DeepMind - AI department at Google – has presented a new algorithm to predict the future.
While unique AI technology developing, I2A – agents with a program code of “imagination” – were used. They allow the system to identify the most probable future state of things. Also, these agents help AI algorithm to plan actions so that they could lead to the wanted result.
The DeepMind representative Demis Hassabis thinks that AI technologies development requires program analogs of such human qualities as imagination, memory and curiosity. I2A agents ensure all of them.
The attained results are satisfactory. The new algorithm with I2A agents appeared to be far better than usual artificial intelligence. It was tested on the puzzle game Sokoban. The system had to calculate different sequences of events to make the only correct move. The AI algorithm oriented quickly and extracted data from the event simulation. Then, developers connected a managing program to compose a plan of future actions assessment. The program started working more accurately.
Of course, AI “imagination” is quite primitive but the system has become more effective due to the new function.
By the way, the Swedish expert Nikl Bostrom is sure that it is DeepMind that will develop the first humanlike AI. It will happen in the nearest decades.