# MIT Researchers Say AI Adoption Depends on Cost Alongside Capability

By Token Post

Canonical URL: https://www.tokenpost.com/news/insights/62077
Published: 2026-11-20T13:38:30.000Z
Updated: 2026-11-20T13:38:30.000Z
Section: Insights & Views

> A study of about 1,000 computer-vision tasks found that 23% of the wages associated with them were worth automating at the costs considered.

Businesses weighing artificial intelligence for workplace tasks need to consider the cost of building, maintaining and checking systems alongside their capabilities, MIT FutureTech researchers Damian Saccocio and Neil Thompson argued in a commentary published Friday.

Their research on about 1,000 computer-vision tasks across roughly 800 U.S. occupations found that only 23% of the wages associated with those tasks were worth automating at the costs considered. The researchers illustrated how fixed costs can make automation worthwhile for a large operation while leaving a smaller business better off keeping the work with employees.

Generative AI can be less expensive to start using because companies can access general-purpose models priced by the token instead of building custom systems. But businesses still need to account for how often model output requires review or correction.

Experienced workers graded AI output on more than 6,000 text-based tasks from the Labor Department’s occupational database, completing more than 60,000 evaluations. The most capable models produced work usable as-is about two-thirds of the time, the researchers wrote.

That success rate means companies must determine which tasks AI handles consistently and where employees need to check or finish the work. The researchers described the move from usually correct results to reliable performance as often the most expensive step.

The effect on wages can depend on which parts of a job are automated. The authors pointed to GPS lowering the expertise needed by taxi drivers, while software that took over routine proofreading checks left proofreaders with more demanding work.

Saccocio is strategic partnerships coordinator at MIT FutureTech and an adjunct professor of Technology Strategy at Georgetown University’s McDonough School of Business. Thompson is director of MIT FutureTech and a principal research scientist at MIT CSAIL and the MIT Initiative on the Digital Economy.

The authors urged business leaders to evaluate AI’s full cost per task, including review and correction. They also said policymakers should track cost and reliability alongside capability, and that workers and employers should consider which parts of a job AI takes over.


