cyberneticzoo.com

pdf - Popular Electronics October 1962

Bionics

Bionic "Mouse." As mentioned earlier, RCA is working on a far more complicated moving-target indicator containing hundreds of neurons which operates on the same principle. But perhaps the most important piece of neural-bionic hardware to come out of the laboratories so far is a "bionic mouse" built by the Melpar Corporation. The "mouse" is housed in a small red plastic case about the size of a matchbox, mounted on wheels. A thin umbilical cord of control wires, suspended from a freely moving arm above, allows complete freedom of move ment and connects the mouse with its "brain" mounted in a relay rack.
Although the mouse looks like a toy, U. S. Air Force scientists working with it aren't playing. They are convinced that the mouse is the first step toward a completely new kind of thinking machine, as different from today's conventional computers as a superhet from a crystal set.
Not long ago, in a laboratory at the Wright Air Development Center in Dayton, Ohio, the author saw this "mouse" put through its paces. A technician placed it in a maze and flipped a switch. The mouse ran down alleys, turned corners, came to dead ends and backtracked, and tried other routes. Forty-five minutes later, after exploring scores of wrong turns and dead ends, it reached the end of the maze. The operator picked it up, and set it back at the beginning.
The second time round the mouse made fewer mistakes, and covered the course in about half the time. On the third attempt it ran through in eight minutes. Six tries later, it whizzed through the course in 45 seconds flat without a single error. The mouse had learned the maze, just as a live mouse would!
Bionic devices display true—though limited—intelligence in the animal sense. The bionic mouse has only 10 neurons in contrast to our 10 billion, but like an animal it can adapt to changing conditions and learn from experience. Change the maze, and it's confused—at first. But then it settles down and learns the new pattern.
A bionic "brain," in other words, can operate from generalized instructions. In the case of the "mouse," the only command was "learn to run the maze." Scientists call the mouse a self-organizing system which, on the basis of generalized instructions, figures out how to do the job. Human beings are self-organizing, too. A computer, on the other hand, has to be "programmed"—instructed in detail on every step. It must be told when to turn, when to store correct steps in its memory, and so on.
 

pdf - Generalisation Of Learning In A Machine – R. J. Lee (Maze learning machine) [I'm not sure where I sourced this from - It was downloaded quite some time ago].

SELF-SYNTHESIZING MACHINE R. J. LEE
Patent number: 3327291
Filing date: Sep 14, 1961
Issue date: Jun 1967   – find patent pdf here.

Although I've attributed this maze to Lee, the pictured version may actually have been made by E. B. Carne – see document ref below:

Title : ELECTRONIC REALIZATION OF FUNCTIONAL NERVE NETS.
Descriptive Note : Final rept., Dec60-Jan 62,
Corporate Author : MELPAR INC FALLS CHURCH VA
Personal Author(s) : Carne,E. B.
Report Date : JUN 1962
Pagination or Media Count : 80
Abstract : The self-organizing Binary Logical Network, a learning system using the reinforcement principal, is developed and used to formulate models of neurological subsystems. One of these systems, a maze running vehicle, has been constructed as an experimental and demonstration device. A detailed description of the maze system is provided. (Author)
 

MELPAR Maze:

From Man, Memory and Machines: An Introduction to Cybernetics by Corrine Jacker 1964. p71 [Note: If article is from Electronics World, then this is dated at 1963.] 

The bottom article has it that the word artron was a re-name of reron (relay neuron 1955) done in 1960.

“There are various kinds of artificial neuron. One, Artron (RH-Artificial Neuron), built by the Melpar Corporation of Falls Church, Virginia, is a component of a very interesting bionic computer that operates a mouse who can learn to run a maze. This miraculous mouse is now being studied by the Air Force at their Wright-Patterson Air Force Base near Dayton, Ohio.  The mouse begins its first time through the maze with no special instructions or tendencies to follow one part of the maze more than another.  It finds its way through by a process of trial and error, bumping into blind alleys, retracing its path, and beginning over and over again.
Each time the mouse begins the maze again it has learned something about how the paths are laid out, and gets through in less time. Finally, it can make its way through the maze with no errors in a matter of seconds. This device is only one of a series of maze-running machines that have been constructed.
One interesting refinement that is being worked on now will teach the mouse to associate colors with right and wrong turns. Green will indicate a right turn, red a wrong one. After the mouse “learns” what the colors mean, it could be put into any maze and, if the turnings are marked, run through it quickly and without error the first time.
The practical applications of such a machine are numerous. It is possible that a bionic control system may eventually be able to pilot planes and function in any way where it can be guided by signposts such as colored lights.”

