Google’s research engineers managed to create AI agents that can make plans and predict the consequences of their actions. In others words, AI can now imagine what could happen in the future based on current facts and situations.
Imagination is a powerful tool of human cognition and a distinctive feature of human evolution. Well, now we are not the only ones that can imagine.
AI algorithms operate with very complex behaviors, therefore they should be able to imagine and create scenarios about the future. Moreover, AI agents also should be able to develop step-by-step plans using this knowledge, in order to adapt their behavior to the respective context. In theory, things look simple, but it’s very difficult to put this idea into practice due to the huge number of variables that the system needs to take into account.
Google’s new AI approaches for imagination-based planning
Google’s AI researchers recently published two new papers, describing new approaches and architectures for imagination-based planning. Thanks to these breakthroughs, AI agents can now learn to construct plans and adopt flexible strategies for exploiting their imagination.
The process that AI agents use is similar to how humans think. The system creates different strategies and then runs simulations to test them out.
The agents we introduce benefit from an ‘imagination encoder’- a neural network which learns to extract any information useful for the agent’s future decisions, but ignore that which is not relevant.
Researchers then tested these enhanced AI agents on the puzzle game Sokoban and a spaceship navigation game. They chose these games as both “require forward planning and reasoning”.
As Google explains, in Sokoban the agent has to push boxes onto targets. Because boxes can only be pushed, many moves are irreversible. On the other hand, in the spaceship game, the agent must stabilise the aircraft by activating its thrusters. The catch is that there is a limited number of times that the agent can use the thrusters.
To limit trial-and-error for both tasks, each level is procedurally generated and the agent can only try it once; this encourages the agent to try different strategies ‘in its head’ before testing them in the real environment.
To see Google’s AI agents in action, check out the video below:
The results of this research are very encouraging for Google. The company aims to improve its AI agents so that they can use their imagination to reason about and plan for the future in complex environments, where unpredictable problems can arise.Follow The AI Center on social media:
Join me as I track the latest progress in AI research.
Latest posts by Maddie Blau (see all)
- True Emoji is an AI app that uses your expressions to create animated emojis - November 21, 2017
- Canada’s first AI exchange-traded fund enters the market - November 2, 2017
- Are you curious to see who’s the smartest AI in the world? - November 1, 2017