L1
·
Quiz
·
Lab
L2
·
Quiz
·
Lab
L3
·
Quiz
·
Lab
L4
·
Quiz
·
Lab
Module Test
Module 4 · Lesson 1

The Mirror That Distorts

How AI learns what "normal" looks like — and why that matters for everyone who isn't in the majority
If an AI was trained mostly on one group's voices, whose reality does it know best?

In the fall of 2018, reporters at Reuters broke a story that had been quietly buried inside Amazon for a year. The company had built an AI tool to sort job applicants — a résumé-screening system meant to identify the best engineers from thousands of applications. It was supposed to save time and find talent faster than any human recruiter could.

There was one problem. The AI had been trained on ten years of historical résumés — mostly submitted by men, because the tech industry had mostly hired men. The AI learned the pattern: successful applicants look like this. And "this" meant male. The system started automatically downgrading any résumé that included the word "women's" — as in women's chess club, women's engineering society. It penalized applicants who had attended all-women's colleges.

Amazon scrapped the tool. But the story raises a question that didn't go away with that one program: when AI learns from the past, does it lock in everything the past got wrong?

What the Amazon Story Actually Shows

The engineers who built that system didn't tell it to discriminate against women. Nobody typed in a rule that said "lower the score for female applicants." The bias emerged on its own — from the data. The AI looked at ten years of people who got hired, found patterns in what their résumés looked like, and built those patterns into its scoring. Since most past hires were men, maleness became a hidden signal for success.

This is what researchers call historical bias: when an AI trained on past data reproduces the prejudices that existed in that past, even when nobody intended it to. The data was real. The hiring records were accurate. But "accurate records of the past" and "fair guide to the future" are two completely different things.

Training data The collection of examples an AI learns from. Whatever patterns exist in this data, the AI tends to repeat — including unfair ones.
Historical bias When training data reflects past discrimination, the AI can learn to repeat that discrimination automatically, without anyone programming it to.

Think about what that means practically. If a hospital system trains an AI on old patient records from an era when Black patients were less often referred for expensive treatments, the AI might learn to recommend fewer treatments for Black patients — not because of malice, but because that was the historical pattern. If a bank trains a loan-approval AI on decades of lending decisions from a time when women were often denied credit, the AI might recreate that denial pattern.

The source of the unfairness isn't always hatred. Sometimes it's arithmetic.

Who Is the "Average Person" in the Dataset?

Here's a second angle on the same problem. In 2015, Google Photos launched a feature that automatically labeled photos with captions — "beach," "dog," "birthday party." It was impressive technology. It was also embarrassing: the system labeled photos of Black people as "gorillas." Google apologized and removed the label. But years later, journalists checking the fix found that Google had simply removed the label "gorilla" — and also blocked "chimp," "chimpanzee," and "monkey" — from all photo results. They hadn't fixed the underlying problem. They had censored it.

Why did the system make that error in the first place? Almost certainly because the training data — the millions of photos it learned from — overrepresented lighter-skinned faces. The AI had simply seen far more of them, so it was better calibrated for them. Darker skin tones were treated as edge cases, exceptions, things to be approximated from what it knew.

This is called representation bias: when the training data doesn't include enough examples from certain groups, the AI performs worse for those groups. It's not a bug in the code. It's a gap in the mirror.

Representation bias When some groups of people appear far less in training data than others, causing the AI to work poorly — or unfairly — for the underrepresented groups.

Joy Buolamwini, a researcher at MIT, documented this problem rigorously in 2018. She tested facial recognition systems from major companies — Microsoft, IBM, Face++ — and found that error rates were dramatically different depending on the subject's gender and skin tone. For lighter-skinned men, the systems were 99% accurate. For darker-skinned women, error rates climbed as high as 35%. That means more than one in three darker-skinned women were misidentified. Her paper, co-authored with Timnit Gebru, was called "Gender Shades." It changed the industry's conversation — and it started because Buolamwini noticed the systems worked worse on her own face.

Ethical Question — No Clean Answer

Amazon's AI learned from real historical data. The past discrimination was real. Does that mean the AI was "wrong" — or just honest about what the world had been? If we fix the AI to treat women equally, are we correcting a bias or overriding accurate historical data? Who gets to decide which past patterns deserve to be continued?

