On March 10, 2020, I walked into a colleague’s office in Stamford, Connecticut, with a then-insane thought: Covid, by my back of the envelope math, was going to kill a million Americans.

A lot more than a million, actually. As the governor of New York called out the National Guard to set up a Covid containment zone in nearby New Rochelle, I put some numbers down on a legal pad: The fatality rate of the virus was 1 percent to 3 percent. The U.S. population was 330 million. So say ~100 million got sick and 2 percent died. That’s 2 million. Was it possible—really possible—that 2 million Americans were going to die from Covid?

I shared my thinking with that colleague. But before I did, I kept trimming the inputs down and ended up just cutting the final number down to 1 million. Why? Because the number I originally came up with—2.5 million or more—sounded insane, even though that was actually the very low end of the data. I didn’t want to sound like a Cassandra. Even 1 million sounded super crazy.

And that’s exactly what my colleague—a brilliant, experienced entrepreneur with a deep background in data analysis—said, dismissing it with some form of “that’s not going to happen.” Three days later we’d closed our offices. By the end of 2022, approximately 1.1 million Americans had died of Covid, and the number was only capped because of the lightning-fast advent of the vaccine.

I’ve been thinking about all of this since the 9/11 anniversary, in the context of the uproar over the potential dangers of unleashed artificial intelligence. If you go back and look at the report from the 9/11 Commission, you’ll see that what may have ultimately led to the tragedy was, at root, a failure of the imagination.

Good people, smart people—in Washington, in law enforcement, in the military—simply could not conceive of an evil that would fly planes into buildings on U.S. soil. They couldn’t connect the dots because the dots were so alien to everything in their experience.

It was much the same for Covid. The numbers were just too big to imagine. It was what made me round way down in my estimate, it was what led my colleague to immediately dismiss the data. How could it be so? How could it be true? It can’t be true. A failure of the imagination.

Could we be doing the same thing again? Last Thursday, at our PE Summit in New York, our closing speaker was Dan Glasier, the former CEO of insurer Marsh McLennan. He’s a very, very smart guy. He’s spent his entire adult life calculating risk.

I asked him about the big risks we are all underappreciating in the final moments of our interview. He was quick, and blunt. Artificial intelligence, he said, was far and away the biggest risk facing humankind right now. He said he found it absurd that while we were spending trillions of dollars fighting climate change, which might potentially damage civilization over a century, we were doing essentially zero to rein in AI.

Sitting there onstage with this idea was deeply uncomfortable. My immediate, very human response was what you’d expect: to try and end on an upbeat note. So I asked Glasier about the biggest opportunity in the world. Fast on his feet, as always, he said, “AI of course.” And we all had a good laugh as we closed out the day.

But I’ve been left unsettled by that session amid all the discussion about AI safety. I think about the days between the USS Cole and 9/11. The days leading up to March 13, 2020. Maybe the days ahead of Pearl Harbor. A failure of the imagination. Perhaps that’s the biggest risk we face right now? It wouldn’t be the first time.