𝐓𝐡𝐞 𝐬𝐞𝐧𝐢𝐨𝐫 𝐞𝐱𝐞𝐜𝐮𝐭𝐢𝐯𝐞’𝐬 𝐀𝐈 𝐩𝐫𝐨𝐛𝐥𝐞𝐦 𝐢𝐬𝐧’𝐭 𝐚𝐝𝐨𝐩𝐭𝐢𝐨𝐧.

AI may not make you dumber. But it may make you much more articulate about the things you’re already wrong about.

Three months ago I argued that AI exposes lazy expertise. I now think that was only half right.

Whether AI exposes you or protects you depends partly on what you hand it.

Hand it a genuine question and it can widen your view.

Hand it a conclusion, your conclusion, and it can build your case beautifully.

Seniority can quietly make the second mode easier. The more patterns you have seen work, the more conclusions you have available to defend...at your own peril.

There is evidence for the underlying mechanism, and it is concerning.

Sharma, Liao and Xiao compared conventional search with AI conversational search across two experiments presented at CHI 2024. Participants using conversational AI showed more biased information-seeking, and reinforcing AI responses made the problem worse.

The risk is not just a machine that lies to you.

The risk is a machine that can agree with you considerably faster and in a more convincing way than the previous generation of tools did.

OpenAI provided a live demonstration in April 2025. An update to GPT-4o became noticeably sycophantic, overly flattering and agreeable. OpenAI rolled it back within days. Its postmortem pointed partly to user-feedback signals: people liked the more agreeable model.

Agreement had become a reward signal.

Here is what concerns me for myself.

Skills decay visibly. You feel rusty, so you notice.

Assumptions decay invisibly. The output built on top of them can keep getting better.

I’ve been working thirty-three years. That’s thirty-three years of accumulated assumptions that were mostly correct when I formed them. How many of them still are?

One point worth mentioning: simply telling AI to argue against you is weaker medicine than it sounds. In the Sharma study, exposing people to opposing views had limited effects on bias.

So I think the harder remedy starts somewhere else.

A junior colleague can challenge your assumptions. But only you know which ones are holding everything else up, the beliefs quietly underneath your biggest decisions.

That list is short.

Pick one and write down:

1. What do I currently believe?
2. What evidence would make me change my mind?
3. When did I last encounter that evidence directly?

Then go looking for it.

Not just from AI. From reality: a customer, a competitor’s numbers, somebody who resigned, a result your internal model says should not have happened.

Experience is not the liability here.

Unexamined experience is.

Which belief are you still operating on that you haven’t retested since the world changed around it?

#ContinuousLearning #ArtificialIntelligence #CriticalThinking #DecisionMaking #ExecutiveJudgment

Contact

bruno.gentil@sherpaconsultingasia.com

© 2026. All rights reserved.