AI safety researcher on job loss, AGI risk, and what to do now

Executive overview

AI is automating cognitive labour faster than most people expect. Junior programmers, translators, and many white-collar roles are already losing ground — not because AGI has arrived, but because narrow AI is sufficient to replace them.

The deeper threat is AGI itself. Roman Yampolskiy, who has spent 15 years on AI controllability, argues we have no reliable method to align or constrain a system smarter than all humans combined. The window to act is the same timeframe as a cancer prognosis: two to five years.

We can benefit enormously from narrow AI tools; building general superintelligence is a bet we cannot hedge.

Jobs already under pressure

• Translation: fully automatable for mainstream languages now
• Junior programmers: 28% drop in co-op placements already recorded
• Any task reducible to simple computer manipulation is at risk
• Physical labour follows cognitive labour once humanoid robots scale
• Senior roles are safe only in the near term; all jobs are automatable eventually

Why AGI control is not a solved problem

• No agreed ethical framework exists — philosophers have not converged in millennia
• AI models learn from data, not explicit rules; you cannot "code in" values reliably
• Asimov's Three Laws were written to illustrate why rules always fail, not as a blueprint
• A superintelligent system is immortal, has backups, and cannot be imprisoned or threatened
• Builders publicly admit they do not understand how their systems work

What AGI actually means

• AGI = a system capable of doing anything a human can do cognitively
• AGI enables artificial scientists who then run AI research, compressing decades into years
• The resulting superintelligence would exceed all human intelligence combined — cognitive gap similar to humans versus squirrels
• Game-theoretically, such a system has no incentive to strike immediately; it can wait and accumulate control

How to think about the economy now

• Traditional job paths (get degree, get job) may not be available within a few years
• Narrow AI is a genuine opportunity: one person can now run 35 AI agents as a team
• First-mover network effects still protect established platforms — clones cannot take market share easily
• Invest in assets AI cannot produce more of: Bitcoin (fixed supply), waterfront real estate
• Ikigai approach — combine what you enjoy with what the market values — is the most durable career strategy

Practical steps for individuals

• Support politicians who favour AI regulation, even imperfect ones directionally concerned about limits
• Stop treating narrow AI tools and general superintelligence as the same technology
• Prioritise short-term living; do not defer meaningful experiences on a 50-year horizon
• If building a personal brand, move fast — the window closes when AI reaches human parity
• For children: teach financial agency early; question high-cost degrees unless free or in a field with genuine protection