Book references an article, but it may not be the source.
Gilmore, Ken, “Bionic Computers.” Electronics World, March 1963.  Pp 25-28, 63-64.

The mouse is tethered via an umbilical cord. The Brand name on the controller is “MELPAR”, and the inscription on the controller says “Artificial Nerve Network Maze Vehicle”.
 

Perceptrons, Regression,
and Global Network Optimization
John F. Elder IV
IPC-TR-92-10
9/4/92

(Extract from above pdf reference)
3.3. Smooth Logic as an Alternative
The path from perceptrons to polynomial nodes can be traced from the history of an early
“neuromime” company: Adaptronics, Inc. of McLean, Virginia. Adaptronics was formed in the early 60s by four young researchers from Melpar, Inc. (shortly after Air Force funding for “bionics” research jumped by a factor of 25 — see the survey of early techniques by Corneretto, 1960). One of the founders, R. J. Lee, inspired by (Ashby, 1952), had devised an artificial neuron called the “reron” (for “relay neuron”), descriptions of which were only privately published (Lee, 1954, 1955). The reron employed a noise source to switch between circuit states corresponding to six transfer functions, and biased the noise to reward (spend more time in) states with good error feedback. That is, the discrete logic function selected inside the reron was simply conditioned on the training data.
Use of a noise source, or random element, was apparently controversial to the scientific
community of the day. In fact, the idea first saw print as a Letter to the Editor in the July 1953 issue of (that "over the leading edge" monthly) Astounding Science Fiction! The letter was, however, recommended for publication by C. E. Shannon, who had explored a similar idea for "machine learning" (Shannon, 1951), though abandoned it 34 Perceptrons, Regression, and Global Network Optimization "… chiefly because it is rather difficult to trouble-shoot machines containing random elements. It is difficult to tell when such a machine is misbehaving if you can't predict what it should do!" (Shannon, 1953).
(Still, the controversial noise element was perhaps not really a factor in practical implementations, as the reron could have been set to the state most representative of the training data set.).
Given inputs a and b, the possible reron transfer functions were {0, a-, b-, ab- , ab,ab-}
(where “#- ” represents the NOT operation). In 1956, the other ten possible logic functions (including the notorious parity function, XOR(a, b)) were added at the suggestion of R. A. Kirsch (Gilstrap and Lee, 1960), and the node was renamed an "artron" for "artificial neuron". Shortly after starting Adaptronics, Lee's colleagues R. L. Barron and L. O. Gilstrap (living off savings for months) apparently discovered that bilinear multinomials (z = w0 + w1a + w2b + w3ab) could match all 16 logic functions of an artron. Not only that, but the polynomials provided a mechanism for interpolating and for using real-valued, rather than strictly binary, inputs. Hence, discrete computational elements were abandoned in favor of the richer real-valued polynomial nodes.

The “Bible” of the Biocybernetics Department
dnipt.irtc.org.ua/history.html
In 1964 Nikolai Mikhailovitch Amosov formulated a hypothesis on the information processing mechanisms of the human brain. Within this hypothesis he expressed his system-level observations on the brain’s structure and the mechanisms that are made operational by a human’s mental functions. Of principal importance was the fact that it was not the separate structures, mechanisms or functions (such as memory, perception, learning and so on) that became the simulation object, but the brain of the human as a social being – the brain of homo sapiens. Such was the main idea of the monograph “Modeling of Thinking and of the Mind”, published in 1965, which for a couple of decades became the bible for several generations of Department’s researchers (and not only for them).

