Winning Games vs. Understanding Intelligence

Research has shifted from a focus on understanding intelligence to simply achieving success in specific games. Early programs like Deep Blue excelled at chess but did not provide insights into broader intelligence. The emergence of reinforcement learning allowed programs to develop learning capabilities by competing against each other, reflecting a more accurate model of intelligence. This approach underlies advanced systems like AlphaGo, which defeated a human Go master. Remarkably, the algorithms of reinforcement learning show parallels with signals in the brain, particularly in dopamine neurons, suggesting it may be a legitimate model of certain intelligence aspects.