The advances in artificial intelligence obtained in recent years are unprecedented: ever more powerful computers and the possibility of having huge amounts of data, thanks to the Internet, have made it possible to create very elaborate software, which in the coming years could significantly change our relationship with computers, machines and the world in general. Some small pieces of these advances and innovations are already part of our lives, from personal assistants on smartphones like Siri to cars that drive themselves, through sophisticated software like AlphaGo of Google, which last year learned to play the complicated Chinese board game "go" better than anyone else, beating a world champion. But while it is true that our experiences increasingly intersect with "intelligent" software, we are still a long way from creating a true artificial intelligence (AI), able to think and behave like a human being, if not better.
Define the AI
There is no single definition of artificial intelligence and there is not even a broad consensus among researchers and computer scientists on how it can be defined, because the concept includes a large number of topics ranging from pure computer science to neurology, passing through studies on how our brain works. Generally speaking, we can say very generically: artificial intelligence is the science that deals with how to create intelligent machines, and that has found in the possibilities offered by computer science the most practical and probable way to obtain such a result. This area of science is closely linked to the even wider one that has long been trying to answer the question of questions: how does human intelligence work? Intelligence discoveries could lead us to develop the best possible AI, but according to other researchers, the opposite could happen: by developing an AI we could find out things about how our brain works.
Simplifying, intelligence is the set of psychic and mental capacities that allow us to think, understand actions and facts and know-how to explain to them until we develop abstract models starting from reality. These processes lead to the ability to obtain a result of some kind, with various levels of efficiency depending on the case. L ' intelligence is almost always referred to human intelligence, the one of which we have the knowledge and direct experience, and this complicates our ability to imagine different bits of intelligence, maybe that might be more suitable for the development of an AI.
Intelligence and software
The particular mechanisms have been identified that are the basis of intelligence. Drawing inspiration from their operation, it was possible to create computers that imitate some of these mechanisms. The problem is that to date we have not yet managed to imitate them and integrate them all, so the AI systems we have are incomplete. A software can, therefore, imitate the mechanisms necessary to win a game of "go" or to drive a car automatically respecting the code of the road, improving these abilities and becoming "intelligent" in a broad sense.
In decades of research, different solutions have been tried to reach a real AI. Basically two approaches were chosen: one consisted in observing human behavior, the way we reason and behave, to build software that imitates our logical processes as much as possible; the other is more creative and expects to start from the problems that reality poses and on the basis of these to have AI develop its method of behavior. The two approaches often intersect and one does not necessarily exclude the other, also because the designers are still human beings, with a way of thinking and reasoning that is reflected in the design of the AI.
Where does the idea of AI come from?
The prospect of being able to one day create the machine that can imitate human behavior has emerged in many historical periods, combining mythology, alchemy, the invention of automata and science fiction. But it was the British Alan Turing in 1950 who put together and formalized many of the concepts underlying artificial intelligence, as we understand it today. In his Calculating Machines and Intelligence, he listed the requirements to define a machine as "intelligent". Turing elaborated on the concept of a test, which today bears his name, in which artificial intelligence is revealed as such only if it succeeds in convincing those who are using it to deal with a person and not a machine.
The problem with the Turing test is that it allows a partial evaluation by the observer: a machine that surpasses it can be considered intelligent, but at the same time it cannot imitate a human being in everything and his way of thinking. According to some observers, in 2014 software passed the Turing test pretending to be a 13-year-old boy of Ukrainian origins, who asked his interlocutors understanding not to have much command of English, to be a little ignorant and not to follow always the conversation linearly.
In recent years, evolutions of the Turing test have been developed to mitigate the fact that many people conclude that a system is intelligent even if it maintains a very low level of conversation. Some propose to subject the AI to tests in which meaningless questions are included such as "Do footballers with wings on their feet score more often?", Which requires much more complex elaborations to answer, without falling into the trap.
Will there ever be an artificial intelligence similar to human intelligence?
The ultimate goal of AI understood as a science, is to create software that can achieve goals and solve problems in reality as a human being would. The most optimistic think that one day it will be possible to get a car with the necessary skills to think independently. The problem is that, despite the progress of the last few years, many pieces are still missing to reach a goal of this kind: new approaches and ideas are needed, which still have to be imagined.
A fascinating idea, which not by chance often occurs in science fiction, involves the creation of a basic AI that mimics the cognitive processes of a child. An AI of this kind could then learn new things, just as it happens during growth, gaining capacity and greater autonomy of thought. Maybe one day someone will succeed, but even in this case they lack the pieces: we are not yet able to create a machine that learns from physical experiences, as well as programs that can effectively interpret the language to understand in depth what they are reading.
Why it matters a lot?
Computer science is the area in which more solutions have been experimented for the creation of artificial intelligence, especially since the second half of the twentieth century. In purely theoretical terms, large computing capacities are not needed to implement an "intelligent" program and, according to some researchers, probably the computers of 30 years ago were already fast enough, and it could not simply program them to create the AI. The progress of the last few years, the possibility of keeping thousands of computers working together on the network and the huge amount of data available via the Internet have been crucial for the evolution of AI towards increasingly intelligent systems, albeit with limited purposes.
Companies like Google, Facebook, Amazon, Uber, and various car manufacturers are investing a lot of resources and money to produce artificial intelligence, at least in the broad sense of the term. Even if in practice it is still impossible to create a computer that thinks like a human being, you can still get some kind of intelligent and skilled software to perform particular tasks. Thanks to neural networks and deep learning ( here we tell you what they are ), Google has, for example, improved its Translator making the automatic translations from different languages such as English, Spanish, French, Portuguese, Chinese, Japanese, Korean and Turkish much more accurate... Facebook has instead created systems to recognize the content in the photographs uploaded by its users more quickly, simplifying the moderation of the contents prohibited on the social network.
Another sector in which the development of artificial intelligence has already given important results is that of cars that drive themselves. The software used by Google, Uber, and others, for now, is not very "smart" and constantly repeat algorithms to recognize road signs, the presence of other vehicles, pedestrians and make predictions about how they will move around the car. The idea, however, is to refine these systems so that through machine learning they learn to improve themselves, making automatic driving increasingly safe and reliable.