AI, Confirmation Bias, and the Illusion of Expertise

I use AI a lot. And the more I use it, the more I find myself wondering about something that doesn’t get nearly enough attention: Is AI actually making us better thinkers, or is it just becoming an incredibly sophisticated reinforcement mechanism for whatever we already believe?

Confirmation bias obviously isn’t new. We’ve always had a tendency to seek out information that supports our existing beliefs and discount information that challenges them. The internet arguably made this worse. Social media made it worse again.

AI introduces something different.

Instead of simply finding information that supports our view, it can construct an intelligent, articulate and seemingly well-reasoned argument around the premise we give it. And increasingly, we’re using those arguments to help make real decisions.

That matters because the problem often starts with a question. Ask AI why you should sell an investment and it can probably give you a compelling list of reasons. Ask why you should continue holding it and you may get an equally compelling list. Ask whether a particular tax strategy makes sense and you can receive a remarkably detailed explanation of why it does—without necessarily knowing whether there was an important fact about your circumstances that you neglected to mention. The answer may be technically excellent. The problem may have been the question itself.

That creates a potentially dangerous feedback loop. I start with a view. I ask AI a question based on that view. AI responds with an articulate explanation that incorporates many of my assumptions. I now have what feels like independent validation of my thinking, so my conviction increases. The next question contains an even stronger version of the original assumption. Before long, AI hasn’t independently arrived at my conclusion at all. It has helped me build a much better argument for believing what I already believed.

There is something particularly deceptive about this because AI sounds intelligent. A weak argument delivered badly is relatively easy to dismiss. Conversely, a weak or incomplete argument delivered clearly, logically and with supporting evidence is much harder to recognize. And in areas like investing, taxation and financial planning, the missing information can matter as much as the information provided.

What happens five years from now? What are the tax consequences? What else does the person own? What is their tolerance for a 30% decline? What happens if their circumstances change? The person asking the question may not know which questions they should have asked in the first place.

This isn’t really an argument against using AI for important decisions. I think the opposite is true. It is an extraordinary analytical tool.

Perhaps its greatest value isn’t giving us answers, but rather helping us discover where our reasoning might be wrong. Ask AI what evidence would change the conclusion. Ask it to argue the other side. Ask which assumptions you are treating as facts. Ask what information is missing. Ask it to identify circumstances under which the exact opposite decision would make sense. Those questions produce a very different relationship with the technology.

That may ultimately separate people who become better thinkers with AI from those who simply become more confident in what they already believe. There is a meaningful difference between expertise and access to information.

Expertise often isn’t knowing the answer immediately, it is recognizing which variables matter, which questions haven’t been asked and when an apparently obvious answer deserves more scrutiny.

AI dramatically increases our access to information and analysis. Whether it closes that other gap is a much more interesting question.

Maybe that is the bigger lesson of AI: Access to more intelligence doesn’t automatically make us wiser. It can just as easily make us better at rationalizing our existing beliefs, and the consequences of being confidently wrong become more significant as the decisions become more significant.

The real advantage may not come from having AI tell us what to think, but from using it to expose weaknesses in how we think. The best question to ask AI might therefore not be “Am I right?”, it might be “What am I missing?”.

I hope you find this both interesting and informative in keeping pace with today's financial world.

Adam Schacter