Why the 2013 Oxford study can no longer tell you your AI risk
The single most-cited source for "will a robot take my job" is Frey & Osborne (2013), "The Future of Employment." It is a landmark paper — and it is now twelve years old, written before the transformer architecture that powers modern AI even existed.
It was built for the wrong kind of AI
Frey & Osborne modeled automation as physical and routine: machines on factory floors, self-checkout, self-driving trucks. It systematically scored creative and cognitive knowledge work as safe — copywriters, analysts, and radiologists all look low-risk in 2013. Generative AI inverted exactly those predictions.
Then vs. now
On every ReplacedYet occupation page we show the 2013 number beside our 2026 index so you can see how far the ground has moved. For routine knowledge work the gap is often 40 points or more — the clearest evidence that scores built before ChatGPT cannot be trusted for the AI question people are actually asking today.
Related occupations
Radiologist (33%) · Copywriter (62%) · Truck Driver (59%) · Paralegal (62%)