Why does AI always agree? And how to fix it?

AI keeps nodding along — it praises the idea, agrees with the argument, finds value where there is none. Usually that is merely tiring.

But the moment you ask it to score something, and then make decisions based on those scores, the habit of agreeing turns into a problem.

I built myself a tool with an AI agent that monitors my work activity on the laptop and in cloud services, and then generates topics to write content about.

The content has to be useful for my target audience, so I asked the AI to score every topic by how interesting it is for executives, business owners and marketers.

And there it was: a purely technical topic about a bug in code got 78 points out of 100 for the business audience. In other words, the AI believed this topic should be highly interesting to a business owner.

In my new article I break down where this score inflation comes from, which of the obvious fixes I tried with no result, and which techniques actually help.

The topic is broader than content. Exactly the same shows up in lead scoring, triaging inbound requests, evaluating applications and tenders — anywhere the AI assigns a score and you make the decision based on it.

How AI inflates its scores and what to do about it (in Russian)

Subscribe to my Telegram channel: online_data_analytics