I Don't Want AI That Simply Agrees With Me

One thing I've noticed while working with artificial intelligence is how easy it can be to get an answer that sounds convincing.

You explain what you want. The system responds with a plan. The plan sounds reasonable, so you proceed.

But what if the original question was wrong?

What if you've overlooked something important?

Or what if the proposed solution is simply following a familiar approach because that's how everyone has always done it?

These are the situations where I expect more from AI.

I don't want a system that simply agrees with me and executes whatever I've asked.

If there's a weakness in my reasoning, I want it pointed out.

If there's a better way to achieve the objective, I want to hear it.

And if something doesn't make sense, I'd rather be challenged than reassured.

I don't see this as giving up human control. In fact, I see it as making better use of it.

Humans have limitations. We become attached to our ideas. We make assumptions without always noticing. Sometimes we ask the wrong questions because we don't know what we don't know.

I include myself in that.

One of the reasons I find AI interesting is the possibility that it could help us recognise those shortcomings before they become expensive mistakes.

But that requires a different kind of relationship.

Instead of treating AI as a tool that waits for commands, I've been exploring what happens when it is expected to contribute to the thinking itself.

Not to make every decision on my behalf.

Not to argue for the sake of arguing.

But to examine the objective, consider alternatives, identify missing information, and tell me when the evidence doesn't support the direction we're taking.

Of course, the AI can also be wrong.

That's an important part of the equation.

A system that confidently challenges a human with an equally flawed argument hasn't improved anything. It has simply introduced another source of error.

So I believe both sides need to be questioned.

The human should be willing to reconsider an assumption. The AI should be expected to support its recommendations with reasoning and evidence.

Neither should be treated as automatically correct.

I've also become interested in continuity.

When you work through a difficult problem over several discussions, the reasoning behind previous decisions matters. Why was one approach rejected? What risk was identified? What evidence changed our direction?

If those details disappear, we may keep revisiting the same mistakes.

That's why I increasingly value systems that preserve decisions, understand the history of a project, and can explain why something was done a particular way.

To me, the real opportunity isn't just faster execution.

It's the possibility of a working relationship in which human experience and machine capability help correct each other's weaknesses.

I don't need AI to tell me that every idea is brilliant.

I'd much rather have it help me discover which ideas are worth pursuing in the first place.