The cybernetics dream
Several new concepts have emerged to form what we call cybernetics. The idea of creating something that resemble humans or reproduce human or animal learning capabilities has always been there, like a dream. However, in the 40s, technology was advanced enough to start building such machines. Let’s see what the pioneers of cybernetics had in mind and how they set the path to one day realizing this dream. This section will first present the work of two American researchers integrating two core ideas at the basis of cybernetics : the simulation of learning and adaptation and the imitation of animals by building machines. Then, it explains how cybernetics lead to the transition to the digital world.
​
​
​
1. Emergence of cybernetic
​
As shown in the previous section, technologies and sciences get developed by the second world war. And precisely, sciences become more specialized regardless to other field. Moreover, new results have been found in different fields such as mathematics, statistic, electrical engineering and physiology that are related to the same notion but received distinct name. Sometimes for a same notion, solutions or improvements have been found in one field but not in another. This non-sense was one of the main reason why the field of Cybernetics has been created by Wiener. Cybernetics implies that scientists that are specialized in one field, should also have some knowledges about the field of their colleges.
​
​
2. Thomas Ross’ electrical memory cell
​
The cybernetics dream is to build machines that can adjust their way of functioning according to the context thanks to what they have learned from past events. The cybernetics dream is to build machines that think.
Nature has always been a source of inspiration to engineers and has led to the creation of many well known algorithms (genetic algorithms, ant colony optimisation, artificial neural networks, ...). Thomas Ross’ idea also came from trying to replicate natural behaviour. Indeed, he tried making a machine that could simulate conditioned reflexes which are well known thanks to Pavlov’s dog experiment. This experiment shows that when a dog is about to eat its mouth starts to water (dominant stimulus). Whenever the dog hears a ring (neutral stimulus), nothing happens. However if the dog was to often hear a ring before eating, it would lead to the salivary response of the dog becoming conditioned to the neutral stimulus. Thus when simply hearing a ring, the dog’s mouth would start watering. This is precisely the behavior that Thomas Ross tried to imitate, based on the principle that neutral stimuli could acquire the power of generating the response of the originally ”dominant stimuli”. In order to do this, Thomas Ross created a mechanical machine with an electric circuit. The basic circuit consists of two electro-magnets and a metallic ball as shown on Figure 11. Normally a stimulus on Sx does nothing since the dominant stimulus is Sd . However if the two stimuli are active at once, the metallic ball will lift and go right, thus making contact and causing Sx to become a dominant stimulus as well. Thus the machine has been conditioned to consider Sx equivalently to Sd . In his paper [102], Thomas Ross shows that by cumulating this principle and chaining several of such cells, a machine could learn to find its way into a maze if it is built to give appropriate stimuli.
This is an example of a mechanical machine that simulates learning by making stimuli and responses discrete. Since then, scientists lean more and more towards the digital world. We took this interesting example as it stands between the analog and digital.
​
Fig. 11 Thomas Ross’ electrical memory cell for simulating conditional reflexes [102]
​
3. Wiener’s model of the brain
​
Norbert Wiener is one of those who also believed in cybernetics at that time. In [124], he studies the models that could have been used to create plausible thinking systems. Contrary to Turing,Wiener did not want to study if and when those machines could be created. His goal was mainly to set up the requirements he believed could not be overlooked for creating them. Feedback loops were a well-studied topic by that time, and therefore, part of his work uses them to create a model of what we would now call an android. However, the chapter that binds his book most closely to Turing’s paper is the fifth one, ”Computing machines and the Nervous System”, in which he depicted an analogy between those two kind of systems.
Let us first recall his model of the brain: it is a simplified version of the McCulloch model [74], for he considers neurons to be units that can transmit a binary information whose value is solely defined by their inputs, binary as well. Going further in this model, he proposed a central clock that orchestrates all operations, or at least a mechanism to wait for previous computations to finish before evaluating a neuron’s inputs at each stage of the pipeline.
To align the brain onto a computing system, it is thus sufficient to exhibit a system that can emulate all those biological operations, even much slower than we actually do. To create the neurons themselves is not much of a problem, using electronic circuits and digital gates. The goal of Wiener is thus to show that the functional assets required for the computation can all be emulated: not only the computing engine itself, but also the short-term memory, the long-term memory, etc.
To create the first type of memory, a technique is simply to have the information to keep looking within the computing engine by propagating it back and forward at each clock tick withing the graph of neurons. For practical considerations,
it is mandatory that the short-term memory is error-free, and this could already be achieved using telegraph-type repeaters in the fifties.
The long-term memory could also be emulated using this kind of device, as some concurrent work was already able to store thousands of bits without errors over more than a million writing cycles. The magnetic tape was also being developed at that time, leveraging the problem of mass storage for a human lifetime. To explain its brain counterpart, Wiener had no choice but to adopt the most trusted idea circulating around, which was to consider that some neurons get devoted to permanent storage over the life of a human being. Because no new neurons are created, this was also used to explain senescence for some part.
This model, although far from perfect, can be simulated on a machine using a finite amount of states. The proposed model creates only fixed patterns, or reflexes, within the learning entity, and nothing prevents a machine from working solely from those patterns, as it anyways is completely defined by the logic it runs. These are the kind of reflexions required by Turing to respond to the toughest arguments he received.
The final remark of Wiener is that any computing system consumes energy. This is obvious for the computers themselves, but brains do so as well, and also have to dissipate some heat created in the process using the surrounding blood. The remark positions brains and computers as two systems for managing information. If not for Wiener, Turing has evolved in a space where the very idea of intelligent machines was rising. His paper democratized cybernetics, and boosted research in this field for decades, though no machine was any close from passing his famous test.
​
​
4. Transition to the digital world
​
All in all, the cybernetics dream comes from the idea of imitating the animal’s capability to learn and adapt. This is done by transitioning from an analog world to the digital world. Indeed, from that moment on, machines are built from models based
on an ensemble of simple units with binary discrete states.