My new piece "Machine behavior needs to be an academic discipline" is out in Nautilus magazine

turingbox_microscope.png

Despite their growing influence on our lives, our study of AI agents is conducted by a very specific group of people. Those scientists who create AI agents—namely, computer scientists and roboticists—are almost exclusively the same scientists who study the behavior of AI agents. We believe the study of AI agents can benefit from greater inclusion of social and behavioral scientists.

Read my Nautilus article with Manuel Cebrian, making a case for a new science of "Machine Behavior".

My student Morgan Frank's paper made the cover of Science Advances

ICEWS.jpg

In 1981 Robert Axelrod famously hypothesized that international cooperation might evolve through direct reciprocity. In our new article in Science Advances, we present new evidence for that hypothesis. Using a new method and dataset, we're able to detect direct reciprocity in a relatively large number of country pairs. Reciprocating country pairs exhibited higher levels of stable cooperation, and were more likely to punish (yet quickly forgive) instances of non-cooperation, in an effort to maintain cooperation. By contrast, countries without reciprocal strategies were more likely to exploit each other's cooperation and abandon mutual agreements. The results indicate that policymakers of powerful countries—who may be tempted to take sweeping, non-cooperative action in areas like trade or environment—should consider the long-term negative effect of non-reciprocal tactics.