Friends, Foes, and First Authors: A Game Theory Model of How Power Plays Rewrite Academic Co-Authorship Networks

Researchers Amit Bengal and Teddy Lazebnik have published a novel game theory model that uses deep reinforcement learning to simulate how strategic behavior affects academic co-authorship networks. Their paper, 'Friends, Foes, and First Authors,' creates a repeated, networked game where AI agents form collaborations, accumulate reputation, and learn when to issue ultimatums about authorship order. The model compares myopic (greedy) authors with forward-looking (strategic) ones in mixed populations, revealing that strategic agents don't issue fewer ultimatums but instead learn crucial social cues.

The key finding is that strategic agents learn to avoid insisting on their demands after facing rejection, which dramatically reduces destructive manuscript termination. Through large-scale simulations, the researchers found that as the prevalence of strategic agents increases, paper destruction rates plummet from 0.120 to 0.000 per paper, while completion rates jump from 85.3% to 97.0%. Strategic agents also gain a substantial utility advantage of 30.8% when rare, and the average number of completed papers per agent rises from 15.2 to 16.9—a 14% productivity increase.

This research provides the first computational framework to systematically study how reputational feedback and long-term incentives can make academic collaboration more resilient. By modeling how power dynamics play out over repeated interactions in evolving networks, the study offers institutions and publishers a testbed for designing authorship policies that minimize conflict and maximize productivity. The findings suggest that teaching researchers to be strategically cooperative—rather than simply less demanding—could transform how scientific teams navigate the high-stakes world of authorship credit.