Why This Matters Beyond Tech Companies

You might think this is a problem for engineers to solve, not for you. But AI systems that show these biases aren't just sitting in corporate offices. They're being used in courts to predict whether someone will commit another crime. They're used in schools to detect "suspicious behavior" in hallways. They're used in hospitals to prioritize who gets treatment. In 2019, a study in the journal Science found that a widely used health algorithm — used for approximately 200 million people in the U.S. — was systematically underestimating the health needs of Black patients compared to white patients with the same actual health problems, because it used historical health care costs as a proxy for health need, and Black patients had historically been charged less.

This is the part that should genuinely disturb you: these systems often look objective. Numbers look neutral. An algorithm sounds scientific. But a biased input plus a calculation still produces a biased output. The math doesn't launder the unfairness. It just hides it.

You Now See What Most People Miss

When someone says "the AI decided" or "the algorithm flagged it," most people treat that as the end of the conversation — as if a machine can't be wrong or unfair. You now know that the decision is only as fair as the data that trained it. That changes how you should read every story about AI making consequential decisions about people's lives.

The word "bias" here doesn't mean the AI has feelings or prejudices. It means the AI has patterns baked in from data — and those patterns sometimes hurt real people from specific groups, consistently, without anyone noticing because the system "looks objective."

The communities most affected by biased AI systems tend to be the same communities that were already disadvantaged before AI existed. The technology doesn't create that inequality from scratch. But it can automate it, scale it up, and make it invisible — which is arguably worse.

The Scale Problem

One more thing worth understanding: the scale at which AI operates makes bias more dangerous than individual human bias. A single biased hiring manager affects maybe a few hundred decisions a year. An AI system embedded in recruiting software used by thousands of companies affects millions of applications simultaneously — all in the same biased direction. A biased judge affects dozens of cases. A biased recidivism-prediction algorithm (one that predicts whether someone will commit another crime) affects every courtroom in the state that uses it.

This is what makes the representation problem so urgent. When something is wrong in the data, it's wrong at industrial scale, applied to millions of people, automatically, around the clock. Human biases are inconsistent — people have good days and bad days, they're influenced by context, they sometimes catch themselves. An AI's biases are completely consistent. It makes the same error in the exact same way every single time, millions of times, until someone discovers the problem and fixes it.

That discovery usually happens because someone notices it happening to them — or to their community.

Lesson 1 Quiz

The Mirror That Distorts · 5 questions
1. Amazon's résumé-screening AI downgraded applications mentioning women's organizations. What caused this?
Correct. Nobody programmed the bias in deliberately. The AI found patterns in who had historically been hired and encoded those patterns — including the fact that past hires were mostly male.
Not quite. The bias wasn't intentionally coded. It emerged automatically from patterns in historical training data that overrepresented male hires.
2. Joy Buolamwini's "Gender Shades" research found that facial recognition error rates were as high as 35% for which group?
Correct. The combination of darker skin tone and female gender produced the highest error rates — up to 35% — compared to near-perfect accuracy for lighter-skinned men. This demonstrated representation bias in action.
Review the lesson. Buolamwini found the highest error rates specifically for darker-skinned women — the group least represented in the training data.
3. A new city is considering using a crime-prediction AI trained on 20 years of arrest data from a neighborhood that has historically been over-policed. What is the most likely risk?
Exactly right. This is historical bias at its most dangerous: more police in an area produces more arrests, which produces data that looks like more crime, which trains an AI to send more police — a self-reinforcing loop.
Think about what historical bias means in practice. The training data records arrests, not actual crime. Areas that were policed more heavily had more arrests — so the AI learns to treat those areas as high-risk.
4. Why is AI bias considered more dangerous than individual human bias in high-stakes decisions?
Correct. Scale and consistency are the key factors. A biased algorithm runs the same error millions of times across many institutions simultaneously, whereas individual human bias varies and affects fewer cases.
The lesson makes a specific point about scale and consistency. An AI doesn't have good days or bad days — it applies the same pattern every single time, across millions of decisions.
5. What did Google do after its photo-labeling AI mislabeled Black people's photos, according to later reporting?
Right. This distinction matters: removing offensive labels is a cosmetic fix, not a technical one. The underlying representation gap in the training data that caused the error wasn't addressed.
According to later journalism, Google censored the label rather than fixing the underlying problem. The training data gap that caused the error in the first place remained.

Lab 1: The Bias Auditor

Investigate a hypothetical AI system for hidden bias · 3+ exchanges to complete

Your Role: Independent Auditor

A school district wants to use an AI system to predict which students are "at risk" of dropping out, based on historical records from the past 15 years. They've asked you to audit it before deployment. Your lab partner — an AI investigator — will help you dig into the data and decisions, but won't hand you conclusions. You have to build the argument yourself.

