What counts as a 𝗱𝗲𝗰𝗶𝘀𝗶𝗼𝗻, really? From the strict sense in 𝗮𝗱𝗺𝗶𝗻𝗶𝘀𝘁𝗿𝗮𝘁𝗶𝘃𝗲 𝗹𝗮𝘄—where a decision formally alters someone’s rights or entitlements—to the everyday sense where almost everything, from selecting a movie to choosing a candidate, feels like a decision. Yet across these very different contexts, the underlying anatomy looks surprisingly similar.

In CSIRO’s Data61 new preprint (link in the comment), 𝙊𝙫𝙚𝙧𝙨𝙞𝙜𝙝𝙩 𝘿𝙚𝙨𝙞𝙜𝙣 𝙛𝙤𝙧 𝘼𝙄-𝙀𝙣𝙖𝙗𝙡𝙚𝙙 𝘿𝙚𝙘𝙞𝙨𝙞𝙤𝙣 𝙈𝙖𝙠𝙞𝙣𝙜 𝙞𝙣 𝙂𝙤𝙫𝙚𝙧𝙣𝙢𝙚𝙣𝙩 𝙎𝙚𝙧𝙫𝙞𝙘𝙚𝙨, we unpack this shared decision stack: fact-finding, rule application, deliberation, discretion, and formalisation. Each stage represents a distinct kind of role in decision-making and therefore a distinct opportunity (and limit) for AI to assist.

Our approach applies our recent framework on meaningful human oversight to this stack: distinguishing between doing and overseeing, between AI assistance and human accountability, and between algorithmic outputs and what law calls the “state of mind” of a specified officer or delegate.

Even where AI can “do” most of the work under human oversight, the discretion and formalisation of a decision usually remain with a natural person — though AI can still perform independent double-checks.

We’re exploring how this layered view can make human–AI decision design more transparent and defensible across contexts.

If you’re interested in designing and testing the effectiveness of AI’s doing and human’s overseeing roles across your decision stack, we’d love to collaborate.