Realization of N.M. Amosov’s ideas through computers and robots
The reaction of the Soviet scientific circles to the appearance of this monograph can in general be evaluated as a weakly aggressive one. On the one hand, by that time physiologists and the “great philosophers” of physiology were clearly fatigued by the wide discussion concerning the firmness of core postulates of I.P. Pavlov’s theories of higher nervous activity, that was hosted by the “Voprosy filosofii” (“Philosophical Issues”) journal and initiated by a well-known physiologist A.N. Bernstein. In this discussion the traditional physiologists suffered a considerable defeat, albeit not a complete one. Much became “allowed”, in particular, the statement on irreducibility of higher brain functions to lower ones was shaken. On the other hand, psychologists and the “great philosophers” of psychology were by that time thoroughly contaminated by widely accepted cybernetics concepts. Studies in which mental processes were explicitly treated as information processing started to appear, and ideological doctrines that dominated psychology for many years had already been weakened, but remained influential. (It is interesting to note that in those years, psychological publications were still found in bookstores under the heading “The Theory of Marxism-Leninism”.) Under these conditions the publishing of N.M. Amosov’s monograph was not met with official rejection (which would absolutely have not been possible five or ten years ago) and the Department’s work continued without interference. The fact that N.M. Amosov was a deputy of Supreme Soviet of the Soviet Union has apparently played its role: at the time, this was an important argument for “ideological acceptability” of his theories.

The ideas, which N.M. Amosov put forward in his book “Modeling of Thinking and of the Mind” were further developed in his subsequent works (“Modeling of Complex Systems”, “Artificial Intelligence”, “Algorithms of the Mind”, “Human Nature”).

Nikolay Mikhailovitch Amosov  (c2000)

Period of Robototechnics
The initial orientation of Amosov school research towards complex modeling of mental functions has determined to a considerable extent the “robot-technical” trend of subsequent research conducted in the Department.

It should be noted that in 1980s, the robot-technical subject area became quite popular in the USSR. Moreover, even direct instructions from the party and the government were given, concerning the particular topicality of research in the given direction. Thus a certain niche was created, in which many developers of issues within AI could find a spot. As time went by, however, it turned out that the country’s industry could not ensure effective use of high-intelligence devices offered by the scientists, thus the robot-technical boom slowly subsided. Nevertheless, for several years this line of research was yielding some quite interesting scientific results.

The signs of approaching crisis in neural cybernetics were stimulated by work aimed at obtaining results that could have a practical, applied value. For N.M. Amosov’s Department, this research turned out to be naturally connected to creating prototypes of autonomous mobile robots and to neural network control systems development. It should be said that in the initial stage of this “prototype epic” Nikolay Mikhailovitch Amosov strongly objected to such a digression to hardware. But he was successfully reassured, and as a result, a whole family of robots of this kind was developed and studied. A research member of the Amosov Department, E.M. Kussul, initiated and supervised this line of research (it was him who headed the Biocybernetics department after N.M. Amosov has vacated that position). In 1972-1975, the first autonomous transport robot in the USSR was created, TAIR (see Figure 3). Its development was completed with a successful presentation of the derived results during the Fourth International Conference on Artificial Intelligence in Tbilisi in 1975 (a short-length film about TAIR was shot and demonstrated at the conference). The robot demonstrated purposive movement in natural environment, obstacle avoidance and similar actions. TAIR was a three-wheel power barrow equipped with a system of sensors (rangefinder and tactile sensors). It was controled by a hardware-implemented neural network (the network nodes – special transistor electronic circuits; links between nodes – resistors).

TAIR is taking a “walk”
TAIR’s testing was conducted in the park near the Amosov Clinic for heart surgery. (At the time, the department of biocybernetics was located on the clinic's premises and N.M. Amosov headed simultaneously both the clinic and the Department). While moving, the robot was supposed to avoid obstacles, such as people, trees, park benches and so on. Coordinates of a point on the environment gave the target of robot’s motion. TAIR’s pilot research demonstrated that it was in principle possible to create an entirely autonomous robot operated by a hardware-implemented neural network. At the same time, it showed the overall complexity of organization of the robot's interaction with the natural environment, as well as the necessity of using trainable neural networks. The analysis of capabilities for creating teachable robots with neural network control systems was carried out using the MALYSH prototype created in 1979.

Later on several more robots were built, on which various motion control and natural environment interaction circuits were tested. In 1980-1981, a STAR transport robot prototype was built, which was used for working through control algorithms for a “large” transport robot created on basis of a commercial truck loader.

