I have taught every age from preschool to graduate school, and while this post may seem a little adjacent to what I usually write, I couldn’t resist because I read it and was reminded myself that a dumb, legitimate question can turn out to be the smartest discussion starter. Politico asks Signal Foundation President Meredith Whittaker questions “about the hype, risks and data-privacy threat of AI.” Before founding Signal (an encrypted-messaging app), she worked for Google.

When you talk to policymakers now, whether about energy or national security or economic competitiveness, AI is inevitably part of the equation. What do you make of how big an impact AI is having?

I would dare you or anyone listening who has contacts with policymakers and politicians to just sit them down and say: What do you mean by AI? I think what you’ll get at that point is a lot of hype, a lot of fog, a lot of magical thinking. And that’s a big problem. We are seeing a wave of hype washing over critical institutions, governments, and key decision makers to trust these technologies with key functions that those who understand the technical reality, the limitations, the conditions for how these actually work would never have advised.

What’s the antidote to that? You also always hear this argument that Washington or policymakers don’t understand the technology well enough to regulate it, or put guardrails on it.

That old trope that all you need is tech brains in Washington to move aside the dusty policymakers and get things on the rails of modernization has been around for a very long time. But they’re not too old or too crusty to understand the domains in which they operate, be that education or health care or national security. And tech has a lot to learn on the fundamentals of those domains.

The antidote — there’s no one weird trick here, but just be brave enough to ask the dumb question. People are deeply afraid of being humiliated for being dumb about AI. And I will hear NATO chiefs, I will hear CEOs of Fortune 100 corporations, repeating as received wisdom claims about AI that make absolutely no sense.

These quote-unquote stupid questions, like, “How does this work? Do we have control over the data? What are the privacy implications? Are there vulnerabilities there?” These are just basic questions that should be the floor before entrusting critical decision making to obscure systems that often don’t, in my opinion, meet that bar for safety use in critical domains.