AI Automation Rate Rises From 2.5% To 16.1% In Under A Year

In the economics literature, task based models treat jobs as bundles of separable tasks. They ask: "which tasks can machines do?" then sum up exposure scores. If 7 of 10 tasks are automatable, the job is "70% exposed." This has been correctly criticized for being overly simplistic and not accounting for O-ring production (amongst other things).

The critique is the following: workers have fixed time they allocate across tasks. If you're doing 10 tasks in 40 hours, each gets 4 hours. Automate 5 tasks, and now each remaining task gets 8 hours, ie doubling quality on those tasks. So if you automate complementary tasks but keep the human on bottleneck tasks, you've scaled the value of those bottleneck tasks and the worker captures a share of higher output.

There's an important weakness to this however. The implicit assumption is that someone wants to buy the high-quality O-ring output at whatever price emerges. But if automation creates a "good enough" substitute at dramatically lower cost, the market for the premium O-ring product could just evaporate.

Imagine lawyers automating document review and discovery - hurray they can now concentrate on strategic bottlenecks, raising their value. But if easily accessible contract drafting through models captures most of the market because clients want "good enough at 10% of the cost," the market for high-quality human-mediated legal work shrinks even if the remaining lawyers become notionally more productive. So in principle they could well end up losing out in the long run still.

Put differently, consumer demand and preferences can change. Even if the fancy sourdough is better, most people are fine with cheap supermarket bread. This doesn't have to be the case across the board of course and depends on the specific context; but it seems worth bearing in mind. Regardless, and importantly, consumer welfare rises in either scenario.