A small change in the title of a report led by MIT makes a big difference in interpretation

In one version of the report (link), the authors discuss the Iceberg Index, a measure of "where AI capabilities overlap with human skills before adoption crystallizes" in a sort of digital twin of the US labor market, "using Large Population Models to simulate the human–AI labor

market, representing 151 million workers as autonomous agents executing over 32,000 skills across 3,000 counties and interacting with thousands of AI tools." They characterize the Iceberg Index as a "skills-centered KPI for the AI economy. It measures the percentage of wage value of skills that AI systems can perform within each occupation, revealing where human and AI capabilities overlap." However imperfect, it is an interesting approach, especially at the level of tasks.

The key difference between tasks, or skills, and jobs, is that most jobs involve multiple tasks, some of which may be partially, or more rarely, fully, automated with intelligent algorithms or machinery. The second version of the very same paper (link) only differs, as far as I can tell, in its title: measuring "Workforce Exposure" as opposed to "Skills-centered Exposure." But computing fractions of jobs that are exposed to AI replacement does not translate into number of jobs at risk. The second title suggests insights that are out of reach.

And as Tom Davenport and Miguel Paredes, PhD argue quite convincingly in a timely article (link), there's no clear rule for when automatable tasks actually translate into job losses. To the question, "Can We Predict What Jobs AI Will Take?", their unequivocal answer is no! They posit that contextual factors like organizational growth, regulatory responses, the slow pace of AI adoption, and the time needed to redesign processes make rigorous quantitative prediction impossible.

Where the MIT report and Davenport and Paredes converge is in their view that society should focus on preparing workers for AI-related changes, training them in digital skills, teaching them to work alongside AI, redesigning jobs and processes, and helping people build meaningful lives if traditional employment arrangements shift. This would be more useful than predicting and often exaggerating negative impacts that may never materialize.