The contagion of fear

In the early days of my university career, I made a regrettable confession to myself. Late one night, while in the computer science lab, I proposed a reckless idea to my fellow students: to spread a computer virus through the general computer lab. We went next door, alarming the humanities students with fake news of a spreading virus.

People panicked, ejecting floppy disks and powering off machines, creating chaos in the lab. We immediately realized our mistake but couldn't contain the fear we had unleashed. The consequences were severe, with students losing their work and the facilities director reprimanding us sternly. We wrote letters of apology, pledging to never repeat such a mistake.

Looking back, I feel compelled to share this tale, as I have witnessed a similar irresponsible fear-sowing with respect to AI and the threat of human extinction. Recently, ex-Anthropic employee Jacob Coxon and Anthropic Alignment Science lead Evan Hubinger claimed a ten percent chance of AI killing all humans within the next decade.

This statement is alarming, as it suggests that a newborn child's likelihood of being killed by AI before reaching middle school is greater than ten percent. These terrifying claims have gained significant traction, causing widespread fear and panic. The answers to how such an event might happen rely on speculative extrapolations into the future.

Coxon, while not an expert, is an expert in spreading fear, which is its own kind of evidence and often overwhelms contrary voices. As a computer scientist, I would like to offer a rebuttal from my own field of expertise. Building computers requires not only intelligence but also our character, actions, and reasoning about the physical world.

Robots cannot currently perform tasks that would lead to human extinction, and there is no time horizon consistent with these fears. The fear of AI's potential harm should not overshadow the physical reality of our limitations and the responsibility we hold in shaping the future.