Updated
Artificial intelligences have managed to specialize themselves in their chosen field without having received specific instructions.
It is an extremely disturbing phenomenon that teams of researchers from the University of Massachusetts Institute of Technology (MIT) and Columbia University have observed. By studying the results of work of artificial intelligences specialized in facial recognition and object recognition, the scientists observed a very clear evolution in the results over time.
Read also: In China, an artificial intelligence becomes a prosecutor
Until now, AIs trained in face recognition and those trained in object recognition obtained almost identical results in both areas. Thus, an AI capable of identifying a face was also able to distinguish a truck from a car and vice versa. But gradually, the AIs, without having received any instruction, began to specialize. “An AI network trained to spot both 1,715 biometric identities and 423,000 object images initially performed well with both faces and objects despite its basic training,” the study explains. read on science.org.
But, subsequently, his performance plummeted with faces without affecting his ability to distinguish objects. In other words, these AIs have decided to lose interest in one job to devote themselves fully to another. This evolution implies that, like the human brain, AIs have spontaneously established a distinction between objects and faces!
For humans, the ability to recognize a congener very quickly has developed for reasons of survival, facial recognition is therefore assigned to a particular area of the brain, distinct from the recognition of objects. But why would an artificial intelligence have reached this same radical separation? Is this a simple consequence of the search for optimal efficiency or are stranger processes at work? This is the question on which scientists will now look.