The current revolution in artificial intelligence is made possible by "a combination of 3 factors," according to Harry Shum: "a vast amount of data; extraordinary computing power, especially thanks to the cloud; and revolutionary algorithms, based on deep-learning ".
Either data, the power to exploit them, and the ability to learn. Because the so-called "artificial intelligence" is inspired by human cognitive processes. In particular, our very great capacity for learning, throughout life.
AI thus frequently uses supervised learning. For example, we "feed" a program with thousands of photos of cars, labeled. After this "training", the program can recognize, alone, cars on the new images that will be presented to it.
Another component of artificial intelligence is machine learning. This time, computers are given access to data and then let them learn on their own, without human intervention or software reprogramming. This allows them to progressively improve, independently. And to go beyond the functions and capacities initially programmed.
Some algorithms are no longer content to recognize images, but are able to produce them and to give eyes to the machines.
As for Deep Learning, it is based on a network of artificial neurons, which mimic the functioning of our brain. And creates a virtual machine composed of thousands of units, each loaded with simple small calculations.
Solve a complex problem by dividing it into several subproblems, with an artificial intelligence assigned to each one, it is the technique which allowed to obtain the best possible score to ... Miss Pac-Man. The researchers used reinforcement learning and artificial neural networks. A result that opens new perspectives for AI and accomplishing complex tasks, in areas where unpredictability was a problem.
Analyze a new environment, understand the rules that govern it, adapt ... Video games are an excellent field of action for researchers in artificial intelligence. In particular, they use Minecraft to train AIs, pushing them to understand their environment, build buildings or transportation, explore labyrinths and better cooperate.
The announced arrival of the quantum computer could further increase these capacities. Another good news: until recently, artificial intelligence required extremely strong expertise and consistent resources, in time, resources, money and people. But today it has become widely democratized, and developers can easily integrate AI into their applications.