In this first article of our series dedicated to the brief history of AI, we will focus on essential achievements in this field in the pre-computer age period. The dominant method of research at the time was to look in nature for ideas for solving severe problems. In the absence of an understanding of the functioning of natural systems, the research could only be experimental. So the most daring of the researchers approached the creation of mobile automatons (pre-robots) as the first attempt to create artificial intelligence.
Grey Walter’s “Tortoise”
Born in the United States but educated in England, Walter failed to obtain a research fellowship in Cambridge and started neurophysiological research in various places over the world. Heavily influenced by the work of the Russian physiologist Ivan Pavlov and Hans Berger (the inventor of the electroencephalograph for measuring electrical activity in the brain), Walter made several discoveries using his version of EEG machine in the field of brain topography. The most notable was the introduction of triangulation as a method of locating the strongest alpha waves within the occipital lobe, thus facilitating the detection of brain tumors or lesions responsible for epilepsy. He pioneered the brain topography based on EEG machine with a multitude of spiral-scan CRTs coupled to high-gain amplifiers.
Walter remained famous as an early contributor to the AI field mainly for making some of the first mobile automatons in the late ’40s, named tortoises (after the tortoise in “Alice in Wonderland”) because of their slow speed and shape. These battery-powered automatons were prototypes to test his theory that a small number of cells can induce complex behavior and choice. As a very simple model of the nervous system, they implemented two neuron architecture by incorporating only two motors, two relays, two valves, two condensers, and one sensor (ELSIE had sensor for light and ELMER had sensor for touch).
ELSIE scanned the surroundings continuously with the rotating photoelectric cell until a light source was detected. If the light was too bright, it moved away. Otherwise, ELSIE moved toward the light source. ELMER explored the surroundings as long as it didn’t encounter any obstacles; otherwise, ELMER retreated after the touch sensor had registered a contact. Both versions of the tortoise moved toward an electric charging station when the battery level was low.
Walter noted that the automatons “explore their environment actively, persistently, systematically, as most animals do”. This is what happened most of the time, except when a light source was attached to ELSIE’s nose. The automaton started “flickering, twittering and jigging like a clumsy narcissus” and Walter concluded that this was a sign of self-awareness. Even though many scientists today believe that robots will not achieve self-awareness, Walter’s experiment succeeded in proving that complex behaviours can be generated by using only a few components and that biological principles can be applied to robots.
Subsequent developments, some remaining only in a theoretical phase, promised substantial improvements in the direction of intelligent behaviour, Walter trying to add “learning” skills – even if they were in a primary form, such as Pavlovian conditioning. For example, the incorporation of an auditory sensor and the whistle immediately before contact between ELMER and an obstacle will cause ELMER to subsequently perform an obstacle avoidance maneuver before contact occurs – if it “heard” the whistle. Although it seems that Walter materialised this attempt, it seems that the echo was not noticeable in the scientific world at that time.
John Hopkins’ “Beast”
Another well-known realisation of a mobile automaton is the “Beast” project from the ’60s of a team of engineers from Johns Hopkins University Applied Physics Laboratory, including Ron McConnell (Electrical Engineering) and Edwin B. Dean, Jr. (Physics). By having a height of half a meter, over 200 cm diameter, and a weight of almost 50 kilograms, “Beast” was built to perform two tasks only: explore the surroundings and survive on its own.
Initially equipped with physical switches, “Beast” moved “freely” following the white walls of the laboratory and avoiding potential obstacles encountered. When the battery level was low, “Beast” “looks for” a black wall socket and plugs it in for power. Without a central processing unit, its control circuitry consisted of multiple transistor modules that controlled analogue voltages; three types of transistors allowed three classes of tasks:
– Make a decision when activating a sensor, by emulating Boolean logic;
– Specify a period to do something, by creating timing gates;
– Control the pressure for the automaton’s arm and the charging mechanism by using power transistors.
A second version also received a photoelectric cell in addition to an improved sonar system. With the help of two ultrasonic transducers, “Beast” could now determine the distance, location within the perimeter, and obstructions along the path – thus exposing a significantly more complex “behaviour” than those of Walter’s tortoises. Performances such as stopping, slowing down or bypassing an obstruction or recognising doors, stairs, installation pipes, hanging cables or people through taking the appropriate actions are perhaps the most significant technical achievement of the pre-computer age.
In his response to Bill Gates, who predicted in 2008 that the “next” hot field would be robotics, McConnell humorously stated about their work from the ’60s: “The robot group built two functioning prototypes that roamed and “lived” in the hallways of the lab, avoiding hazards such as open stairwells and doors, hanging cables and people while searching for food in the form of AC power on the walls to recharge their batteries. They used the senses of touch, hearing, feel and vision. Programming consisted of patch cables on patch boards connecting hand-built logic circuits to set up behaviour for avoidance, escape, searching and feeding. No integrated circuits, no computers, no programming language. With a 3-hour battery life, the second prototype survived over 40 hours on one test before a simple mechanical failure disabled it.”
