This new computer algorithm created by researchers at the University of Berkeley cares more about knowing than achieving goals
Curiosity is no longer just a human faculty, the Artificial Intelligence he also wants to get hold of this ability. This new computer algorithm created by researchers at Berkeley University cares more about knowing than achieving goals. This software challenges the limits of Artificial Intelligence that it moves with firm step towards a future where the machines that have it will be capable of everything.
Children, the most curious human beings, base their knowledge of the world through this skill: they look, touch, express themselves and interact with their environment to know it as best as possible. What if machines could do the same? The IT software based on this new model of curiosity has been put to the test with a virtual character of a video game so that he knows his scenario without having to achieve specific objectives. Knowing for knowing is the objective of this algorithm that has been incorporated in one of the most classic games such as Mario Bros. It therefore becomes a reinforced Artificial Intelligence and more human than ever.
It’s not one of the first jobs to try to add new skills to Artificial Intelligence. In fact, a Deepmind program, AlphaGo, managed to beat a professional player of Go, a strategic board game originally from China. Although in this case the goal was defined: he had to beat his opponent. However, with this new study, the machine that incorporates this technology will perform machine learning by simply knowing its environment. And so he did through the game Mario Brosalthough he could not pass the first level, the character knew his environment better and better without having to collect coins or beat the monsters. But the more he learned, the easier it was for him to overcome the small obstacles that lay in his way. Therefore, effectiveness grows when curiosity is on their side.
Artificial curiosity, tested
These curious machines with a desire to know “the world” can even learn more than those that only focus on very specific objectives. Although this curiosity is selfish if it is related based on the same ability, but in humans. Machines are curious only about what directly affects them. Unlike humans we are able to be observers of the environment, although it does not affect us.
It is expected that thanks to this reinforced learning robots will be able to respond to spontaneous adverse situations as quickly as possible. One of its creators, Pulkit Agrawal, points out that in the face of these adversities the androids wasted a great deal of time. The arrival of this artificial curiosity may facilitate the reaction of these robots to unforeseen situations, such as that of a robotic arm that has to hold an irregularly shaped object.
This new algorithm was tested in two games, the aforementioned Mario Bros and VizDoom, a 3D shooter rpg. The implantation of this curiosity in the characters reinforced their learning process. For example, in the 3D video game the protagonist explored space and moved in it on his own very effectively. Even so, this reinforced learning has its limitations, because in both games it was shown that it is going to be safe. Rather than launching into further exploration, the AI retraced its steps in order to continue. In addition, without discovering possible shortcuts or those hidden scenarios such as, for example, the inside of the wells in the game of Mario Bros.
Undoubtedly, this most curious advanced autonomous learning is a great progress for these intelligent machines that grow to be able to overcome challenges, until now impossible for us. Will this learning process in robots be equated to that of a child who learns to take his first steps? Let Artificial Intelligence take its course and keep surprising us.
Even so, in this path that AI travels not only robots are the great beneficiaries of cutting-edge technology, health or education also have a place for it. From the BlogThinkBig.com we reveal the potential of Artificial Intelligence in education. Education and technology can go hand in hand, moving away from the stereotypes that place it against it. And where it can benefit us most is in health. Increasingly early and effective detections are the best ally of AI to overcome problems as serious as cancer or depression. In fact, we were overtaking in the BlogThinkBig.com how Artificial Intelligence is closer to overcoming depression.
It only remains to wait to witness what technology can bring us.