What Happens When We Ask AI for Feedback?

Have you ever asked your refrigerator what you could do to improve your working relationship?

I haven’t. But while following the Thinkers50 London Summit yesterday, I encountered research discussed by Keith Ferrazzi that made the question seem rather less ridiculous.

His research team interviewed people who regularly work with AI agents. They also asked the agents themselves: what makes humans difficult to work with?

Ferrazzi describes eight types, including these three:
• The vague requester: asks for help without making clear what needs doing.
• The context hoarder: provides the assignment but leaves out the background needed to understand it.
• The overdelegator: leaves decisions to the agent, then objects to the choices it makes.

Uncomfortably recognizable. Perhaps our human colleagues have been filling in more gaps than we realized.

But what kind of relationship are we imagining when we ask an AI agent how it experiences working with us?

Its answer can be useful without being an account of lived experience. An agent saying that we frustrate it is generating a response to our question. That response still needs examination.

There is a dilemma here. Treating an agent as a collaborator may encourage us to explain ourselves better, invite challenge and reconsider our habits. The same language may encourage us to assume shared understanding and grant its responses more authority than they warrant.

The French philosopher Bernard Stiegler examined how technologies reshape our knowledge, attention and capacities. His work offers a useful way to consider this exchange.

Asking an AI agent for feedback might help us notice habits we otherwise overlook. But if we repeatedly accept its assessment of what makes us a good collaborator, we may also begin organizing our behavior around the system’s expectations.

We invite AI to challenge our judgment so we can improve it. Yet we may gradually defer to its judgment in the process.

I can see value in asking an agent to challenge how I work. I would also want it to show which exchanges support its assessment, explain what it might be missing, and leave room for disagreement.

The interesting possibility is that this exchange helps us become more thoughtful about how we work, with technology and with each other.

Even if my refrigerator eventually has something to say.