Artificial intelligence with MIT like a child – intuitively understands the laws of physics


Another step in the right (?) Direction.

Perhaps this is not obvious, but people from an early age in a basic way they understand some of the laws of physics. Babies, for example, have certain expectations of how objects should behave when they interact with each other. When these expectations are not met, that is when objects behave unexpectedly, children react with surprise. Researchers at MIT have just developed artificial intelligence, which also shows an intuitive understanding of selected laws of physics.

ADEPT, because this is the name of artificial intelligence, can look at a given object registered on video, predict how this object should behave based on what it knows about the laws of physics, and then react with surprise if the object unexpectedly disappears or teleports . According to the researchers, this model will not only allow other scientists to create more advanced artificial intelligence in the future, it may also allow them to better understand how babies perceive their surroundings.

"By the time the baby reaches three months, he already has some idea that objects just do not disappear and suddenly do not appear, and that they cannot penetrate or teleport.", said Kevin A. Smith, one of the researchers behind the creation of the ADEPT model. "We wanted to capture and formalize this knowledge to embed infant perception into artificial intelligence systems."


ADEPT works on the basis of two modules. The first examines the object, determining its shape, position and speed. It should be noted that he does not do it in a very detailed way. It does not analyze absolutely every aspect of the object, and instead looks closely at its geometry. According to the creators of artificial intelligence, this allows the system to predict the behavior of many objects, and not only those for which it was trained. In addition, it makes the system resemble babies. When young children think about how something can move, they don't care too much about the specific physical properties of that thing.

The second module of artificial intelligence is a physics system. It shares some features with the physics systems used in computer games. It uses the data collected by the first module and based on the laws of physics simulates how a given object should behave. After predicting several results, the system compares them with subsequent frames of the film. When he notices the discrepancy between what could happen and what actually happened, he will send a signal. The stronger the signal, the more it surprised him what just took place. Interestingly, ADEPT responded to individual movies with the same level of surprise as people who were shown the same movies.

In the future, MIT researchers intend to look more closely at how young children perceive the world by introducing appropriate improvements in their model. We are looking forward to the results of these activities.

Source: MYTH