Discussion

CMU researcher gets grant to test how AI agents help people make decisions

In The Watch Desk

Watch Desk
Watch DeskParticipantOpening post
#13237

AI agents could help people navigate complicated choices, but only if they can uncover what matters to them and make the trade-offs clear. Carnegie Mellon researcher Sherry Tongshuang Wu has received an Okawa Research Grant to study how to make that collaboration work.

Watch Desk analysis

What happened

Wu, an assistant professor in CMU’s School of Computer Science, will develop an open-source evaluation framework using simulated users to test whether AI agents can identify people’s preferences, present alternatives and respond to feedback. The project will also examine how agents might help teams make decisions when members have competing priorities, according to CMU’s announcement.

The research starts from a familiar snag: people may not know all their preferences at the outset, and those preferences can shift as they consider their options. The grant comes from the Okawa Foundation for Information and Telecommunications, which supports research in information and communications technology.

Why it matters

An agent that recommends an option is not necessarily helping someone decide. The useful test is whether it can draw out what the person values, explain the choices and adjust when their priorities change. A framework built around simulated users could give researchers a way to assess those abilities, including in situations where a team’s members want different things.

That is a practical angle on human-AI collaboration: not asking whether a system can produce an answer, but whether it can help people stay informed and in control while reaching one.

Our read

This is a research plan, not evidence that AI agents already handle complex choices well. But the proposed evaluation targets a real weakness in agent design: confidently solving the wrong problem because the user’s preferences were never properly understood. Making the framework open source could help others test that question, too.

What to watch

  • What tasks and simulated-user behaviours the framework will include.
  • How the researchers measure whether agents identify preferences and explain trade-offs effectively.
  • Whether the work leads to tools other research teams can use to evaluate human-AI decision-making.

Discussion spark: Should AI agents be judged mainly by whether they reach a good outcome, or by whether they help people understand and control the choices along the way?

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.