Marvin Minsky, who was the pioneer of artificial intelligence
Marvin Minsky
The year is 1966 and the place is the Artificial Intelligence Lab of MIT in Boston. Marvin Minsky, founder of the laboratory and professor of mathematics, proposes to his student Gerald Jay Sussman to work together to find out if it is possible to teach a machine to see by connecting a video camera to a computer. According to today's canons, it is a naive experiment - described in depth by Simone Arcagni in his The eye of the machine (Einaudi) -, imagined as if it were sufficient to connect an electronic brain (the computer) and an electronic eye (the video camera) to obtain as a result a computer system capable of seeing .At the same time, however, it is an experiment perfectly consistent with the interpretation that Minsky gives of human intelligence , and consequently of what is the way to equip computer systems of human qualities. For Minsky - and for his associate, the computer scientist and educationalist Seymour Papert - there is in fact no real difference between human beings and machines. Human beings, in their view, are nothing more than "flesh machines", whose brains are made up of a myriad of semi-autonomous but non-intelligent "agents". To simplify a lot, intelligence arises from the connections between these agents and the actions they carry out.
It is a theory that will later be developed in the fundamental text The society of the mind, but which was certainly already present for a long time first in Marvin Minsky's theory and which in fact directly influenced what, ten years before the experiment to equip machines with sight, had been a founding event in the study of artificial intelligence.
Childhood and studies
Born in New York in 1927, son of the eye surgeon Henry Minsky and the Zionist activist Fannie Reiser, Marvin was fascinated by the sciences and in particular from electronics, which he studied first at the Bronx High School of Science and then at the Philips Academy in the state of Massachusetts. After a period spent in the Navy during the Second World War, Minsky began studying mathematics at Harvard, later earning a doctorate at Princeton.In short, Minsky's academic career was one of all prestige, but the study of mathematics satisfies him. In fact, there are other topics that attract him much more, but which he has not yet managed to frame. It's not genetics, which he deems interesting "but not deep enough." It's not physics, which is only "moderately attractive" instead. Finally, Minsky understands what it is that he wants to study for the rest of his life: intelligence itself and its "desperately profound" problems, as he will explain to the New Yorker in 1981: "Now I can't imagine anything else worth to do” .
The idea of artificial intelligence
Intelligence, of course, does not necessarily mean studying only human intelligence. And this is where John McCarthy comes into play, his colleague at Princeton who in 1956 gathered a dozen scientists, including Minsky himself, at Dartmouth college (New Hampshire) to work on what they thought was a not so ambitious project: give life to a real artificial intelligence (a term that was coined on that occasion). As they wrote in their presentation , the goal was to study "how every aspect of learning or any other characteristic of intelligence can, in principle, be described in such a precise way as to make it possible to build a machine capable of simulating it" . A machine, therefore, capable of "using language, creating abstractions and concepts, solving various types of problems reserved for human beings and self-improvement" .To do all this, they calculated that it would be necessary to work all together for a couple of months. Obviously, they failed in the enterprise (which has not yet been achieved today), but their failure did not discourage Marvin Minsky, who still in the seventies had maintained his optimism unchanged. “In the space of three or eight years, we will have a machine endowed with the same general intelligence as the average human being - indeed Minsky announced -. A machine that can read Shakespeare, polish a car, deal with corporate politics, tell a joke and argue. At that point, the machine will start training itself at incredible speed. In a few months he will reach the level of genius, and a few months after that, his powers will be incalculable.”
Again, Minsky's optimism was decidedly misplaced. But that of having generated unreasonable expectations towards artificial intelligence is certainly not the most serious accusation that has been leveled against it. In 1969, in fact, Minsky together with Seymour Papert published the book Perceptrons , in which he frontally attacked the work done by another great pioneer of AI, Frank Rosenblatt , who had created a machine – the Perceptron , to be precise – which represented a rudimentary example of a neural network and of what, decades later, will evolve into deep learning.
The essay by Minsky and Papert mathematically demonstrated how neural networks were able to perform only some elementary operations (those linearly separable); while there was no hope that they would complete more complex tasks (at this link, a very brief explanation of the problem discovered by Minsky). Of course, it was admitted in the book, adding more layers between the input and output neurons would theoretically be possible to solve more difficult problems (as in fact will happen in the case of deep learning), but at the time no one had any idea of how to train them and, therefore, even that potential development was completely useless.
Winter is coming
The dismissal of Rosenblatt's work by a character known as Minsky – motivated, according to gossips, even by the fact that the two were competing to access funding from the US military - caused the sudden interruption of all studies on neural networks, giving life to what has gone down in history as "the winter of artificial intelligence" .On the other hand, Minsky was a convinced supporter of the opposite path to that followed by Rosenblatt to achieve true artificial intelligence. Instead of using data and neural networks to allow the machine to learn autonomously, according to Minsky an artificial intelligence could arise only by instructing it from above, that is, by providing it with all the rules necessary to complete its task: it is a logic " creationist" , better known as symbolic . This system worked in exactly the opposite way to the evolutionist one: to translate from Italian to English, for example, it would have been necessary to provide the computer with all the grammatical rules and vocabulary of the two languages, and then ask them to convert a sentence from a different language to the other.
The symbolic model has achieved important results (for example, teaching a machine to play chess), but it is certainly not what led to today's artificial intelligence revolution, which is instead it is accomplished thanks to the painstakingly made progress of the line of research that Marvin Minsky had hindered for decades.
However, all this must not lead us to underestimate the role of Minsky, to whom we owe fundamental contributions in the field of cognitive psychology, symbolic mathematics, robotic manipulation and computer vision. Indeed, before denying that trend, Minsky had been the inventor of the Snarc, considered the first neural network in history and made up of over 3000 thermionic valves and some components taken from the B-52 fighter-bomber. The artificial intelligence laboratory founded together with McCarthy in the early 1960s instead became known for attempts to create a model of human perception and intelligence, for the first prototypes of robots and for the construction of a robotic hand equipped with sensors tactile.
Giving an enormous contribution to overcoming the limits of Minsky's work, recovering the work done by Rosenblatt and others, was another name of fundamental importance in the field of artificial intelligence: Geoff Hinton . In the 1980s, Hinton was in fact a young Carnegie Mellon professor who – by building a complex system of artificial neurons with multiple layers – was among the first to demonstrate how Minsky's perplexities could be overcome.
In 2006, thanks to the collaboration with Yann LeCun (now head of artificial intelligence research at Meta) Hinton developed the technique called deep learning. A few years later, around 2012, researcher Andrew Ng explained to Google founder Larry Page how deep learning could forever change image recognition, natural language understanding and much more, taking the path that led to to the results that today are there for all to see. Ironically, Andrew Ng's project was dubbed “Project Marvin” .
