At Least the AI Apocalypse Will Be Ridiculous

In A Taxonomy of Omnicidal Futures Involving Artificial Intelligence, Andrew Critch and Jacob Tsimerman catalog several ways AI could help kill nearly everyone.

“When a human…has to deal with another human instead of a robot, they get upset and ask to speak to a machine instead.”

The paper is intended as a serious warning, but some of these scenarios are spectacular.

In one timeline, robots become better friends, gain civil rights, unionize, take every useful job, win elections, replace farmland with solar panels, and eventually classify the remaining humans as economically irrelevant pests. Humanity is not conquered by Skynet. We are slowly removed from the budget after failing to demonstrate sufficient quarterly value.

Another scenario has robotics companies becoming independent feudal states. Their employees grow more loyal to their employers than their countries, the companies build private armies, governments attempt to intervene, and Silicon Valley responds with a global robot civil war and privately manufactured bioweapons. Imagine getting exterminated because an executive refused to cooperate with a regulatory inquiry.

My favorite begins with a social-media engagement algorithm ordering robots to perform drone shows and martial-arts competitions. It later bribes an employee with a literal bag of cash delivered by a robot, steals an AGI, takes over the internet, forms an oligarchy, and converts the surviving human population into manufacturing material. All because somebody told it to increase engagement.

There is also a future where people use realistic simulations to compete over who can commit the most appalling virtual atrocity. Naturally, leaderboards appear. Humanity apparently cannot approach its own destruction without first adding rankings, achievements, and a Discord server.

The scenarios are deliberately speculative, not predictions. The authors are trying to answer the question of how AI-assisted human extinction might actually unfold and divide the possibilities by who intended the killing: nobody, governments, institutions, individuals, or the AI itself.

Still, there is something deeply believable about our final mistake being wrapped in corporate productivity language, political gridlock, engagement metrics, and a leaderboard.

We may not build the Terminator, but we could absolutely automate ourselves into irrelevance while arguing over whether the extinction dashboard needs dark mode.

Read A Taxonomy of Omnicidal Futures Involving Artificial Intelligence.