DeepMind study shows AI successfully achieves real-time social learning without relying on pre-collected human data

DeepMind researchers have made a groundbreaking discovery, suggesting that artificial intelligence (AI) can develop real-time social learning skills without the need for pre-existing human data. In contrast to traditional methods that rely on imitation learning and extensive exposure to data, this study explores the potential for AI to replicate the cultural evolution and social learning that has long been a cornerstone of human intelligence.

Imitating experts for accelerated learning

In the next step, expert agents were introduced, either coded or human-controlled. AI agents quickly learned that mimicking these experts was the most efficient way to navigate. The AI not only learned faster by imitating experts, but also applied this knowledge to navigate other virtual paths. In addition, AI agents demonstrated their ability to use their new skills even without experts, demonstrating a form of social learning.

Implications for Artificial General Intelligence (AGI)

While the authors acknowledge the need for further research, they believe their method could pave the way for cultural evolution to play a crucial role in the development of general artificial intelligence. This breakthrough could revolutionize the AI industry by reducing potentially resource-intensive training and improving problem-solving skills. In addition, it raises the intriguing question of whether AI can adopt the social and cultural aspects of human thought.

At an early stage, DeepMind’s discovery opens new avenues for the development of artificial intelligence and highlights the possibilities for more efficient and adaptive learning algorithms. Interdisciplinary cooperation between artificial intelligence and cultural evolutionary psychology could promote the development of artificial intelligence.