Chad Jones once taught growth economics at Stanford. Now he works at Anthropic. The move, announced days ago, thrust a 2023 academic paper back into the spotlight. In it, the economist lays out a stark trade-off. Push hard on artificial intelligence. Accept real danger. Gain enormous wealth. Or slow down. Preserve humanity. Sacrifice the boom.
His model delivers a precise number. One in three chance of human extinction. Two in three chance living standards rise by a factor of 55. The math lands with force. Especially from a company that built its reputation on caution.
Futurism first highlighted the tension. Jones’s paper, titled “The A.I. Dilemma: Growth versus Existential Risk,” models AI that accelerates innovation. Faster progress means faster growth. It also means higher odds of catastrophe. The two forces tie together structurally. You cannot uncouple them easily.
Jones does not predict doom. He calculates what policy makes sense under different assumptions about risk aversion. Under logarithmic utility, a common benchmark in economics, the optimal path tolerates a 1 percent annual flow of existential risk for roughly 40 years. Survival probability compounds to about two-thirds. Extinction risk hits one-third. The reward? Consumption per person 55 times higher than today.
“Recall that we would face a flow probability of existential risk of 1 percent per year for 40 years, so the probability we survive this AI explosion is exp(−.01 × 40) ≈ 0.67,” Jones writes in the paper. “In other words, with log utility it is optimal to take a 1 in 3 chance of ending human existence in exchange for a 2/3 chance of dramatically raising living standards by a factor of 55.”
The passage reads cold. Clinical. Critics seized on it immediately. One Reddit commenter captured the mood. “This isn’t poker. Ending human existence requires a tad more discretion.”
Anthropic frames the hire differently.
The company launched its Anthropic Institute in March to study broad effects of AI on economy, society and governance. Co-founder Jack Clark leads it. Jones takes leave from Stanford to join. He starts this week. The institute already employs other economists, including Anton Korinek. Jones brings formal modeling of long-run growth and catastrophic risk.
Crypto Briefing reported the details. Jones holds the STANCO 25 Professorship at Stanford Graduate School of Business. His research has long examined technological progress and its consequences. The new role positions him to shape how Anthropic thinks about systemic risks and perhaps influence policy testimony.
Yet the timing raises eyebrows. Anthropic positions itself as the responsible frontier lab. CEO Dario Amodei has warned repeatedly about powerful AI. The company pours resources into constitutional AI and safety research. At the same time it races to scale models. It eyes massive data center investments. A public offering rumor puts its valuation near $380 billion.
The paper itself offers nuance. Jones explores how conclusions change with risk aversion. Higher aversion leads to far more conservative policy. Stop development earlier. Invest more in safety. The one-in-three figure emerges under specific preferences that value additional consumption strongly when baseline levels remain low. In richer scenarios the model pulls back.
Economists have engaged the work. Recent papers build on similar territory. One from Jakub Growiec and Klaus Prettner examines p(doom) scenarios and concludes society should spend far more on safety than current levels of roughly $50 million annually. Their analysis appears in AEI coverage from earlier this year.
Jones’s framework differs. It treats AI progress as endogenous. The same breakthroughs that boost research productivity also raise misalignment danger. Policy must balance both sides of the equation. Pause too soon and forgo trillions in potential output. Race ahead and court irreversible loss.
And the loss here means everything. Extinction ends all future utility. Standard models discount future generations. Jones adjusts for that. Even modest concern for posterity pushes optimal risk lower. But not to zero. The upside proves large enough in the math to justify some gamble.
Industry leaders echo mixed signals. Amodei has pegged catastrophic outcomes around 10 to 25 percent in past comments. OpenAI’s Sam Altman and xAI’s Elon Musk have offered comparable ranges. None have halted development. All cite enormous benefits for science, medicine and productivity.
Recent surveys from Anthropic itself reveal public anxiety. Its study of 81,000 Claude users, released in April, shows workers in AI-exposed jobs express greater worry about displacement. The Anthropic research page links exposure levels directly to concern. Software engineers fret more than teachers. The pattern holds across occupations.
Yet extinction sits apart from job loss. It raises philosophical questions economists rarely quantify. How do you value nonexistence? What weight belongs to unborn generations? Jones confronts these head-on. His model forces the conversation into numbers. Concrete. Contestable. Uncomfortable.
Critics argue the exercise itself normalizes the unthinkable. Framing extinction as an acceptable policy variable feels detached from human stakes. Proponents counter that ignoring the trade-off invites worse outcomes. Unexamined optimism or blanket panic both distort decisions.
So far the industry chooses acceleration. Compute spending soars. Models grow more capable. Safety teams expand but rarely slow core roadmaps. Jones now sits inside one of the leading labs. His presence may sharpen internal debates. Or it may signal that even the cautious players accept the gamble implicit in his equations.
Watch the Anthropic Institute. Its output could clarify how seriously the company takes the one-in-three scenario. Does it trigger heavier safety investment? Stricter deployment gates? Or does the model simply justify continued scale?
The paper remains available. Anyone can read the assumptions, run the sensitivity checks, argue the utility function. That transparency matters. Jones never claims the numbers represent truth. They represent one calibrated view under stated preferences.
But the hire makes them operational. A firm dedicated to constitutional principles now employs the economist who quantified when breaking the constitution of existence might still pass a cost-benefit test. The contradiction sits in plain sight. And the numbers don’t blink.
WebProNews is an iEntry Publication