Benedict Evans argues that predicting which jobs will be most exposed to AI is fundamentally impossible, drawing on historical parallels with past technology waves. Despite decades of accounting automation, the number of CPAs kept rising — illustrating the Jevons paradox, where cheaper tasks lead to more of them being done rather than fewer workers. He also highlights how businesses can be disrupted even when the core job itself isn't directly automated (e.g., journalism's business model collapsing due to the internet, not the craft of writing). A third problem is that job descriptions like those in O*NET are too incomplete to model accurately — the same 'Gell-Mann Amnesia' that makes people underestimate complexity in unfamiliar fields. Evans concludes that directional claims about AI job exposure may be truisms rather than predictions, and that any quantitative model should be tested against whether it would have predicted the CPA, newspaper, or Uber effects.