In 1984-1986, analysis and refining of algorithms for control of robot movement in a natural environment were continued using the MAVR prototype. This research work was ordered by the USSR Ministry of Defense and was directed at building an autonomous robot capable of moving under conditions of a complex uneven environment. Unique structural solutions provided MAVR with high maneuverability and reliable protection of control circuits, located inside barrel-shaped wheels. Environmental data came through rangefinders, optical and tactile sensors onto the hardware-implemented (vehicle-borne computer) neural network. Decision about the movement direction or other actions within the decision-making block was made upon processing the input information. The decisions taken were activated by appropriate actuating mechanisms.

The MAVR robot
The MAVR research marked the end of the “robot-technical” period of the N.M. Amosov School. The obtained results are presented in a composite monograph “Neurocomputers and Intelligent Robots”. It should be noted that this monograph summed up the main results of departmental research work during 1980s, and its contents were well beyond the robot-technical subject-matter itself. In particular, this monograph presented the findings of intensive research work directed at creating methods and practical techniques for using neural networks in building expert systems based on formalization of estimative expert knowledge, which is usually quite vague and difficult to formalize. Creation of a development program system VESTA was the practical outcome of this research work. VESTA supported expert's efforts on autoformalization of its own knowledge in the form of a neural network structure, capable of automatically transforming itself into a decision-making support system.

In 1951 Marvin Minsky teamed with Dean Edmonds build the first artificial neural network that simulated a rat finding its way through a maze.

They designed the first (40 neuron) neurocomputer, SNARC (Stochastic Neural Analog Reinforcement Computer), with synapses that adjusted their weights (measures of synaptic permeabilities) according to the success of performing a specified task (Hebbian learning) The machine was built of tubes, motors, and clutches, and it successfully modeled the behavior of a rat in a maze searching for food.

As a student, Minsky had dreamed of producing machines which could learn by providing them with memory "neurones" connected to "synapses"; the machine would also have to possess past memory in order to function efficiently when faced with different situations.

In 1951 the "machine" was born, consisting of a labyrinth of valves, small motors, gears and wires linking up the various "neurones". Some of these wires were connected up at random to the various memory banks in order to achieve a degree of causality of events. The reason such a machine had been put together was to try and find the exit from a maze where the machine would play the part of a rat whose progress would be monitored on a light network.

When the system was completed it was possible to follow all the movements of the 'rat' within the maze and it was only through a design fault that it was found more than one 'rat' could be introduced which would then interact together. After various casual attempts the rats started 'thinking' on a logical basis helped along by reinforcement of correct choices made and the more advanced rats would then be followed by the ones left behind. This first practical example, built by Minsky with the help of Dean Edmonds, also included numerous casual connections between its various 'neurones', acting like a sort of nervous system able to overcome any eventual information interruption due to one of the neurones failing.

Image courtesy Gregory Loan:

Gregory visited Marvin Minsky and enquired about what happened to his maze-solving computer. Minsky replied that it was lent to some Dartmouth students and it was disassembled. However, he had one "neuron" left, and Gregory took a photo of it.

An extract from an interview Jeremy Bernstein did with Marvin Minsky – The New Yorker, Dec 14, 1981

