Could humans and AI become a new evolutionary individual?

This PNAS opinion piece (open source) by Rainey and Hochberg is well worth a read.  It suggests that as different AI implementations suffuse into cultures, current competition between different cultures could evolve into accelerating competition between different culture - AI hybrids.  I pass on the introductory paragraphs to their arguments:

Artificial intelligence (AI)—broadly defined as the capacity of engineered systems to perform tasks that would require intelligence if done by humans—is increasingly embedded in the infrastructure of human life. From personalized recommendation systems to large-scale decision-making frameworks, AI shapes what humans see, choose, believe, and do (1, 2). Much of the current concern about AI centers on its understanding, safety, and alignment with human values (3–5). But alongside these immediate challenges lies a broader, more speculative, and potentially more profound question: could the deepening interdependence between humans and AI give rise to a new kind of evolutionary individual? We argue that as interdependencies grow, humans and AI could come to function not merely as interacting agents, but as an integrated evolutionary individual subject to selection at the collective level.

Lessons from the History of Life

The theory of major evolutionary transitions (METs) provides a framework for understanding this possibility (6). METs have punctuated the history of life. Those of particular relevance to our thesis here are those involving transitions in individuality. Such transitions are defined by events in which lower-level autonomous units—such as genes, cells, or organisms—become components of a higher-level individual subject to selection as a whole (7). Examples include the evolution of chromosomes from independent genes, multicellular organisms from single cells, and eusocial colonies from solitary ancestors (8). A particularly instructive case for our purposes is the eukaryotic cell, which arose from the integration of two ancient microbes—an archaeon and a eubacterium (9).^*

 

To better understand the evolution of human–AI interactions, consider a globally embedded system: diffuse in structure, nonreplicating, yet unified in function. Such a system might emerge not by design alone, but via competitive, economic, or ecological pressures (10), selected for its capacity to integrate, persist, and coordinate across human networks. Initially shaped by human design and deployment, this AI could become embedded across societal and cultural infrastructures—advising, mediating, and responding. This would entail a persistent informational presence that learns from humanity and, in turn, comes to shape it. Importantly, even if yet incapable of autonomous replication, such an AI is likely to modify its own code in response to environmental feedback.