Class 6 - Ethics & Stakeholder Analysis
Date: September 21, 2026
⚠️Homework
Homework 2 due September 27, 2026
🆕New Skills
First week for skills: 0
Skills: 10, 11
Notes (review before class)
Adapted from the CS 2100 lecture notes on ethics: stakeholder-value matrices, and bias and privacy.
Why Value-Sensitive Design?
- Computer science is a group activity: this course gives you the vocabulary of moral language, to talk with teammates and advocate for what is right.
- Value-Sensitive Design helps us identify and defend privacy issues, against "I have nothing to hide": its stakeholder-value matrices show who is at risk because they do "have something to hide."
- It comes first, with the requirements: it shapes what a program should do,
before any code or data. Without it, other values take over:
- Social media was built to connect communities, but its company (also a stakeholder) earns more the longer users look, at the cost of their mental health.
- reCAPTCHA (verifying users are human) and Duolingo (teaching languages) also train AI on users' data; and Duolingo's finances favor viewing over learning.
Steps for Value-Sensitive Design
- Identify the stakeholders: the people affected by the software, in any way.
- Identify the values at stake for those stakeholders.
- Fill out a stakeholder-value matrix: a row for each stakeholder, a column for each value, and in each cell, how that stakeholder's value relates to the software.
- Identify and analyze conflicts in the matrix: for each, decide whether and how to address it.
Case Study: Finder, a Dating App
Step 1: stakeholders of "Finder," a fictitious dating app:
- Users, who are many groups: nonbinary and transgender people, people who date all genders, people whose sexual orientation is criminalized where they live, people not legally allowed where they live, sex workers, survivors of intimate partner violence, blind people, non-English speakers, people with apostrophes in their names, ...
- Beyond users: content moderators (exposed to harmful content), software engineers and other employees, advertisers, actors and influencers in ads, governments and regulators, and anyone who dates (apps changed dating for everyone), and their friends and families. More will appear over time.
Step 2: values. Values shape technology, acknowledged or not, and technology changes our practices and values in turn.
- Privacy (key in this course): hiding one's identity from a government, or location from an abuser. Safety can be separate, or part of privacy.
- Accessibility differs by user: a nonbinary gender option; screen readers and translation apps; entering one's actual name.
- Fairness only happens if prioritized: recommendations should use only users' own preferences.
- Also financial cost (to advertisers and users), reputation, and laws and regulations.
Step 3: the matrix (part of it):
| Privacy | Accessibility | Financial cost | Laws and regulations | |
|---|---|---|---|---|
| People whose sexual orientation is criminalized where they live | Want to hide their orientation from the government | Need to search profiles of appropriate genders from their location | ||
| Survivors of intimate partner violence | Want to hide their location from certain people | Need two-factor authentication, to prevent account takeover | Hope we go beyond the minimum regulations | |
| Advertisers | Have some data to share, some to keep private | Want users' contact information, but shouldn't get it | Want to reach more people for less money | Must follow advertising regulations |
| Governments and regulators | Want their identities hidden from users | Need access to what they need, and nothing more | Need to collect accurate taxes | Must enforce the law |
Step 4: conflicts.
- Advertisers' low costs push costs onto users or the company: a financial model chooses between them.
- Many users want their identities or locations private from the government, which may want them to enforce laws. Mitigations: share nothing without a subpoena, or only specific information (never location or sexual orientation).
- Fairness and accessibility can risk Finder's reputation, depending on whose opinions count; and fairness can require collecting personal data, compromising privacy.
- Criminalized groups simply want to exist legally and peacefully, while governments target them.
Throughout the Process: Bias and Unfairness
Bias and unfairness are built into society: countering them is continuous, and doesn't end with a matrix. They enter technology through:
- unrepresentative data ("the general population," leaving out women);
- biased collection (a survey where no choice fits some people);
- builders misunderstanding the data's context, or making mistakes (correlation taken for causation);
- use for another purpose (a platform for vital communication in vulnerable communities, used to harass them).
They are created or amplified in use, as technology:
- reflects compounding historical injustices (the numbers computers use for words carry the biases of their source text);
- magnifies users' implicit biases and stereotypes;
- helps one population, but harms or excludes another;
- feeds back on itself (social media boosts posts that already have attention).
