"If we get this wrong we all die. If we get this right we all lose our jobs." This comment was relayed to me this week by a fellow attendee at the Center for Human-Compatible AI's annual workshop. The quote comes from someone about as close to the frontier as it's possible to be.
On Friday, I went on ABC prime time live, talking with Rebecca Jarvis about whether we should pause frontier AI development. I'll save you the suspense: I said we might need to. That's a change for me. Until recently, I didn't think a pause made sense.
The technology isn't what changed my mind. This capability has been coming for a while but the context has changed. The people building these systems are saying two things at once. On one hand, we have no choice but to keep going. On the other, this is likely to be really bad for most people, though they themselves will be fine. When the same individual holds both of those, yet claims they are unable to do anything and "society" needs to solve this, I think, yes, they should pause at the frontier.
By pause I mean the absolute edge, and only the edge. Behind it, the work should speed up. The frontier has raced ahead of our collective ability to use these tools well, and it's entirely sensible, indeed rational, to hold steady while the rest of us catch up.
Could we actually pump the brakes? A huge number of people work on AI worldwide, but the decision sits with a small group who all know each other. It is not an exaggeration to say they could agree to this over dinner. The way sensible people describe these leaders matters: they self-identify as powerless, carried along by the race, and yet they hold all the power. So why don't they use it?
The incentives they work under don't support it. Which means the big question is what the rest of us can do to change those incentives.
The labs are flip-flopping in public right now. Will AI cause mass unemployment or not. Is the next model amazing or too unsafe to ship. Should they race ahead for profit or pause along with everyone else. There's no steady vision that makes sense to the rest of us. Those at the CHAI annual meeting were split on where the danger even sits, roughly half pointing to frontier models and half to open source, with cyber risk tipping it back toward the frontier. And the usual public defense of the race, "but China," drew little but dismissal from the people closest to the work, who described actually collaborating with Chinese researchers who care about safety too. I take their read on that over the version sold by the loudest voices furthest from the work, and I know that could be naive.
I worry too about bringing knives to a gun fight. When the safety field began, the researchers studying these systems and the companies building them worked on roughly the same plane, with access to roughly the same technology. That's gone. The academic work I see mostly—not always—runs on older, smaller open models, while the labs talk about enormous compute and systems that improve their own code. The people best placed to scrutinize this from the outside have a fraction of the access of the people building it. Independent oversight was always a condition for trusting any of this, and it has eroded. One thing the rest of us can push for is funding and access that lets independent researchers keep up. Maybe that's a real place to start—that safety access doesn't have to pivot on safety researchers leaving their independent posts.
Rebecca asked me: who's missing from the conversation? The seat that's empty belongs to the people imagining something different. Today the table holds those who want to ship and those who want to stop. Missing are the people with other ideas about what these products could be. Instead of building AI to replace humans, we could build AI for humans. People want meaningful work and real connection. They look at AI and see companies building products that deliver the opposite.
That empty seat is the one we're trying to fill at the Artificiality Institute. We work to understand what makes AI good for people, and then to show how to build those products well.
Here's what you can do in the meantime. Stay human and stay the author of your own mind. It is human to use technology to further ourselves and make our lives better, and AI can do that. So we need to understand what makes us collectively more intelligent, and refuse to let this race turn into collective stupidity. It starts with each of us. Notice how you use AI. Choose consciously. Keep showing up for others. Hold the edge steady and let the rest of us catch up. That is what matters, and it is what we need AI to support. Cognitive sovereignty over cognitive surrender.
More soon. I've also written a longer piece on what I saw at a closed safety meeting that pushed me to this view, and I'll send it your way shortly.
Artificiality Summit 2026
Join us October 22-24, 2026 in Bend, Oregon for 2.5 days with a fantastic group of speakers—academics, authors, designers, investors, photographers, and more.
We'll explore the theme of Unknowing—not ignorance, but a necessary release of inherited assumptions. We don’t yet know what AI will become, and we don’t yet know what we will become in relationship to it.
Unknowing is the space between—the place where neither side is fixed, and something new can emerge.
Rethinking what AI does to human work—and the index behind it
In 2016, we found that unpredictable work resisted automation. That held for a decade, then generative AI started handling unpredictability on its own. The barrier instead turned out to be irreducibility: work whose parts each depend on all the others at once, so it can't be pulled apart and solved piece by piece. We built the Irreducible Complexity Index to measure it, scoring 894 US occupations on five dimensions of resistance—whether it needs abody in a place, whether someone must own a contestedcall, how many knowledge domains it braids together at once, whether it requires solving problems nobody has posed yet, and whether it demands a sense oftaste no rubric can capture. The findings invert standard career advice and point toward AI that extends human judgment rather than replacing it. Read the full research on the IRX and what it means for your work, here.
We particularly appreciate reader support for this work, it was a big lift and we hope it can help shift the dialogue to human-supporting AI rather than human-replacing. We also hope it helps shape important conversations with kids, bosses, and coworkers. Support Our Work
Follow Us: our Insta is going off and TikTok has some spicy comments.
Our video production is up, especially short videos. Follow us on your favorite channels—and please like, share, and repost to help us spread the word.
Helen Edwards is a Co-Founder of Artificiality. She previously co-founded Intelligentsia.ai (acquired by Atlantic Media) and worked at Meridian Energy, Pacific Gas & Electric, Quartz, and Transpower.