Consider: What questions would you ask about the training data? What communities might be misrepresented? What harms could a biased prediction cause for a student who gets incorrectly flagged?

Start here: "What's the first thing I should check about this dropout-prediction AI before the district deploys it?" — or challenge any assumption in the setup.
Investigator AI
Lab 1
You've been hired to audit a dropout-prediction AI before a school district deploys it. I'm your research partner — I'll push back on weak reasoning, but I won't hand you conclusions. What's your first question about this system?
Module 4 · Lesson 2

The Amplifier

How recommendation algorithms build communities — and what happens when those communities go somewhere dark
If an algorithm's only job is to keep you watching, who is responsible for where it leads you?

In 2019, a former YouTube engineer named Guillaume Chaslot went public with something he had seen from the inside. Chaslot had worked on YouTube's recommendation algorithm — the system that decides which video plays next. He told The Guardian and other outlets that the algorithm had one overriding goal: maximize watch time. Keep people on the platform as long as possible. Every recommendation was a bet: which video will this person click on and then keep watching?

What Chaslot and later researchers found was a consistent pattern: the algorithm kept recommending more extreme content. A person who watched a mainstream political video would be recommended a more partisan one. A person who watched a moderate fitness video would gradually be led toward videos about extreme diets. A teenager watching flat-earth curiosity content would be recommended conspiracy content about vaccines, then government cover-ups, then darker material still. Researchers called it "radicalization by recommendation."

The algorithm wasn't trying to radicalize anyone. It was optimizing for clicks. But outrage, fear, and extreme claims turned out to produce more engagement than calm, measured information. The machine learned: escalation keeps people watching. So it escalated.

How Recommendation Algorithms Actually Work

A recommendation algorithm is a system that looks at what you've watched, what people like you have watched, and what tends to keep people watching, and then picks the next thing to show you. It sounds helpful — and sometimes it is. But the metric it optimizes for isn't "what's true" or "what's good for you" or even "what you'd actually enjoy if you thought about it." The metric is usually engagement: clicks, time spent, shares, replies.

Recommendation algorithm Software that chooses what content to show you based on patterns in your behavior and others' behavior — optimized for engagement, not accuracy or wellbeing.
Engagement optimization Designing a system to maximize clicks, watch time, or interactions — which often rewards provocative or outrageous content over accurate or calm content.

The problem is that extreme content is often more engaging than moderate content. A headline that says "scientists find mild correlation" gets fewer clicks than "study proves your phone is killing you." A video that calmly explains a political issue gets fewer watch-minutes than a video that declares the other side is evil. The algorithm doesn't know or care which is more accurate — it just knows which one you kept watching.

Researchers from the University of California, Berkeley, and other institutions documented what they called a "rabbit hole effect" in YouTube data around 2018–2019: when someone started watching videos on politically charged topics, the recommendation system consistently pushed toward more extreme content — regardless of whether the user started on the left or the right of the political spectrum. The escalation pattern was symmetric.

The Community That Algorithms Build

Here's what this creates over time: online communities organized around escalating outrage. Not communities that formed because people sought each other out and decided they had something in common. Communities that were assembled by an algorithm choosing, thousands of times a day, to show each person slightly more intense versions of what already made them click.

This affects who "gets heard" in an important way. If you're part of a group that expresses things calmly and moderately, the algorithm has less incentive to amplify your voice. If you're part of a group that expresses things with anger, fear, or extreme claims, you get more reach. The algorithm doesn't pick voices based on accuracy, wisdom, or even popularity — it picks them based on who generates more engagement. This means the most extreme voices within any community often end up being the most visible ones online, even if they're a small minority.

In 2021, internal research at Facebook (later revealed by whistleblower Frances Haugen) showed that the company's own researchers had found that its engagement algorithm was amplifying divisive content and that users who followed political content were being pushed toward "more and more extreme content." One internal slide, reported by the Wall Street Journal, summarized the problem starkly: "Our algorithms exploit the human brain's attraction to divisiveness."

Ethical Question — No Clean Answer

YouTube and Facebook's algorithms weren't designed to spread misinformation or radicalize users. They were designed to keep people watching — which is what their businesses depend on. Does a company have a responsibility to limit its reach when pursuing its business goal causes social harm? Who should decide what "harm" means in this context — the company, the government, or users themselves?

