The latest

A behavioral experiment found that people can adopt strategies discovered by artificial intelligence, transmit them to others and preserve them over successive generations. Researchers at the Center for Humans and Machines at the Max Planck Institute for Human Development, working with the Toulouse School of Economics and Humboldt University of Berlin, tested when machine-generated solutions can become durable human knowledge. They found transmission depended on the strategy remaining learnable and its advantage being clearly visible.

Details

• Task design: The study involved 1,155 participants, each seeking to maximize points in specially designed reward-based tasks. The optimal approach required accepting small early losses to secure much larger later gains, a pattern running against the common human tendency to avoid short-term losses whenever possible.
• Learning chains: Participants were arranged in groups that learned across multiple generations. Some chains contained only humans; others began with AI agents that had already completed the task using a learning algorithm and discovered the optimal strategy. Later participants could observe successful solutions from the preceding generation.
• Results gap: Among 600 participants in human-only groups, just one independently found the best solution. The machine-discovered strategy spread much more often in mixed groups and remained in use through the entire experiment in nine of the 15 human-machine groups.
• Model choice: Participants strongly favored the most successful available role model, usually choosing the person or agent with the highest result. That preference gave the machine’s unconventional strategy a route into human behavior and helped sustain it over time.
• Transmission conditions: Researchers identified three requirements for lasting uptake: the strategy must be difficult for humans to discover initially, remain learnable once observed, and offer a clearly recognizable advantage. Together, those features allow a machine discovery to spread within a community and endure across generations.
• Cultural effect: Lead author Levin Brinkmann said cultural evolution depends on knowledge moving among many individuals and generations. Co-lead author Thomas Eisenmann said comprehensible machine discoveries could help people develop new skills, preserve them over time and broaden their cultural repertoire.

Background

Modern AI systems have surprised human experts with unusual but successful play in Go and chess. The study frames cultural exchange as potentially two-way: advanced systems learn from human knowledge, while people can build capabilities from machine discoveries. This idea underpins “machine culture,” in which humans and machines each supply innovations the other can inherit and extend. Large language models illustrate the reverse direction: they are trained on vast collections of human-written text and can process and generate fluent text across nearly any topic.

What’s next

The next indicator for machine-to-human cultural transfer is whether an AI-discovered strategy remains understandable, learnable and visibly advantageous when passed through further multigenerational learning chains.