Friction Factors That Make AI Predictions Difficult

To accurately predict AI's impact, one must understand not only technical capabilities but also how the entire Ecosystem reacts when the technology is adopted. This is because even if one step in a specific workflow is accelerated, the overall throughput does not increase if a Bottleneck occurs at another stage.

Predicting how capabilities are applied to industry and society is much more difficult than simply determining what AI can do (Capability). In many cases, the gap between the ability to access information and the ability to actually act on that information is overlooked.

The case of Radiology is representative. In 2016, there were predictions that AI would replace radiologists, but in reality, demand actually increased. This is because while AI improved image interpretation capabilities, the work of radiologists involves much more complex processes beyond simple image reading, such as communicating with patients, analyzing clinical context, and supporting treatment decisions.

Ultimately, the failure of AI predictions arises from being buried solely in technical possibilities and failing to sufficiently consider the social incentives and the reality of workflows in actual work environments.