In the long run,
the facts are on the side of the optimists.
To paraphrase Martin Luther King Jr., the arc of history bends toward progress. But progress doesn't just happen. People work hard to discover ways forward.
The What Works Initiative
highlights positive outcomes on difficult issues – and how people achieved them.
Those of us who consult Claude, Gemini, or ChatGPT — which by now is almost all of us — know that our artificial consultants and coaches are tirelessly agreeable. They’re programmed sycophants, trained to please and take our cues as they answer our questions and respond to our prompts.
Is this a setup for digging us even deeper into our own separate rabbit holes of opinion — dividing us further?
Most AI experts, surveyed at the start of a recent study, believed the answer to be yes: AI, and its tendency to flatter and appease, will deepen the divides among us.
But the study went on to test the theory on 1,500 users across 30 different decision environments (types of topics) and found the opposite effect. The very tendency of large language models to flatter, agree, and support a user’s initial leanings made the users more receptive to contrary facts.
The upshot is that AI in fact has a depolarizing effect on its users. “Depolarization occurs across moral and non-moral, objective and subjective, strategic and non-strategic, and complex and simple tasks,” the study found.
It’s a new version of an older lesson: no one convinces anyone of anything by winning a debate. People listen when they feel like you’ve heard them and you get it.