Fei-Fei Li: Compare Two People's Experience of an AI Tool

The Lesson

A club is considering a tool that drafts short summaries of its discussions. The organizer wants less writing after each meeting. A participant wants their tentative suggestion to remain tentative. Both may want a useful summary, but they are judging different parts of the experience.

In the 2018 article Introducing Stanford’s Human-Centered AI Initiative, Fei-Fei Li and John Etchemendy argue that AI development should be accompanied by study of its effects on people and should enhance humanity. They describe a multidisciplinary effort extending beyond engineering. Li’s Stanford profile identifies her AI research and role in the human-centered AI institute.

That perspective invites a practical question: useful for whom? The person operating a tool is not always the only person affected by its output. A summary has a writer, readers, and people whose contributions it describes. A workflow can feel convenient from one position and confusing from another.

Begin with a simple walkthrough of the proposed experience. What goes into the tool? What comes out? Who reads, checks, corrects, or acts on it? You do not need a working AI system to discuss these questions. A paper example can make the proposed process easier to examine.

Then compare two perspectives without assuming either is universal. The organizer may care about the effort needed to prepare a summary. A participant may care about whether a suggestion appears as an agreed decision. Those concerns can coexist. They point to different questions you need to answer before judging the process.

Ask what each person wants to be able to do. Rather than asking only whether they like the idea, ask where they would check a detail and how they would raise a correction. Their answers can reveal a requirement that a polished demonstration did not make visible.

Keep a tradeoff visible when preferences conflict. An organizer may want to circulate the summary immediately; a participant may want time to check how their contribution is described. Calling the tool “helpful” does not resolve that difference. The group needs an explicit decision about review and timing.

This is a conversation about a proposed workflow, not proof that an AI system is accurate, accessible, or suitable for every setting. Use invented examples for the practice so you can explore perspectives without submitting anyone’s actual discussion to a tool.

Reflection

• Who operates the proposed tool, and who else experiences its output?
• What does each person need to understand or be able to change?
• Have I mistaken one person’s convenience for everyone’s benefit?
• Which tradeoff needs a group decision rather than a technical demonstration?

Practice

This is an original SelfGrowthVideos exercise. It is not an evaluation protocol prescribed by Li or Stanford and does not imply their endorsement.

Sketch a modest proposed workflow and make an invented sample output. For a discussion summary, include a suggestion, a decision, and an unresolved question.

Invite two willing people who bring different perspectives to walk through the same paper example. Explain that no real AI system is being tested. Ask each:

1. What would you use this for?
2. Which detail would you want to check?
3. What would you do if something were missing or wrong?
4. What would make you prefer the existing process?

Make a comparison note with three rows: shared needs, differences, and unresolved questions. Preserve the distinction between what participants said and your interpretation.

For example, the organizer may want a short decision list while a participant wants suggestions clearly labeled. Your note might identify a shared need for a readable summary and a difference about when it should circulate. A possible next step is to agree on a review process before testing any tool.

Two conversations provide local input, not a representative result. Keep room for other affected people whose perspectives you have not yet heard.

Review

Before the next project decision, return to the comparison note. Did the proposed workflow change in response to either person’s input? What remained unresolved? Who else should be included?

Choose one next step: clarify the process, consult another affected person, revise the paper example, or keep the current approach. Record why. Human-centered thinking becomes practical when people’s perspectives can change what you choose to build or use.

Go Deeper

Explore Fei-Fei Li’s teacher profile and the AI study area. If you have not yet identified a person and task, begin with Start an AI Project With a Human Need. For checking a belief against limited evidence, see Adam Grant: Test a Working Assumption.