theantimedia | Will artificial intelligence get more aggressive and selfish the more intelligent it becomes? A new report
out of Google’s DeepMind AI division suggests this is possible based on
the outcome of millions of video game sessions it monitored. The
results of the two games indicate that as artificial intelligence
becomes more complex, it is more likely to take extreme measures to
ensure victory, including sabotage and greed.
The first game, Gathering, is a simple one that involves
gathering digital fruit. Two DeepMind AI agents were pitted against each
other after being trained in the ways of deep reinforcement learning. After 40 million turns,
the researchers began to notice something curious. Everything was ok as
long as there were enough apples, but when scarcity set in, the agents
used their laser beams to knock each other out and seize all the apples.
The aggression, they determined, was the result of higher levels of complexity in the AI agents themselves. When they tested the game
on less intelligent AI agents, they found that the laser beams were
left unused and equal amounts of apples were gathered. The simpler AIs
seemed to naturally gravitate toward peaceful coexistence.
Researchers believe the more advanced AI agents learn from their
environment and figure out how to use available resources to manipulate
their situation — and they do it aggressively if they need to.
“This model … shows that some aspects of human-like behaviour emerge as a product of the environment and learning,” a DeepMind team member, Joel Z Leibo, told Wired.
