Artificial intelligence DeepMind learns to predict the future based on past

A system of artificial intelligence achieve goals, based on ideas about the world based on past experience. They generalize them and pass on to a new situation that allows them to perform tasks even in conditions with which they had not previously encountered. And if as a result of those actions “mission is accomplished”, the AI gets some kind of “reward.” This method of learning is called reinforcement learning. However, this method of training artificial intelligence suitable to not only obtain new knowledge but also to predict the future.

The AI learns to predict the future. What’s next?

How the AI can predict the future?

Researchers from the laboratory of DeepMind and the University of Toronto in the course of studying the dynamics of models of reward set up AI so that it, based on past experience, did not take any action, and predicted it based on known cases. In this context, about the full prediction of events is not yet mentioned. Now the model only examines non-obvious possible location of the object in the future, given their non-linear speed and winding path.

Model of the behavior of the AI has been called Dreamer and it consists of five basic elements. The first is a representative component, which produces the observation of the object and determines its next steps. The second transition module that anticipates the state of the object, not knowing their original properties. The watchdog current status provides feedback signals which works in conjunction with the module observations of the final state which allows to compare the prediction with the result. Well, and crowned the whole complex system component of the promotion.

In several experiments, scientists tested the Dreamer to 20 tasks in the program DeepMind Control Suite. First, they trained him with a single graphics chip Nvidia V100 and ten CPU cores. Each training cycle went for 9 hours. For comparison, the PlaNet, the predecessor of the Dreamer, achieved a similar result for 17 hours. In the result, the Dreamer was able to use the learned models to draw conclusions about possible outcomes and did it quite accurately. What do you think, how accurately the AI can predict the future? And is it possible to make such a prediction at all? Write about it in our chat in Telegram.

See also: Artificial intelligence has learned to track down the hackers

There are AI’s that use big data to analyze the situation and predict outcomes, but they sharpened it to work with, which is logical, a large amount of information you need to study in advance. Dreamer has developed an algorithm that can predict the outcome is not worse than specialized systems. Yes, at the moment AI only predicts the position of randomly moving objects, but he has a possible scope.

For example, models like the Dreamer, can be used for short-term planning. That is, in theory, he will be able, based on the data operation Department of the company to predict what if these performance indicators and the schedule they will achieve in the future. You can then optimize the activities of people so that “the future has changed,” and their performance improved.

Leave a Reply

Your email address will not be published. Required fields are marked *