Digital Divide, Computing Bias, and Crowdsourcing
Your school district sends homework to a website. Mia finishes hers in ten minutes on fast home Wi-Fi. Jayden has to sit in the library parking lot to get a signal. Same homework, very different night. Is that fair? And what happens when a computer program makes decisions using data that already reflects unfair patterns? In this page you'll learn how to explain, analyze, and judge the effects of computing on people.
Beneficial and harmful effects
Every computing innovation is built to solve a problem. But it can also cause effects nobody planned. Two pairs of words matter:
-
Intended effects are what the creators wanted to happen.
-
Unintended effects are side effects the creators did not plan, good or bad.
An effect can be beneficial or harmful, and a single innovation can have both at once. Think about a social media app. It connects you with friends (beneficial, intended). It can also keep you scrolling at 2 a.m. or spread rumors (harmful, often unintended).
A telehealth app lets patients in rural Montana video chat with doctors who are hundreds of miles away. a) Describe two beneficial effects. b) Describe two possible…
A telehealth app lets patients in rural Montana video chat with doctors who are hundreds of miles away.
- a) Describe two beneficial effects.
- b) Describe two possible harmful or unintended effects.
- c) Explain why a single innovation can have both beneficial and harmful effects at the same time.
The digital divide
Access can differ because of:
-
Income: home broadband and devices cost money.
-
Geography: rural areas often have fewer providers and slower service.
-
Age: some older adults have little experience with new tools.
-
Disability: sites that don't work with screen readers shut people out.
The divide matters because school, jobs, health care, and government services increasingly happen online. People on the "wrong side" fall further behind, which can make existing inequality grow.
Ways to reduce the divide: loan Wi-Fi hotspots or Chromebooks, open school and public library computer labs after hours, expand broadband to rural counties, offer low-cost internet plans, and teach digital skills.
Define the digital divide. Give one example of a group of people who might be on the "wrong side" of it, and name one way it affects them.
Define the digital divide. Give one example of a group of people who might be on the "wrong side" of it, and name one way it affects them.
A (hypothetical) survey of home broadband access in one state found: | Area | Households with home broadband | |---|---| | Suburbs | 93% | | Urban low-income…
A (hypothetical) survey of home broadband access in one state found:
| Area | Households with home broadband |
|---|---|
| Suburbs | 93% |
| Urban low-income neighborhoods | 71% |
| Rural counties | 62% |
- a) Which areas show the biggest gap compared with the suburbs, and how large is it?
- b) Describe two ways this gap could affect students' homework.
- c) Suggest two actions that a school district or city could take to reduce the divide.
Computing bias
Computers don't have opinions, but people do, and bias gets in through people. There are three main sources:
-
Data: a face-recognition system trained mostly on photos of light-skinned people makes more errors on darker skin.
-
Design: the creators choose which features to measure and what counts as "success."
-
Use: a program built for one purpose is used on a group it was never tested on.
Bias can hide behind a proxy: a feature that acts as a stand-in for something sensitive. A program may never see a person's race, but ZIP code often correlates with it. Removing the sensitive input does not remove the bias.
Ways to reduce bias: use training data that is representative, test results separately for different groups, remove or audit proxy features, keep a human in the loop for important decisions, and include diverse people on the design team.
Which of these is an example of algorithmic bias? Explain your choice. i) A calculator app that rounds results to two decimal places. ii) A face-recognition system…
Which of these is an example of algorithmic bias? Explain your choice.
- i) A calculator app that rounds results to two decimal places.
ii) A face-recognition system trained mostly on photos of light-skinned people that makes more errors on people with darker skin.
iii) A weather app that updates every hour.
A company builds a program to screen job applications. It is trained on resumes of employees hired over the past 20 years, and most of those employees were men. The…
A company builds a program to screen job applications. It is trained on resumes of employees hired over the past 20 years, and most of those employees were men. The program never receives an applicant's gender as an input.
- a) Explain how the program can still produce biased results.
- b) Identify the source of the bias (data, design, or use) and give one example of a feature that could act as a stand-in for gender.
- c) Describe two ways the company could reduce the bias.
Feedback loops in biased systems
Sometimes a program's output changes the data it will learn from next. This creates a feedback loop.
A city plans to use a predictive program to decide where to send extra police patrols. It is trained on past arrest records. The (hypothetical) data show: | Neighborhood…
A city plans to use a predictive program to decide where to send extra police patrols. It is trained on past arrest records. The (hypothetical) data show:
| Neighborhood | Patrol level in past years | Recorded arrests last year |
|---|---|---|
| A | Heavy | 400 |
| B | Light | 120 |
- a) Explain how the program could send even more patrols to Neighborhood A. Describe the feedback loop.
- b) Does the higher number of arrests in A prove that more crime happens there? Explain.
- c) Make two recommendations to the city to reduce the harm, and argue for or against using the program at all.
Crowdsourcing
Common forms:
-
Citizen science: volunteers help scientists, for example classifying galaxy photos on Zooniverse or counting birds.
-
Crowdfunding: many people each give a little money to a project, as on GoFundMe.
-
Collaborative knowledge: Wikipedia is written and edited by volunteers.
-
Reporting apps: residents report potholes or broken streetlights to the city.
Benefits: huge amounts of work done fast, many viewpoints, and low cost. Weaknesses: quality can vary, some people may be missing from the crowd, and the crowd can be wrong or manipulated.
A company's engineers privately writing code is not crowdsourcing, because the work is not open to a large group of outside contributors.
Fairer designs: allow reports by phone call, text message, or at the library; send city workers to inspect all areas regularly; and adjust priority by the number of reports per household or by actual inspections.
Define crowdsourcing. Which of these are examples of crowdsourcing? Explain each choice. i) Thousands of volunteers edit articles on an online encyclopedia. ii) A city…
Define crowdsourcing. Which of these are examples of crowdsourcing? Explain each choice.
- i) Thousands of volunteers edit articles on an online encyclopedia.
ii) A city app lets residents report potholes.
iii) A company's engineers privately write the code for a new app.
A city launches a smartphone app where residents report broken streetlights, and the city fixes streetlights in the order of the number of reports. The (hypothetical)…
A city launches a smartphone app where residents report broken streetlights, and the city fixes streetlights in the order of the number of reports. The (hypothetical) data after one month:
| Area | Households with a smartphone | Reports |
|---|---|---|
| North (higher income) | 85% | 140 |
| South (lower income) | 55% | 35 |
- a) Does the number of reports accurately reflect where streetlights are broken? Explain using the digital divide.
- b) Name one benefit of crowdsourcing in this case and one weakness.
- c) Redesign the reporting system so that it is fairer, and explain how each change helps.
Summary
Back to the start: Jayden in the library parking lot is on the wrong side of the digital divide. Lending hotspots and keeping school labs open are fair ways to give him the same chance as Mia.