Google’s DeepMind achieves breakthrough in AI social learning
Researchers from Google’s DeepMind have demonstrated that AI can acquire skills through a form of social learning similar to humans and animals. The study highlights the potential for cultural evolution to contribute to the birth of AGI, and calls for further interdisciplinary research in the field.
Google’s DeepMind team of machine learning researchers claims to have achieved a significant breakthrough in AI by demonstrating that AI can acquire skills through a process similar to social learning in humans and animals. The researchers found, as per the recent publication in the peer-reviewed open-access journal Nature Communications, that AI agents can learn from human and AI experts in a simulated task space called GoalCycle3D, even though they have never encountered a human or had any concept of one. The team used reinforcement learning to train an AI agent capable of identifying new experts, imitating their behaviour in novel contexts, and retaining acquired knowledge within minutes, proposing that cultural evolution may play a role in developing artificial general intelligence (AGI). The researchers hope that their work will encourage further exploration of how cultural evolution can be applied more broadly in AI, and they look forward to interdisciplinary collaboration between the fields of AI and cultural evolutionary psychology in the future.