Who Gets Heard — and Who Doesn't

The flip side of amplification is suppression. When algorithms preferentially surface some voices, others get buried. Researchers studying social media platforms have found multiple patterns here. Civil rights organizations documented in 2020 that Instagram's algorithm appeared to systematically deprioritize content tagged with Black Lives Matter hashtags at certain moments. LGBTQ+ creators on TikTok documented in 2019 and 2020 that content with their identities flagged or hidden, in what TikTok eventually acknowledged was an overly broad "safety" filtering system.

This creates a layered problem: the voices the algorithm amplifies are often the loudest and most extreme; and the voices it suppresses are often from communities that are already marginalized. The result is that the "public conversation" you see online isn't a neutral reflection of what people think — it's a distorted picture shaped by what an engagement-maximizing system decided to show you.

You Now See What Most People Miss

Most people experience their social media feed as "what's happening" or "what people think." You now know it's an edited version — edited not by a journalist with professional standards, but by an algorithm that rewards provocation over accuracy. The communities you see most clearly online are partly the communities the algorithm wanted you to see.

Understanding this doesn't mean everything you see online is false. It means the selection process is invisible and driven by incentives that have nothing to do with your actual interests or with truth. Knowing the algorithm exists — and what it optimizes for — gives you one tool for questioning why you're seeing what you're seeing.

At a policy level, this question is being debated seriously: the EU's Digital Services Act (passed 2022, enforced 2023) requires large platforms to assess and report on "systemic risks" from their recommendation systems — one of the first laws anywhere to treat recommendation algorithms as a matter of public concern rather than pure private business. Whether that law actually changes things remains to be seen.

Lesson 2 Quiz

The Amplifier · 5 questions
1. Former YouTube engineer Guillaume Chaslot said the recommendation algorithm's primary goal was to:
Correct. The algorithm's singular objective was engagement — keeping people on the platform as long as possible. Everything else, including what content it recommended, was shaped by that goal.
Chaslot was explicit: the algorithm optimized for watch time, not for quality, accuracy, or diversity of viewpoint.
2. Why did extreme and outrageous content tend to be recommended more by engagement-optimizing algorithms?
Exactly. The algorithm didn't "prefer" extreme content ideologically — it learned that extreme content produced more engagement, so it recommended it more. The bias came from human psychology, amplified by the algorithm.
The algorithm has no preferences. It follows patterns in user behavior. Extreme content tends to generate more engagement — clicks, watch time, shares — so the algorithm learned to recommend it more frequently.
3. A 15-year-old starts watching videos about a minor conspiracy theory on YouTube. Based on what the lesson describes, what is the most likely outcome over the next hour of recommendations?
Correct. This is the "rabbit hole effect" documented by researchers. The recommendation system consistently pushes toward more extreme content because escalation produces more engagement — regardless of the user's starting point.
Research documented a consistent escalation pattern: the algorithm pushes toward more extreme content over time because that keeps people watching. It doesn't self-correct toward accuracy.
4. What did Facebook's internal research (revealed by Frances Haugen) find about its engagement algorithm?
Right. Facebook's own researchers documented that the algorithm was amplifying divisiveness and pushing users toward extremes — a finding the company apparently knew about before it became public.
The internal research, as reported by the Wall Street Journal, found the algorithm was exploiting human attraction to divisive content and pushing political users toward more extreme material.
5. LGBTQ+ creators on TikTok documented in 2019–2020 that their content was being filtered or hidden. What does this illustrate about recommendation algorithms?
Exactly. The lesson makes this "layered problem" explicit: engagement algorithms amplify the loudest and most extreme voices while suppressing communities that are already marginalized. The public conversation gets distorted in both directions at once.
The lesson describes this as a "layered problem": amplification of extreme voices plus suppression of marginalized voices creates a doubly distorted picture of who gets heard online.

Lab 2: The Algorithm Designer

Redesign a recommendation system with different goals · 3+ exchanges to complete

Your Role: Platform Designer

You've been asked to redesign a social media recommendation algorithm. The current one optimizes purely for watch time — and it has the problems described in the lesson. Your task: propose a different optimization goal or a set of constraints. Your lab partner will challenge your design and ask you to defend it.

There's no perfect answer here. Every design choice involves trade-offs — between engagement and safety, between freedom of speech and harm reduction, between what users say they want and what actually benefits them.