p69
For a while, I studied topology, and then I ran into a young graduate student in physics named Dean Edmonds, who was a whiz at electronics. We began to build vacuum-tube circuits that did all sorts of things."
As an undergraduate, Minsky had begun to imagine building an electronic machine that could learn. He had become fascinated by a paper that had been written, in 1943, by Warren S. McCulloch, a neurophysiologist, and Walter Pitts, a mathematical prodigy. In this paper, McCulloch and Pitts created an abstract model of the brain cells—the neurons—and showed how they might be connected to carry out mental processes such as learning. Minsky now thought that the time might be ripe to try to create such a machine. "I told Edmonds that I thought it might be too hard to build," he said. "The one I then envisioned would have needed a lot of memory circuits. There would be electronic neurons connected by synapses that would determine when the neurons fired. The synapses would have various probabilities for conducting. But to reinforce 'success' one would have to have a way of changing these probabilities. There would have to be loops and cycles in the circuits so that the machine could remember traces of its past and adjust its behavior. I thought that if I could ever build such a machine I might get it to learn to run mazes through its electronics— like rats or something. I didn't think that it would be very intelligent. I thought it would work pretty well with about forty neurons. Edmonds and I worked out some circuits so that —in principle, at least—we could realize each of these neurons with just six vacuum tubes and a motor."
Minsky told George Miller, at Harvard, about the prospective design. "He said, 'Why don't we just try it?' " Minsky recalled. "He had a lot of faith in me, which I appreciated. Somehow, he managed to get a couple of thousand dollars from the Office of Naval Research, and in the summer of 1951 Dean Edmonds and I went up to Harvard and built our machine. It had three hundred tubes and a lot of motors. It needed some automatic electric clutches, which we machined ourselves. The memory of the machine as stored in the positions of its control knobs—forty of them—and when the machine was learning it used the clutches to adjust its own knobs. We used a surplus gyropilot from a B-24 bomber to move the clutches."
Minsky's machine was certainly one of the first electronic learning machines, and perhaps the very first one. In addition to its neurons and synapses and its internal memory loops, many of the networks were wired at random, so that it was impossible to predict what it would do. A "rat" would be created at some point in the network and would then set out to learn a path to some specified end point. First, it would proceed randomly, and then correct choices would be reinforced by making it easier for the machine to make this choice again—to increase the probability of its doing so. There was an arrangement of lights that allowed observers to follow the progress of the rat—or rats. "It turned out that because of an electronic accident in our design we could put two or three rats in the same maze and follow them all," Minsky told me. "The rats actually interacted with one another. If one of them found a good path, the others would tend to follow it. We sort of quit science for a while to watch the machine. We were amazed that it could have several activities going on at once in its little nervous system. Because of the random wiring, it had a sort of fail-safe characteristic. If one of the neurons wasn't working, it wouldn't make much of a difference —and, with nearly three hundred tubes and the thousands of connections we had soldered, there would usually be something wrong somewhere. In those days, even a radio set with twenty tubes tended to fail a lot. I don't think we ever debugged our machine completely, but that didn't matter. By having this crazy random design, it was almost sure to work, no matter how you built it."
Minsky went on, "My Harvard machine was basically Skinnerian, although Skinner, with whom I talked a great deal while I was building it, was never much interested in it. The unrewarded behavior of my machine was more or less random. This limited its learning ability. It could never formulate a plan. The next idea I had, which I worked on for my doctoral thesis, was to give the network a second memory, which remembered after a response what the stimulus had been. This enabled one to bring in the idea of prediction. If the machine or animal is confronted with a new situation, it can search its memory to see what would happen if it reacted in certain ways. If, say, there was an unpleasant association with a certain stimulus, then the machine could choose a different response. I had the naive idea that if one could build a big enough network, with enough memory loops, it might get lucky and acquire the ability to envision things in its head. This became a field of study later. It was called self-organizing random networks. Even today, I still get letters from young students who say, 'Why are you people trying to program intelligence? Why don't you try to find a way to build a nervous system that will just spontaneously create it?' Finally, I decided that either this was a bad idea or it would take thousands or millions of neurons to make it work, and I couldn't afford to try to build a machine like that."
I asked Minsky why it had not occurred to him to use a computer to simulate his machine. By this time, the first electronic digital computer— named ENIAC, for "electronic numerical integrator and calculator"—had been built, at the University of Pennsylvania's Moore School of Electrical Engineering; and the mathematician John von Neumann was completing work on a computer, the prototype of many present-day computers, at the Institute for Advanced Study. "I knew a little bit about computers," Minsky answered. "At Harvard, I had even taken a course with Howard Aiken"—one of the first computer designers. "Aiken had built an electromechanical machine in the early forties. It had only about a hundred memory registers, and even von Neumann's machine had only a thousand. On the one hand, I was afraid of the complexity of these machines. On the other hand, I thought that they weren't big enough to do anything interesting in the way of learning. In any case, I did my thesis on ideas about how the nervous system might learn.

To date, I have not been able to locate a diagram of SNARC.  It's possibly in Minsky's thesis.