Alphabet’s DeepMind recently developed an artificial intelligence (AI) system that taught itself how to navigate the world like a mammal, according to a blog post. The discovery signifies that AI could become more human, and that it could to it all on its own.
The research began as an effort to determine how accurately AI could imitate certain aspects of an actual brain. Specifically, the post said, the research team was looking into the idea that grid cells can help support vector-based navigation.
Using velocity measurements, the team trained the AI to figure out where it was in a virtual environment, the post said. This is similar to the way mammals, including humans, move through areas they aren’t familiar with, or areas that are dark.
SEE: IT leader’s guide to the future of artificial intelligence (Tech Pro Research)
However, once they let the AI loose with these parameters, it began to mimic mammalian behavior patterns on its own.
“We found that grid-like representations (hereafter grid units) spontaneously emerged within the network – providing a striking convergence with the neural activity patterns observed in foraging mammals, and consistent with the notion that grid cells provide an efficient code for space,” the post said.
The team then trained the AI to navigate toward a goal area in a virtual reality game environment. Using its own developed grid cells, the AI did very well.
“This agent performed at a super-human level, exceeding the ability of a professional game player, and exhibited the type of flexible navigation normally associated with animals, taking novel routes and shortcuts when they became available,” the post said.
However, when the team cut off the AI’s access to its grid cells, it performed poorly.
According to the post, the team believes that these results are evidence that imitating brain-like algorithms could lead to more powerful machine learning tools. A similar approach could be used to perceive sounds or control limbs, the post said.
For businesses, this could mean that more lifelike AI systems and maybe even robotics are much closer on the horizon than once thought. It’s also evidence that, if an organization is planning to work with AI, it must be prepared for unintended consequences–as they all might not be as pleasant a surprise as the one found by DeepMind researchers.
The big takeaways for tech leaders:
- Alphabet’s DeepMind developed an AI system that spontaneously taught itself how to navigate like a mammal would, using what are known as grid cells.
- DeepMind’s grid cell research could be used to create more human-like robots and AI, and highlights the unintended consequences that could come from working with AI.