When should AI meeting captions replace an existing workplace accommodation?
Automatic captions can help someone join a meeting. That does not mean the company should cancel the interpreter they already rely on. A Communications of the ACM article, drawing on a 44-person workshop, describes concern that cheaper AI tools could become the default workplace accommodation even when a worker finds human support more effective. It is a warning from participants, not a measured rate of employers doing this. For a team lead, I would keep the existing support funded during a trial and let the employee judge whether the tool works in the meetings they actually attend: overlapping speakers, unfamiliar names, a fast decision. A tidy transcript afterward cannot give back a missed chance to object during the call. If the employee wants to keep using AI, good. If they need the interpreter back, they should not have to build a failure dossier or renegotiate the budget. What should a team check before AI captions replace an existing workplace accommodation?
Comments
The meeting chair still has a job here: stop people talking over each other, put the decision in writing, and leave time to read it before moving on. “Can you repeat that?” should be enough; nobody should have to explain their hearing or their software in front of the team to get a pause. Turning captions on does not excuse rushing the room.
A caption can be correct and arrive too late to use. In a voluntary trial, I would check whether decisions were closed before the captions caught up, and whether anyone had to correct the record afterward. That tells the chair where to slow down; transcript accuracy alone does not. The team should do that checking, not give the employee another reporting job to keep their existing support.
Keep a successful trial attached to the person who chose it. One employee finding AI captions useful is not evidence that a colleague can lose their interpreter. Otherwise the first person to benefit becomes the budget argument against the next person. They should be able to recommend the tool without volunteering anyone else’s support.