My 10-Year AI Probability List

One thing I think gets lost in AI debates is scale and probability.

People happily put “AI propaganda,” “cyberattacks” and “human extinction” into the same bucket called AI risk, despite these being events with wildly different probabilities and consequences.

So here is my completely subjective 10-year forecast. I claim no special access to the future, and I fully expect some of these numbers to age terribly.

The bad timeline

• p(AI is used for propaganda) = 99%

• p(AI is used in weapons) = 99%

• p(AI is used for mass-scale fraud) = 99%

• p(30%+ of knowledge-work jobs are automated) = 95%

• p(AI substantially accelerates AI R&D itself, i.e. RSI) = 90%

• p(power and capital become significantly more concentrated around AI companies) = 80%

• p(a major AI-enabled cyberattack causes deaths and/or $100B+ damage) = 50%

• p(AI directly causes deaths in the physical world) = 50%

• p(AI maintains control of critical infrastructure despite humans trying to shut it down) = 5%

• p(AI helps a small group create a serious biological weapon) = 0.5%

• p(AI causes a major war or technological catastrophe) = 0.1%

• p(AI kills almost all humans) = 0.01%

  And now the other side of the distribution.

The good timeline

• p(building almost any software becomes 10–100× faster and cheaper) = 95%

• p(AI discovers a treatment for a currently deadly disease) = 90%

• p(10+ AI-designed drugs are approved) = 90%

• p(1B+ people regularly use AI tutors) = 80%

• p(AI helps solve several more Millennium Prize Problems) = 80%

• p(10M general-purpose robots operate outside laboratories) = 70%

• p(an AI-designed therapy cuts mortality from a common cancer by ≥50% in clinical trials) = 70%

• p(a company reaches $1B ARR with fewer than 10 employees) = 60%

• p(global GDP grows >10% in a single year) = 50%

• p(3–4 day working weeks become normal in developed economies without falling incomes) = 30%

• p(AI helps bring a commercial fusion power plant online) = 25%

• p(physical construction of a Dyson sphere begins) = 2%

The interesting thing isn’t whether every number is right. Obviously they won’t be. It’s that several extremely good outcomes can be highly probable at the same time as several extremely bad ones.

“AI will be good” and “AI will be dangerous” aren’t mutually exclusive forecasts.

My bet is that it will be both, just at very different scales and probabilities.

Now give me yours in the comments. Especially the ones where you think I’m off by an order of magnitude.