One Concept, One Example, One Question

The concept

AI can do much of the work behind a decision without being the right party to make it. The third question in my Practical Delegation Test is: What judgment must remain human?

I ask this before handing over a task because a plausible AI answer can contain a choice that nobody has explicitly made. The judgment I'm looking for is where someone decides which trade-off is acceptable or what commitment the organisation can stand behind. If that choice goes unnamed, a recommendation can become a decision nobody can explain to the people affected.

The person making that choice needs enough evidence to challenge the proposal and the authority to take a different course. Sometimes policy settles the choice and a check is enough. Where it doesn't, someone has to make the judgment and answer for it.

The example

Imagine a service lead who can decide what their team addresses next. They ask AI to group recent complaints and show what each delay means to the people affected. About 200 people have waited a week for email replies, including some waiting to confirm appointments due soon. A dozen say the online form loses their answers before they can submit it. AI recommends clearing the emails first, citing the number waiting and the appointments due soon.

The lead checks examples from both groups. People can apply by phone, but only during office hours; some of the twelve can't call then, leaving them unable to submit this week if the form waits. Delayed emails could mean missed appointments. The task was clear and the cases checked; neither settles whose delay the service should accept.

Another lead might reasonably put the form first, but this lead clears the emails, accepting that some of the twelve can't apply this week. They'll have to explain the delay to those twelve and give the email group a reply date the team can meet.

The question

In a task you're thinking of giving AI, what must someone still decide, and who has to explain that choice if challenged?

Related

• Last week's issue, Can I Verify the Result Cheaply Enough?, covers the second question in the Practical Delegation Test.
• A Human in the Loop Needs a Job goes further into the evidence and authority a person needs to challenge an AI proposal.

Support

Everything I publish here is free to read, and always will be.

I maintain this newsletter and my website in my own time. If something here has been useful to you, and you'd like to help cover the costs behind that work, you can support me through Ko-fi.

Generative AI is a powerful tool. I openly use it to conduct research, challenge my thinking, generate drafts, and review my work. AI never owns what I think, or my personal judgement. Everything I publish is a reflection of my own thoughts and opinions.