Participants with access to semantic knowledge:
🔹 Explored fewer, more focused combinations
🔹 Showed lower entropy (less random search)
🔹 Used semantic generalization to build on prior success 🔹 Combined semantically dissimilar items to innovate

Using the same innovation task, we tested 1,243 participants combining items to make new inventions.

When items had meaning (semantic condition), people innovated far more—especially with social learning. Without meaning (non-semantic condition)? Performance was no better than random "bots".

Populations with both semantic knowledge and social learning produce far richer cultural repertoires.

Agents that with both semantic knowledge and social learning dominate over time, exploring efficiently—not randomly. Crucially, semantic knowledge and social learning act in synergy.

In our model, agents start without knowing how items combine. Through success, they build semantic knowledge. structured associations between items and functions, and pass this knowledge to their offspring.

How do humans keep inventing tools and technologies that no single person could create alone?

Our new preprint, led by
@anilyaman.bsky.social & @ts-brain.bsky.social
shows that semantic knowledge guides innovation and drives cultural evolution. 🧠📘 arxiv.org/abs/2510.12837

Participants with access to semantic knowledge:
🔹 Explored fewer, more focused combinations
🔹 Showed lower entropy (less random search)
🔹 Used semantic generalization to build on prior success
🔹 Combined semantically dissimilar items to innovate

Using the same innovation task, we tested 1,243 participants combining items to make new inventions.

When items had meaning (semantic condition), people innovated far more—especially with social learning.
Without meaning (non-semantic condition)? Performance was no better than random "bots".

Populations with both semantic knowledge and social learning produce far richer cultural repertoires.

Agents that with both semantic knowledge and social learning dominate over time, exploring efficiently—not randomly.

Crucially, semantic knowledge and social learning act in synergy.

In our model, agents start without knowing how items combine.
Through success, they build semantic knowledge. structured associations between items and functions, and pass this knowledge to their offspring.

The second @rldmdublin2025.bsky.social poster from the lab is by fantastic PhD student Shen Tian. Using modeling and experiments, he investigated the key role of semantic knowledge in a complex innovation task (originated by @maximederex.bsky.social). On Thursday, poster #86

Following up on our previous work (pnas.org/doi/10.1073/...) we
(led by Ida Selbing) find that extinction learning does not reduce the maladaptive transfer of social threat learning to decision-making. osf.io/gyzm3/