Tim H's model shows AI could worsen well-being when work loses meaning

In a LessWrong essay titled "What if people value their work mattering?", economist Tim H tackles a gap in economic modeling: standard utility functions can capture enjoyment of an activity but not the dimming of that enjoyment when "AI will have usually gotten there first." He proposes formalizing this by assuming the marginal product of one's labor enters utility directly, representing how much work "matters" rather than just being pleasant. His toy production model splits capital between fully-automated production and human-complementing production, with a threshold above which additional automation raises total output but lowers the capital allocated to complementing human labor.

The key result comes from combining this with constant-elasticity-of-substitution (CES) preferences. If consumption and "mattering" are substitutes (elasticity > 1), utility eventually grows without bound as automation productivity increases — the classic story that tech disruption hurts at first but benefits everyone later. But if they are complements (elasticity < 1), then as automation productivity goes to infinity, utility approaches zero. Unlike arguments about a few mathematicians losing purpose, this model applies to everyone equally, sharing both material gains and the loss of meaningful work. Interestingly, with a sufficiently high weight on consumption, there can be an interior optimum where marginal consumption gains exactly offset marginal harm to mattering — and that optimum may yield substantially higher utility than no automation at all. The piece is purely theoretical but raises uncomfortable questions about whether aggregate well-being necessarily rises with transformative AI, even when everything goes well safety-wise.