iiinova — where your digital self keeps evolving

Reposted by Paul Bergmann

Björn Siepe @bsiepe.bsky.social · 17/07/2026

To celebrate the openESM paper release, I wrote a brief blog post summarizing recent developments: openesmdata.org/blog/2026-07... We: 💠added more data 💠provided descriptive visualizations for each item 💠created a semantic similarity mapping for items, making it easier to find related items

openesmdata.org

Updates: Paper published, new dataset & features

Our tutorial paper on openESM has been published, and we have added new datasets and features to the platform.

24523

Reposted by Paul Bergmann

Dominik Deffner @dominikdeffner.bsky.social · 10/07/2026

Learning from the most successful people makes sense? We show that social learning, and especially payoff bias, is bad when choice outcomes are dependent on prior success💰 As such "rich-get-richer" dynamics are everywhere, this shows fundamental boundary conditions for the benefits of SL!

13010

Paul Bergmann @pkbergmann.bsky.social · 10/07/2026

In short: when success is cumulative and partially due to luck, success-biased copying leads to worse outcomes for most. Disproportionally conforming to the majority protects against this misleading influence. Very happy to receive feedback on this and many thanks to @dominikdeffner.bsky.social 7/7

010

Paul Bergmann @pkbergmann.bsky.social · 10/07/2026

As a result, learning from the most successful individuals over all prior decisions (wealth-bias) or a given round (payoff-bias) is misleading for most, leading to worse outcomes and greater population-level inequality. Findings are robust even under time-optimal utility transformations. 6/7

110

Paul Bergmann @pkbergmann.bsky.social · 10/07/2026

This is because a fortunate but risk-seeking minority in the multiplicative environment receives repeated favourable outcomes, learning to prefer the riskier choice and producing systematically misleading social cues. In larger populations, this effect becomes more pronounced. 5/7

120

Paul Bergmann @pkbergmann.bsky.social · 10/07/2026

By varying the available social information, we find that most well-known social learning strategies reduce decision quality in the multiplicative environment. Payoff- and wealth- biased learning strategies, which perform best in additive environments, led to the worst outcomes. 4/7

Heatmaps of individual learning parameters across several social learning strategies and across additive and multiplicative environments. Individual learning rate and inverse temperature are systematically varied. In the additive environment, all social learning strategies (including conformity, unbiased copying, payoff bias and wealth bias) improve performance by increasing the probability of choosing the riskier option that has a higher expected value. In the multiplicative environment, all social learning strategies with the exception of conformity deteriorate performance, also increasing the probability of choosing the riskier option, which is worse in the multiplicative environment given its higher variance (i.e., lower time-average growth rate).

120

Paul Bergmann @pkbergmann.bsky.social · 10/07/2026

Repeated independent decisions are insensitive to variance, while multiplicative accumulation is sensitive to variance. Thus, in our model risky choices are optimal in additive env. (higher expected value), while safer choices are better in multiplicative env. (higher time-average growth). 3/7

110

Paul Bergmann @pkbergmann.bsky.social · 10/07/2026

Using agent-based, social reinforcement learning models we simulate agents making repeated decisions between relatively risky and safe options. We contrast decision environments where decision outcomes accumulate either independently (i.e., additive) or depend on another (i.e., multiplicative). 2/7

120

Paul Bergmann @pkbergmann.bsky.social · 10/07/2026

New preprint with @dominikdeffner.bsky.social: While social learning is broadly adaptive when repeated decision outcomes are independent, we show that when outcomes depend on prior success, most social learning strategies deteriorate performance and increase inequality: doi.org/10.31234/osf... 1/7

doi.org

OSF

1184

Reposted by Paul Bergmann

Dominik Deffner @dominikdeffner.bsky.social · 02/03/2026

Come work with us! And get in touch with any questions you might have about the position, our labs or living/working in Germany #PostdocWanted

01612

Reposted by Paul Bergmann

Charley Wu @thecharleywu.bsky.social · 02/03/2026

🚀 Postdoc Alert! Are you passionate about social learning & cultural evolution? @dominikdeffner.bsky.social & I have a 3-year position with freedom to develop your research and work on cutting-edge multiplayer and immersive experiments. Apply by March 30! hmc-lab.com/SocialLearni... Pls share 🙏

hmc-lab.com

Postdoc position -- Social Learning and Cultural Evolution

Postdoc position -- Social Learning and Cultural Evolution posted on March 2, 2026 We are currently seeking a highly motivated individual...

26365