Start here: "What should a recommendation algorithm optimize for instead of pure watch time?" — or push back on whether algorithms can ever be truly neutral.
Design Critic AI
Lab 2
You're redesigning a recommendation algorithm. I'll challenge every design choice you make — not to be difficult, but because every trade-off matters. What goal should this system optimize for, and why is that better than watch time?
Module 4 · Lesson 3

Language Has an Accent

Why AI language models know some people's worlds far better than others — and what that silence costs
If an AI can write fluently about one culture but stumbles on another, what does that say about whose knowledge was collected?

In 2021, a paper published by researchers at the University of Washington and other institutions analyzed which languages were well-represented in the training data of large AI language models. They found that the models had been trained on data drawn overwhelmingly from the internet — and the internet was not a fair sample of humanity.

English made up an estimated 46% of content on the internet at that time, despite being the native language of roughly 5% of the world's population. Languages like Yoruba (spoken by 45 million people in West Africa), Igbo, Swahili, and dozens of others had almost no representation. When researchers tested GPT-3's ability to understand and generate these languages, performance was dramatically worse — sometimes nonsensical.

A separate study published in 2023 by researchers at Masakhane, an African AI research organization, found that AI translation tools produced translations into African languages that were often grammatically wrong, culturally inappropriate, or simply confused. For the nearly one billion people who speak languages native to Africa, the most powerful AI language tools in the world worked poorly or not at all. Not because of any deliberate exclusion. Because the data that shaped these systems was gathered from where data was most easily gathered.

The Internet Is Not the World

Large language models — the kind of AI that powers chatbots and writing tools — learn by processing enormous amounts of text. The more text they see in a language, from a culture, about a topic, the better they understand it. The less text they see, the worse they perform.

The problem is where that text comes from. Most of it comes from the internet. And the internet is not a neutral sample of human experience. Access to the internet is unequal. Who creates content on the internet is unequal. English-speaking, Western, college-educated voices have historically produced a vastly disproportionate share of the text that ends up in training datasets.

Large language model (LLM) An AI system trained on massive amounts of text to understand and generate language. Its fluency and knowledge depend entirely on what text it was trained on.
Data desert A community, language, or knowledge domain that has produced little digital text, leaving AI systems with little to learn from — and performing poorly for that group.

This creates a compounding inequality. Communities with less internet access produce less training data. Less training data means the AI works worse for those communities. An AI that works worse for those communities is less useful to them and less likely to be adopted. Less adoption means less feedback, fewer improvements — the gap grows wider over time.

A student in Lagos trying to use an AI writing assistant in Yoruba isn't getting the same tool as a student in London using it in English. They're not even getting something roughly equivalent. They might be getting something actively unreliable.

The Deeper Problem: Whose Knowledge Counts?

This isn't just about language. It's about what knowledge gets preserved and what gets lost.

Consider indigenous knowledge systems. Many communities around the world have developed sophisticated understandings of medicine, ecology, agriculture, and navigation over hundreds of generations — knowledge passed down orally, in community practices, and in languages that have no large written corpus. None of this is in the training data. An AI asked about a plant's medicinal properties might give you information from Western scientific journals and miss entirely what local healers in that plant's native region have known for centuries.

Or consider dialects and informal language. The research on AI and African American Vernacular English (AAVE) — a legitimate, grammatically complex dialect spoken by millions of Americans — has documented that AI writing tools often flag AAVE as incorrect English, that speech recognition systems perform worse on AAVE than on standard American English, and that AI tools trained mostly on formal written text can treat vernacular language as a mistake rather than a valid form of expression.

Ethical Question — No Clean Answer

Should AI companies be required to represent all human languages equally, even if collecting that data is expensive and some communities have explicitly said they don't want their language scraped for commercial AI training? What happens when the goal of "better AI for everyone" conflicts with a community's right to decide what happens to their cultural knowledge?

In 2023, the Māori people of New Zealand published a formal position statement on AI and their language, te reo Māori — expressing concern that AI companies might train models on te reo without the community's consent, producing systems that misrepresent their culture and strip the language from its cultural context. They were raising a question that applies globally: who has the right to decide whether a community's knowledge and language gets turned into AI training data?

What This Means For You

Even if you live in an English-speaking country, this isn't someone else's problem. First, because the world you live in includes billions of people whose knowledge and language AI is getting wrong. Second, because even within English, the same imbalances appear at smaller scale.

