from the title.
The Institute has continued to receive office space and secretarial help through the good graces of ND's Department of Philosophy. And KS has continued to publish occasionally in the areas of both AI and applied ethics (see Bibliography). But most of his research effort after 1980 was devoted to Plato. In the present web site, his most recent work on Plato is represented under the category "Plato and the Greek Commentators," and recent work in applied value theory appears under "Energy, Environment, Economics, and Ethics."
Publications stemming from work in the Philosophic Institute (by date)
KS - “Intelligence, Bodies, and Digital Computers,” The Review of Metaphysics, Vol. 21, No. 4, 714-23.
Philosophic Institute | Plato | Energy | Opinion | E-mail | Philosophy Department