DeepMind, a British artificial intelligence firm acquired by Google in 2014, is building an AI capable of “imagination” and understanding the consequences of previous actions.
In two research papers submitted last week, DeepMind describes how the AI would be able to “construct a plan” and remember information that may be important in the future.
“What differentiates these agents is that they learn a model of the world from noisy sensory data, rather than rely on privileged information such as a pre-specified, accurate simulator,” said the DeepMind research team to Wired.
“Imagination-based approaches are particularly helpful in situations where the agent is in a new situation and has little direct experience to rely on, or when its actions have irreversible consequences and thinking carefully is desirable over spontaneous action.”
Like most of DeepMind’s research, it used video games to test the AI’s proficiency. The AI played Sokoban, a puzzle game, without knowing the rules. In the video (below), as the AI begins to understand how the game is won, it becomes more accurate and moves faster.
DeepMind said the new AI showed “improved data efficiency, performance, and robustness to model misspecification compared to several baselines.”
In 2015, DeepMind showed its Deep Q-learning AI figuring out how to play Atari breakout. After 120 minutes, it had become an ‘expert’ at the game, capable of breaking all the blocks without missing.
The company is most commonly known for its AlphaGo AI, which defeated many human champions of the abstract strategy board game Go. DeepMind has not said if it will build another competitive AI agent in the future, for other strategy games like chess or shogi.