When public debate turns to artificial intelligence catastrophes, imagination usually paints a dramatic, cinematic picture: rogue superintelligences, autonomous killer robots reminiscent of The Terminator, or AI-synthesized bioweapons. Warnings from industry insiders—such as former Anthropic employee Evan Hubinger’s estimate that there is a greater than 10% chance AI could cause human extinction within a decade—often focus on these explosive extinction risks.

However, researchers and industry observers are drawing urgent attention to a far subtle, yet equally existential threat to humankind: Gradual Disempowerment.

Speaking in an interview on CBS News, Jay Choy, Founder and CEO of RoboCurve, explained why this overlooked trajectory requires immediate focus and resources.

1. What Is “Gradual Disempowerment”?

Unlike conventional “rogue AI” scenarios, gradual disempowerment requires no malice, no rogue agents, and no bad actors taking over the world.

Instead, disempowerment occurs as AI systems steadily become better, faster, and significantly cheaper than human beings across cognitive and physical tasks. As societies and market forces optimize for efficiency, humans incrementally surrender critical thinking, labor, decision-making, and economic agency to automated agents.

“In gradual disempowerment, there’s no villain to stop. Nobody has to make a mistake. AI just needs to keep getting better, cheaper, and faster, and they will slowly replace every human out there.”

— Jay Choy, CEO of RoboCurve

2. Historical Analogies: Horses, Cars, and Neanderthals

To explain how a species or workforce can be displaced without a war or single catastrophe, Choy pointed to two powerful historical parallels:

The Horse and the Automobile

When automotive technology matured, society did not hate horses, nor was there a war between cars and horses. Cars simply proved far more efficient for transport and commerce. Once cars became “good enough,” economic demand for working horses collapsed.

Neanderthals vs. Homo Sapiens

For thousands of years, Neanderthals and Homo Sapiens shared territory across Europe and Asia. Neanderthals were not wiped out in a single grand conflict. Instead, because Homo Sapiens were slightly more energy-efficient and adaptable, Neanderthals were gradually displaced over generations until they disappeared.

If AI agents become fundamentally more efficient and capable across every domain, humanity risks facing a similar gradual displacement.

3. Why Is This Threat So Severely Neglected?

Despite its likelihood, gradual disempowerment receives a disproportionately small share of safety research and regulatory attention compared to acute risks like cyberattacks or bioweapons.

AI Risk Model	Narrative Structure	Primary Driver	Public Perception
Rogue AI / Cyber / Bioweapons	Acute catastrophe with a distinct villain or bug	Malice, loss of control, or systemic flaw	High media attention & film tropes
Gradual Disempowerment	Slow economic & cognitive erosion	Market incentives, cost reduction, efficiency	Often ignored or misclassified as routine tech adoption

Because there is no dramatic moment of attack, society “sleeps at the wheel,” handing over critical capabilities step by step until human oversight becomes obsolete.

4. It Is Already Happening: Early Economic Signals

Choy emphasized that gradual disempowerment is not a speculative theory for the far future—it is already taking shape within frontier AI research labs themselves:

• 75% Task Automation Target: An internal survey among employees at Anthropic revealed that staff expect AI will be capable of performing 75% of their job duties by the end of the decade.
• Shift in Hiring Signals: At Anthropic, the proportion of junior staff hiring dropped sharply from 27% to just 9% in less than two years.
• The Competitive Incentive Trap: In a global market economy, if competing firms use AI agents because they are faster and cheaper, businesses face immense pressure to stop hiring human labor.

Conclusion: Re-Allocating Attention to Human Agency

The threat of gradual disempowerment suggests that humanity’s greatest AI risk may not be a violent rebellion, but rather our own voluntary surrender of economic utility, cognitive skills, and autonomy. Avoiding this future requires shifting safety resources toward preserving human-in-the-loop governance, maintaining human workforce incentives, and actively preventing systemic over-reliance on autonomous systems.