LessWrong proposes AI-sorted 'artificial families' for social support

KatjaGrace, writing on LessWrong, proposes a radical social experiment: use AI to create artificial families. The idea involves taking a large set of people, running personality similarity and physical proximity data through an algorithm, and grouping them into units of 4–10 across different ages. Each group would then receive substantial incentives—such as free access to local amenities—conditional on interacting together and maintaining good terms. The goal is to replicate the forced commitment and loyalty of biological families, including the inability to choose or leave, which Grace argues is central to family's virtue. She imagines this could be implemented by a national government or by a large group voluntarily opting in.

The proposal acknowledges missing elements like 'meaning and loyalty-to-one's-blood' but suggests that a perfect personality match might mimic that depth. The post has sparked discussion about whether technology can (or should) engineer kinship, and what risks such a system might introduce—from coercion to loss of autonomy. While speculative, it highlights a growing interest in using AI not just for productivity but for restructuring human social bonds.