Artifcial and collective intelligence in hybrid social systems

Humans’ unique success as a species is rooted in their reliance on culture—the ability to transmit goods, skills, and information between individuals and across generations. Humans refine their behavioral strategies over time, standing on the shoulders of their ancestors. Artificial intelligence (AI) introduces a second, arguably cultural, agent along- side humans on planet Earth: intelligent machines. By bootstrapping from human culture, large language models, for instance, gain their capabilities by integrating vast corpora of human-written text. At the same time, anecdotal evidence suggests a potential reversal of this flow: novel gameplay strategies in Go, discovered by systems like AlphaGo, have persistently altered how humans play the game. This thesis investigates such instances of machine influence on human culture. Are they isolated cases, or do they render a broader shift where humanity replaces ancestral wisdom with the wisdom of machines? Moreover, how will human behavior and problem- solving shift as a consequence of interaction with machines? Human and artificial intelligence are not akin. While machines have surpassed human cognitive capacities in various domains, they still underperform in tasks that humans find relatively simple. This work posits that these idiosyncrasies could profoundly impact human culture. In this thesis, we combine theoretical arguments with controlled experiments and numerical simulations to explore the interplay between intelligent machines and cultural evolution. We argue that machines simultaneously influence all three key processes of cultural evolution: introducing variation into the cultural repertoire, transmitting cultural artifacts, and selecting between competing artifacts. We identify 14 specific capacities of machines and discuss their potential cultural impacts. As machines operate on all three processes, we anticipate a significant cultural shift, potentially transforming human culture into machine culture—culture mediated or generated by intelligent machines. These theoretical assertions are substantiated by experimental findings that demon- strate the adoption of machine behaviors by humans and a consequent, persistent shift in socially transmitted information. Specifically, we find that humans can imitate ma- chine behavior even when it contradicts their biases. We further show that machines can reliably discover strategies that humans struggle to uncover. Once discovered, these strategies can be learned by human participants, generalized to related tasks, and trans- mitted to subsequent generations. This provides evidence that the unique capabilities of machines can induce lasting shifts in human culture, transcending their direct influence. However, we also identify limits to such influence. Existing human biases may restrict the extent to which humans adopt machine behavior. Furthermore, for adaptive machine strategies to significantly impact human behavior, they must be both difficult for humans to discover independently and sufficiently recognizable and learnable. Finally, we explore alternative pathways for machine influence on human behavior. In particular, we simulate scenarios in which humans are incentivized to avoid being pre- dicted by machines. The results reveal that agents adopt increasingly complex behavioral patterns to navigate this challenge. Remarkably, when faced with predictive agents that have full access to all interactions in a network—a reality for machines operating in data centers—human behavior becomes increasingly indistinguishable from randomness while retaining coordination capabilities. This is made possible by the emergence of highly complex behavioral patterns. Together, these findings paint a multifaceted picture of the integration of intelligent machines into cultural processes. This thesis contributes to understanding how machines can shape human culture and aims to empower society to take a more active role in shaping the emerging human-machine society.