CIA report on Soviet artificial intelligence: primary documents

Problem: To assess artificial-intelligence research in the USSR.

Conclusions:

1. In the Soviet Union, substantial governmental and Party support and encouragement are being given to extensive studies on the theory of artificial-intelligence. Soviet scientists of high caliber are conducting artificial-intelligence studies and are exposing young students to the state of the art in this interdisciplinary field, laying a sound foundation for further advances.

2. The importance which the Soviet regime attaches to artificial-intelligence research is attested by the unusual freedom of open discussion allowed scientists working in the field, as reflected in the published literature.

3. Soviet research in the theoretical and hardware aspects of artificial-intelligence is now about as advanced as US work and can be expected to continue at a rate equal to or greater than that observed in the West. Significant theoretical achievements within the next 5 years are highly probable. When theories are converted into designs, Soviet engineers probably will be able to produce the equipment.

A highly significant discussion took place during June 1962 at a conference in Moscow on "The Philosophical Problems of Cybernetics." This meeting was co-sponsored by the Scientific Council on the Complex Problem "Philosophical Questions of Natural Science," the Scientific Council on Cybernetics, and the Party Committee, all of the Presidium of the Academy of Sciences, USSR. It was attended by about 1,000 specialists, including mathematicians, philosophers, physicists, engineers, biologists, medical scientists, linguists, psychologists, and economists, from many cities of the USSR.

A report of this conference stated that "the problem most animatedly discussed was the most disturbing of all, the problem of the technological operation of complicated psychic processes, the problem most frequently designated by the short and convenient although not, in our view, entirely correct formula of 'can a machine think?'"

The discussion brought out three questions as approaches to this problem. First, is it possible in general to reproduce with models the complex intellectual activity of man? Second, can a machine surpass man in the realm of intellectual activity and particularly in the performance of creative tasks, such as the formulation of new problems? Third, is it possible in principle to achieve the existence of consciousness in a machine similar to that exhibited by man?

A. N. Kolmogorov, world renowned mathematician, is optimistic about the chances of success in brain-modeling ventures. He called for the freeing of definitions of life and thought from arbitrary premises and for the redefinition of these concepts along purely functional lines. Kolmogorov believes that if the functional point of view toward life and thought is subscribed to, the conclusion is inevitable that replication of the organization of a system can be accomplished by organizing different elements into a new system which would have the essential traits and structure of the system being modeled. From this Kolmogorov concluded that a sufficiently complete model of a living being can in all justice be called a living being and the model of a thinking being, a thinking being.

[The report continues with the counter-argument from neurophysiologist P. K. Anokhin, who held that a machine's capabilities remain limited to problems and decision algorithms assigned to it by its human designer, and records that questions about machine consciousness specifically were argued most sharply of all at the conference.]

A somewhat different "machine that reads" has been developed at the Ukrainian Academy Computer Center under the direction of Candidate of Technical Sciences, V. A. Kovalevskiy. This system reportedly failed to recognize only 2 of 35,000 numbers produced with a portable typewriter, including half printed and otherwise distorted samples.

Tests were conducted with five digits (0, 1, 2, 3, 5), each written 160 different ways. Using 40 samples of each digit for training, and 120 for test, the machine achieved 83 to 89 percent correct recognition with the first algorithm. With "paralleling" of seven variations of a digit in the training process, 98 percent correct response was achieved.

Pattern-recognition techniques are employed at the Institute of Surgery imeni Vishnevskii, Moscow, to achieve rapid assessment of the area and seriousness of burns. The algorithm, for recognition of objects with many parameters, employs learning. The system is "trained" on case histories. When vital information, such as burn area and location and patient's age, is fed in, the computer identifies and stores symptoms and other factors. It also identifies objective criteria for forecasting the outcome of the illness.

Glushkov, V. M., Director. Machine intelligence, self-teaching and self-organizing systems, pattern recognition with learning, reading devices, recognition of meaningful sentences.

Ayzerman, M. A. Learning systems for recognition. Braverman, E. M. Learning systems for recognition, originator of the "compactness hypothesis." Feldbaum, A. I. Machine intelligence.

Bongard, M. M. Mathematical modeling of recognition processes, learning systems for recognition.

Kolmogorov, A. N. Parallel-operating automata, modeling thinking beings. Lyapunov, A. A., editor, Problemy Kibernetiki. General cybernetics.

[The full appendix names approximately 60 additional institutes and individual researchers across the Academy of Sciences, Moscow State University, the Ukrainian Academy of Sciences, and other institutions, with each entry specifying the researcher's documented area of AI-relevant work.]