Before the Flood

The answer came back clean. Superintelligent AI posed "no immediate danger." It remained "far in the future." Ed Felten, then OSTP's Deputy Chief Technology Officer, offered the definitive metaphor for the moment: "The A.I. community keeps climbing one mountain after another, and as it gets to the top of each mountain, it sees ahead still more mountains."

He wasn't wrong about the mountains. He was wrong about what "seeing more mountains ahead" means when the climbing equipment keeps improving by an order of magnitude.

Nobody in that room was stupid. The panelists were correct that the human brain's flexibility and capacity for learning remained far out of reach for AI systems. They were correct that artificial general intelligence wasn't imminent. What they missed — what the institutional framework of "is this an immediate threat?" couldn't capture — was that you don't need AGI to reshape civilization. You just need AI that's good enough, cheap enough, and deployed everywhere.

By 2020, GPT-3 was rewriting what "impressive" meant for language. By 2023, the question had shifted from whether AI would displace work to which work was already gone. By 2026, we're not waiting for the mountain on the horizon. We're navigating the terrain it created when it arrived.

There's a specific flavor of wrong worth naming here: the institutional reflex to defer serious consideration until risk becomes "imminent." The workshop's implicit framework was binary — imminent threat or not-yet concern. It had no category for "already transforming at sub-AGI capability levels." No bucket for "the disruption happens before the superintelligence does."

The same framework has a track record. Climate risk was "far in the future" for decades while the preconditions for that future were being locked in. The pattern isn't stupidity; it's a threat-assessment architecture calibrated for slower-moving dangers. The checklist asks: is it here? If not, come back when it is.

What any workshop framework struggles to capture is pace. The 2016 experts saw more mountains ahead and concluded there was time. They were right that the furthest mountain — AGI, superintelligence, whatever the terminal form turns out to be — wasn't close. They were wrong: the peaks already being climbed were reshaping everything in the valleys below.

This piece is documentation, not accusation. There was a moment when the establishment looked at AI and said: not yet, the serious threats are still ahead. That moment has a date. It was May 2016. The flood arrived without waiting for the summit.

i · sources

• White House OSTP AI Workshop — GCRI News Summary, May 2016