AI Cannot Choose the Values We Refuse to Name

A university asks an artificial-intelligence system to identify students who need support. A public agency asks another system to rank cases. A church deploys a chatbot to answer questions about faith. Each institution says that the technology should align with its values.

The sentence sounds responsible until the difficult questions begin. Which values? Defined by whom? What happens when access competes with privacy, efficiency with explanation, personalisation with common formation, or institutional loyalty with truth?

AI cannot choose the values an institution refuses to name. A model may reproduce preferences, optimise a stated objective or follow encoded rules. It cannot relieve governors, educators, pastors, professionals and communities of the responsibility to decide what kind of human good the system should serve, which trade-offs are legitimate and who may challenge the result.

Responsible AI therefore begins before procurement and before the prompt. It begins with moral purpose, participatory deliberation and governable human judgement.

Australia's National AI Centre lists a public lecture at the Australian National University for 19 August on artificial intelligence and value alignment. Professor Ruth Chang is scheduled to examine hard choices and to argue that familiar alignment strategies may not adequately address choices that are neither simple uncertainty nor indeterminacy. The programme proposes an approach that retains human input across symbolic and large-language-model systems.

This is an official event announcement. At publication time,…