Ashby’s “Mobile Homeostat”
Indeed, the most intriguing prototype of care saw the light of day before the computer age was The Homeostat¹, created by W. Ross Ashby, Research Director at the Barnwood House Hospital in Gloucester, in 1948 and presented at the Ninth Macy’s. Conference on Cybernetics in 1952. The Homeostat contained four identical control switch-gear kits that came from WW2 bombs (with inputs, feedback, and magnetically driven, water-filled potentiometers), and each transformed into an electro-mechanical artificial neuron. The purpose of this prototype was extremely challenging for that time, namely to be an example for all types of behaviour – by addressing all living functions.
During the presentation, The Homeostat was able to perform tasks that indicate some cognitive abilities, i.e., the ability to learn and adapt to the environment. But the approach was at least strange: while other automaton of the time exhibited a dynamic character by exploring the environment, the goal of the Homeostat was to reach the perfect state of balance (i.e. homeostasis). This approach was intended to support the author’s principle of ultra-stability and the law of a variety of requirements. Based on the concept of “negative feedback,” the Homeostat approached incrementally the path between the current state and the final state of equilibrium, the steps representing the concrete responses of the automatons to changes in the environment (which affected the state of equilibrium). In detail, the principle of “Law of Requisite Variety” (as the author called it), stated that in order to break the variety of disturbances from the external environment, a system needs a “goal-seeking” strategy and a wide variety of possibilities to respond to them. For the animal world, a final goal like “no goal” was equivalent to achieving immortality. The part of “cognitive intelligence” embedded in the activity of automatons was precisely this “goal-seeking” approach, and, from a technical standpoint, “its principle is that it uses multiple coils in a milliammeter & uses the needle movement to dip in a trough carrying a current, so getting a potential which goes to the grid of a valve, the anode of which provides an output current”. But the audience was not very convinced of this principle, and, on the whole, its activity could be classified as a “goal-less goal-seeking machine.” It was Gray Walter, who called The Homeostat a “Machina sopor,” of which he said “fireside cat or dog which only stirs when disturbed, and then methodically ﬁnds a comfortable position and goes to sleep again,” in contrast with his creation, “The Tortoise,” called “Machina speculatrix,” which embodies the idea that “a typical animal propensity is to explore the environment rather than to wait passively for something to happen.” It was later learned that Alan Turing advised Ashby to implement a simulation on the ACE² computer instead of building a special machine.
However, The Homeostat received a significant comeback in the 1980s, when a team of cognitive researchers from the University of Sussex led by Margaret Boden created several practical robots incorporating Ashby’s ultrastability mechanism. Boden was fascinated by the idea of modeling an autonomous goal-oriented creature, arguing that the future of cognitive science is one based on The Homeostat.
The cybernetics of the ’60s are long gone, and the current possibilities of computer simulation are infinitely more capable than anything that could be imagined or created by the geniuses of those times, and within reach of any school student. Suffice it to say that the level of tropism of Tortoises is equivalent to that of a simple bacteria and The Beast equals the ability to coordinate of a large nucleated cell’s like Paramecium, which is a bacterial hunter; or that what was then presented as a continuous adaptation of responses to external stimuli is far from what we understand and have today in terms of learning – supervised or unsupervised. But evolution has not been just the result of the appearance of computer technology and its fantastic development. As I mentioned in the introduction, the history of AI overlaps the history of cognitive science. So at today’s AI level, achievements in multiple fields have contributed, including linguistics, psychology, philosophy, neuroscience, anthropology, and, of course, mathematics.
Simply put, even though in most cases it was agreed that it was a success, we can say that these mobile automatons of the pre-computer-era were nothing more than experiments before theoretical research and not during it. The rudimentary means of construction, the lack of a common language in the field and the non-adjustment between the model and the implementation mechanisms have often made the researchers of the time doubt each other’s achievements³; unimaginable today, when everyone understands that an self-driving car can anticipate complex accidents better than all the drivers involved or that a software robot crushes the world chess champion without even training by playing with someone other than himself.
- In biology, homeostasis is the state of steady internal, physical, and chemical conditions maintained by living systems.
- The Automatic Computing Engine (ACE) was a British early electronic serial stored-program computer designed by Alan Turing.
- With regard of The Homeostat of Ashby, the cyberneticist Julian Bigelow famously asked, “whether this particular model has any relation to the nervous system? It may be a beautiful replica of something, but heaven only knows what.”
- Steve Battle – “Ashby’s Mobile Homeostat”
- Margaret A. Boden – “Mind as Machine, A History of Cognitive Science”
- Margaret A. Boden – “Creativity & Art, Three Roads to Surprise”
- Stefano Franchi, Francesco Bianchini – “The Search for a Theory of Cognition: Early Mechanisms and New Ideas”