Is value sensitive design only for AI?

No. It applies across technologies and predates many current AI applications.

Is it the same as user-centered design?

They overlap, but value sensitive design explicitly investigates values and includes indirect stakeholders beyond direct users.

Does it provide one fixed list of values?

It offers concepts and methods for identifying and examining values in context, rather than eliminating the need for judgment.

AI Glossary / Governance, ethics and safety / Advanced

Value sensitive design

Definition

Value sensitive design is an approach to technology design that systematically considers human values throughout the design process. It connects analysis of values and affected people with empirical investigation and technical design. The aim is to make value considerations part of concrete decisions, rather than an abstract statement added after development.

What it means

Values are concerns about what matters to people, such as privacy, autonomy, fairness, accountability, or well-being. Their meaning and practical implications require investigation in the particular setting. Naming a value does not determine the correct design by itself.
The approach considers direct stakeholders who interact with a technology and indirect stakeholders who may be affected without using it. For an AI meeting assistant, participants are not the only relevant people: someone discussed in a meeting may also be affected by the resulting record.
Value sensitive design distinguishes designers' own commitments from values expressed by stakeholders and values the technology explicitly supports. Making these distinctions visible helps prevent a team's assumptions from silently standing in for everyone's interests.

How it works

The approach integrates three kinds of investigation. Conceptual work clarifies relevant values, stakeholders, and tensions. Empirical work examines how people experience the situation and understand its implications. Technical work investigates how existing or proposed system properties support or hinder the values under consideration.
These investigations inform one another. A prototype can reveal a previously overlooked concern; interviews can challenge a proposed requirement; technical constraints can prompt exploration of a different design. They are not three boxes to complete once in a fixed sequence.
Value tensions require reasoning and, where appropriate, design alternatives. More detailed records might support accountability while creating privacy concerns. The task is to investigate those effects and justify choices, not assume that a single metric resolves every tension or that all values can be maximized simultaneously.

Three practical examples

A team designing an AI meeting assistant considers participants, meeting organizers, and people mentioned in discussions. It examines who can access summaries, correct errors, and understand what is retained.
An educational assistant offers personalized suggestions. Designers investigate whether students can understand and question recommendations, alongside the usefulness of those recommendations for instructors.
An internal analytics tool makes individual activity visible to managers. The team examines whether the intended coordination benefit requires that level of detail and explores alternatives that may support the task with less exposure.

Think of it this way

Think of designing a shared building. Decisions about entrances, visibility, private space, and access embody judgments about how people should live and work together. Technology choices can likewise support some values and obstruct others. Value sensitive design makes those judgments a subject of investigation before they become difficult to change.

Worked example

Consider a fictional AI meeting-summary project with 40 interview participants: 30 regular meeting participants and ten people responsible for managing shared records. Suppose 24 of the regular participants favor keeping detailed transcripts for convenience, while six express strong objections. All ten record managers favor retaining them.
Overall support is (24 + 10)/40 = 85%. The team could mistakenly treat that percentage as sufficient justification for the design. However, it says nothing about people discussed in the meetings who were not interviewed, the reasons for the six objections, or the actual consequences of retention.
A value-sensitive investigation examines those missing perspectives and reasons. The technical team prototypes alternatives, such as shorter retention, more limited access, and user review of summaries. It evaluates whether those alternatives support the intended work and how they affect the identified concerns.
The arithmetic describes the sample's responses; it is not an ethical approval threshold. Majority preference, stakeholder representation, technical feasibility, and justified treatment of affected people are related but different questions.

Illustrative case study

Fictional company Birch Meetings plans an assistant that creates searchable records of team discussions. Its initial requirements emphasize saving time and making previous decisions easy to find.
A stakeholder investigation identifies additional concerns: participants want to correct inaccurate summaries, people mentioned in discussions want inappropriate exposure considered, and record managers need clarity about responsibilities. The team distinguishes these perspectives rather than combining them into a single satisfaction score.
Designers explore access, review, and retention options. They test prototypes with relevant users and examine whether proposed controls work in practice. Findings lead to changes in both the interface and the process for handling disputed records.
This fictional case illustrates connecting values to operational choices. It does not establish a universal retention policy or claim that a particular set of controls settles every privacy or accountability question.

Common misconceptions and mistakes

“Listing values completes the method.” Values must inform investigation, requirements, design, and evaluation.
“The customer is the only stakeholder.” People who never operate the system can still be affected.
“Majority preference settles every issue.” Reasons, consequences, and excluded perspectives also matter.
“Technical design is neutral.” Access, defaults, data flows, and other choices can support or hinder values.
“The approach guarantees a perfect solution.” It supports reasoned progress while leaving room for uncertainty and revision.

Business use

Begin stakeholder and value analysis while meaningful design alternatives remain available. Keep a traceable connection between identified concerns, design requirements, and evidence about the resulting system. Document which perspectives were included and which remain difficult to represent.
Use prototypes to investigate value tensions rather than assuming that the first apparent tradeoff is unavoidable. For AI systems, examine how data collection, output presentation, feedback, and human review distribute control and consequences. Record the reasoning behind decisions and revisit it when the system's use or affected population changes.

Check your understanding

Why is 85% support insufficient in the worked example?

Answer: It does not resolve the reasons for objections, missing stakeholders, or consequences.

What connects conceptual and technical investigation?

Answer: Clarified values and tensions guide design choices, while technical findings can change the analysis.

Why revisit the investigation after deployment?

Answer: New uses and experiences can reveal effects that were not previously understood.

Related terms

Learn next

Choose an AI feature and identify both direct and indirect stakeholders. Select two relevant values, describe a possible tension, and propose a technical alternative whose effects could be investigated with those stakeholders.

Sources

Value Sensitive Design Lab, About VSD, theoretical constructs and methods: https://vsdesign.org/vsd/ Friedman, Kahn, and Borning, Value Sensitive Design and Information Systems: https://old.vsdesign.org/publications/pdf/non-scan-vsd-and-information-systems.pdf

Previous: Value of perfect information Next: Value symmetry