New study reveals AI can create more popular income distribution schemes than humans.
Not easy. Building a system that delivers outcomes humans desire is dubbed value alignment in AI research. People frequently differ on the best way to tackle social, economic, and political challenges
"One important obstacle for value alignment is that human society permits a variety of opinions," researchers write in a new article lead by DeepMind's Raphael Koster.
The Human Centered Redistribution Mechanism (HCRM), built using deep reinforcement learning, used input from human players and virtual agents meant to mimic human behaviour.
"The AI redressed initial economic disparity, sanctioned free riders, and won the majority vote," researchers said.