DeepMind: Its AI company DeepMind Google, based in London, created two different types of AI that can use their imagination to plan the future and perform work with a higher rate of success than AI without imagination.
Are you confused?
In a publication on their website, DeepMind researchers give a brief overview of the "new family of imaginative design approaches". The so-called imaginary agents, or I2As, use an internal fantasy encoder that helps AI decide what is and what is not useful predictions for the future.
The researchers argue that AI's imagination is crucial for tackling real-world problems, where you sometimes need to mentally try out some possible outcomes of actions to predict or decide which one is best.
Recently, the founder of DeepMind, Demis Hassabis wrote a paper published in Neuron about how the development of general purpose AI depends on understanding and codifying human capacities such as imagination, curiosity and memory. So his company appears to be making progress in at least one of these human capabilities.
The I2A carrier machines were tasked with controlling different situations for scientists to observe their predictive abilities. Among the tests was a Sokoban puzzle game as well as a spaceship navigation game. Sokoban is a puzzle game where a little alien has to push boxes into the right place – he can't pull boxes though, so a wrong movement it can make you lose the entire round.
I2A carrier vehicles performed much better than machines not using I2A. They learned how to navigate the less experienced jigsaws by extracting more information from internal simulations that they were doing.
Of course, the type of fantasy described in Demis Hassabis's paper is not even approaching the way people use their imagination. But the experiment shows that AIs can work much better when imagining different scenarios before they act.
As Hassabis told Neuron, creating imaginative machines that can compete with what we can do "is perhaps the most difficult challenge for AI research: to build an agent that can design hierarchically, be truly creative and it produces solutions to challenges here and now without the need for the human mind. ”
Step by step, we'll probably get there too.
