A team at the Oxford Internet Institute led by Professor Sandra Wachter argues that EU support for AI should produce a measurable public return in well-being, not only economic growth. Their paper, “Optimising for happiness in EU technology policy to repay public subsidies for AI,” appeared in Royal Society Open Science.

Public backing reaches AI through direct funding, tax breaks, and access to land, water, energy, infrastructure and data. Creative work, social media posts and user interactions are being used to train frontier models, often without meaningful public control. That use has triggered a stream of lawsuits.

The researchers place well-being beside economic growth and risk mitigation as a third policy objective. They distinguish reducing harm from creating social benefit. Their framework treats relatedness as one measure. Competence and autonomy are the other two. Emotion-mimicking chatbots could weaken the value of human relationships. Heavy AI use could erode skills, and AI influence could make independent choices harder. Data centers consume land and water, leaving fewer trees and green spaces for people to connect with nature.

Companies benefiting from public support would have to estimate how their technology will improve happiness and face accountability for the promised gains. The EU has allocated €20 billion for AI gigafactories. GenAI4EU and the Digital Europe programme are using €700 million through Horizon research projects focused on generative AI and supercomputing.

The proposal comes amid rising material and social costs. Data centers are increasing electricity demand and could push up household bills. Memory makers are prioritizing higher-margin chips, making ordinary DRAM and NAND more expensive along with the devices that use them. Server cooling consumes water. In 2026, Anthropic warned that poorly controlled powerful models could threaten human extinction. In 2025, Anthropic and OpenAI raised the prospect of mass unemployment as employers adopt AI to cut costs. Limited access to data from model makers and cloud providers makes the environmental damage harder to assess.