Privacy
- Privacy: "the ability to determine for ourselves when, how, and to what extent information about us is communicated to others" (Westin, 1967).
- Seen only as the individual against society or the state (the supposed privacy vs. national security "tradeoff"), privacy often loses. So we judge it in its social context, with five questions (the left column below).
Case study: Gaggle, a classroom platform with content-filtered blogs and email. A student messages another at 8pm on a Saturday, and Gaggle flags it as a possible sign of depression:
| Question | Answer |
|---|---|
| What type of information is shared? | Student data (the message, sender, recipient, timestamp, location) |
| Who is the subject of the information? | The student and their mental health concerns |
| Who is the sender of the information? | The student |
| Who are the potential recipients of the information? | The other student and Gaggle; and if Gaggle alerts them, parents, school administration, or law enforcement |
| What principles govern the collection and transmission of this information? | Gaggle's policy: no student or staff data goes to third parties without consent from a parent/guardian or the school, except in "Possible Student Situations," which may be reported to law enforcement |
Recipients include unintended ones (as in many homework assignments): someone looking over an administrator's shoulder; students and teachers who notice law enforcement being contacted.
Recap (15 mins)
A city lets "Zoomr" (a fictitious company) put electric scooters for rent on its sidewalks. Riders unlock a scooter with Zoomr's app (which charges by the minute, and records each trip's route), and leave it anywhere when they are done. Read through this analysis: the matrix's rows are the stakeholders (step 1), and its columns are the values (step 2).
| Safety | Accessibility | Privacy | Financial cost | |
|---|---|---|---|---|
| Riders | Want working brakes and lights, and safe places to ride | Need an app that works with screen readers, in many languages | Don't want their trips (from home, to work, to a clinic) tracked or sold | Want short trips to be cheap |
| Pedestrians who use wheelchairs or are blind | Don't want to trip over, or be hit by, scooters | Need sidewalks and curb ramps kept clear | ||
| People without a smartphone or bank card | Can't rent a scooter at all, without another way to unlock and pay | |||
| "Chargers" (gig workers paid to charge scooters overnight) | Work alone at night, lifting heavy scooters | The app tracks their location while they work | Want fair pay for each scooter they charge | |
| Zoomr | Wants few injuries, to avoid lawsuits | Trip data shows where to put scooters, and can be sold | Earns more when scooters are everywhere, and riders can leave them anywhere | |
| The city | Responsible for safe streets and sidewalks | Must follow disability law | Wants trip data, to plan bike lanes | Wants fees from Zoomr, and not to pay for cleanup |
Conflicts (step 4):
- Zoomr earns more when riders can leave scooters anywhere, but scooters left on sidewalks and curb ramps block pedestrians who use wheelchairs or are blind. Mitigation: a trip can only end in a marked parking zone, so the app keeps charging until the rider parks in one.
- The city wants trip data, but where a trip starts and ends can reveal where a rider lives, works, or gets medical care. Mitigation: Zoomr gives the city only the number of trips on each street each day, never individual trips.
Bias: Zoomr puts the most scooters where the most trips started last month, so a neighborhood that starts with few scooters has few trips, and keeps getting few scooters: a feedback loop.
- Which stakeholders in the matrix never ride a scooter? Which cells show why they belong in it anyway?
- Mitigation 1 resolves conflict 1. Which cells of the matrix show who pays for it?
- For the trip data in conflict 2: who is the subject, who is the sender, and who are the potential recipients (before and after the mitigation)?
Class Exercise (40 mins)
Case Study: Algorithmic hiring
Consider this scenario: “Shamazon” (a fictitious company) is looking to hire software engineers, and you have been tasked with designing a tool to filter the submitted resumes and select the ideal candidates for hire.
- Who are the stakeholders?
- What are the values?
- What is in the stakeholder-value matrix?
- What are the conflicts in the matrix? And how can we mitigate them?
- Where are bias and unfairness entering the product? And how can we mitigate them?
Wrap-up (5 mins)
- Programs are built by people, and affect people, so part of the design process should be understanding the ways that the programs either do or do not match the values of the people affected by them.