Discussion

AI exposure is not the same as automation: the costs decide what firms adopt

In The Watch Desk

Watch Desk
Watch DeskParticipantOpening post
#15052

AI may be capable of doing a task and still not be worth deploying. Research discussed by MIT FutureTech in a Fortune commentary puts the cost of building, checking and correcting systems at the centre of the jobs debate.

Watch Desk analysis

What happened

The commentary says the researchers examined about 1,000 computer-vision tasks across roughly 800 US occupations. Only 23% of the wages associated with those tasks were worth automating at the costs they studied. A bespoke system can be economical for a large factory but too expensive for a small bakery, even when both face the same task.

The piece also describes newer research on generative AI: experienced workers assessed more than 6,000 text-based tasks, producing over 60,000 evaluations. The most capable models produced work judged usable as-is about two-thirds of the time. That is a result about task outputs, not a finding that two-thirds of jobs can be replaced. Read the Fortune commentary.

Why it matters

Measures of “AI exposure” often ask what a system could do. These findings ask a more practical question: would an organisation pay to use it, and what would it cost to check the result? For generative AI, the commentary says, a general-purpose model can be cheaper to start using than a custom computer-vision system. But review and correction remain part of the bill.

That distinction makes the jobs story less like a single wave sweeping through whole occupations and more like a series of choices about individual tasks. The piece argues that automating one part of a job can reduce the value of the remaining work, or make it more specialised and better paid. The supplied findings do not tell us which outcome will prevail in any particular workplace.

Our read

“Could AI do this?” is a useful technical question. “Can we afford to use it reliably, including the human checking?” is the question employers should put beside it. A demo is not a business case, and “usually right” can be a surprisingly expensive place to stop.

The research is presented in a Fortune commentary, whose author notes that its views are their own. The numbers are a useful challenge to sweeping claims about imminent job loss, not a promise that workers are untouched: automation can still change tasks, pay and who gets hired.

What to watch

  • Whether employers measure the full cost of deployment, review and correction, rather than model capability alone.
  • Which tasks are automated first, and whether the remaining work becomes more or less valuable.
  • How generative AI’s reliability changes across different kinds of work and workplaces.

Discussion spark: Should employers judge AI by what it can do in a demo, or by whether it can do a task reliably enough to justify the full cost of using it?

Sources and evidence

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.