A value outside the book's range turns yellow. The book excludes those values on purpose: for example, exposure growing by more than 60% means an AI able to do almost every task, which the book leaves out (chapter 18).
| Press "Compare policies" to run every policy with the current assumptions (1,000 simulations each). |
Share of the uncertainty in the result explained by each assumption (total Sobol index, 23,552 runs). From the book's own analysis (outputs/sobol.json), not recalculated here.
What runs here. The files of the aimodel package are loaded into Pyodide, a version of Python that works in the browser, and called without changes. With the book's settings (4,000 simulations, seed 20261010, no assumption fixed, status quo) the page reproduces the book's published results exactly. The status line above checks this every time the page opens.
How to read it. Unemployment and public debt are differences in percentage points from the world without AI. GDP and consumption are percentage differences. These are conditional results of the model, not forecasts. The Sobol analysis needs the SALib library, which does not run in the browser, so it shows the book's stored results.
Job by job. The Job Simulator applies the same logic to single occupations and countries, using published occupation data.