Studies have found that AI writing tools trained on predominantly formal, educated English text perform differently on text from different socioeconomic backgrounds, different regions, and different writing traditions. An AI trained mostly on Wikipedia and academic papers will handle those styles better than street-level language, personal narrative, or culturally specific references from communities underrepresented in the training corpus.

You Now See What Most People Miss

When an AI seems "smart" or "accurate" on a topic, that intelligence is partly just a reflection of how much data existed about that topic. An AI that sounds confident about one culture's history and vague about another's isn't more knowledgeable — it's just better supplied with data. That difference has nothing to do with which culture's knowledge matters more.

This is important for how you use AI tools in your own life. If you're asking an AI to help you understand a topic that has been heavily documented in English — a mainstream scientific question, a major historical event covered in Western sources — you're likely to get something reliable. If you're asking about a community's history that has been predominantly oral, a non-Western cultural practice, or a perspective from the global South, you should treat the AI's answer with much more skepticism, and look for human sources from within that community.

The tool is not equally calibrated for all questions. Knowing which questions it handles well — and which it probably handles poorly — is a skill that makes you a much more careful thinker.

Lesson 3 Quiz

Language Has an Accent · 5 questions
1. Why do AI language models perform much better in English than in languages like Yoruba or Igbo?
Correct. English dominates internet content at roughly 46% despite being the native language of about 5% of the world's population. AI trained on internet data inherits this imbalance.
The lesson doesn't argue AI companies deliberately chose to prioritize English, or that English is simpler. The imbalance comes from where training data was collected: the internet, which is heavily English-dominant.
2. What did researchers from the Masakhane organization find when testing AI translation tools for African languages?
Correct. The Masakhane research documented that AI translation into African languages was often unreliable — not because of malice, but because of data scarcity.
The lesson is specific: Masakhane researchers found that translations were grammatically wrong, culturally inappropriate, or simply confused — a direct consequence of those languages being underrepresented in training data.
3. A student in New Zealand asks an AI to explain the history of the Māori people using only AI knowledge. Based on the lesson, what should the student be most aware of?
Exactly. The lesson uses the Māori example specifically. Knowledge systems that have been oral or have limited digital text representation will produce less reliable AI responses — and communities have the right to be concerned about how their culture is represented.
The lesson specifically discusses the Māori example and the concern about AI misrepresenting cultures whose knowledge has not been digitized in large amounts. This is precisely a case where AI skepticism is warranted.
4. Research on AI and African American Vernacular English (AAVE) found that AI tools often:
Correct. AI trained predominantly on formal written English treats legitimate dialect variation as error — another example of the same underlying problem: training data that doesn't reflect the full range of human language use.
The lesson documents that AAVE — a legitimate, grammatically complex dialect — is often treated as incorrect by AI tools trained on formal written text, and that speech recognition performs worse on it.
5. The lesson says that when an AI sounds "confident" about one culture's history and vague about another's, this indicates:
Exactly right. AI confidence is a reflection of data density, not truth or importance. This is the core insight: how much the AI knows is determined by what data it trained on, not by the inherent value of different knowledge traditions.
The lesson is explicit: AI confidence reflects how much training data existed on a topic — not whose knowledge is more valid. Data availability does not determine cultural importance.

Lab 3: The Knowledge Cartographer

Map what the AI knows — and what it doesn't · 3+ exchanges to complete

Your Role: Knowledge Investigator

Your task is to think through what an AI language model probably knows well versus poorly — not by testing it directly, but by reasoning about what's in its training data. Your lab partner will challenge your reasoning and push you to be specific about which communities and knowledge systems are likely underrepresented — and what real consequences that has.

Consider: oral traditions, indigenous practices, minority languages, working-class experiences, historical voices who never had access to writing or publishing. Where are the gaps, and who pays for them?

Start here: "What knowledge domains do you think an AI trained on the internet probably handles worst — and why does that matter?" — or take a specific community and argue what the AI probably gets wrong about them.
Knowledge Critic AI
Lab 3
I'll challenge your reasoning, not just agree with you. Start by telling me: what's a community or knowledge system you think AI gets systematically wrong — and what's your evidence or reasoning?
Module 4 · Lesson 4

What We Do About It

From understanding bias to demanding accountability — at the individual, institutional, and policy level
If you can now see the problem, what are you actually supposed to do with that knowledge?

In January 2020, Robert Williams, a 42-year-old Black man living in a suburb of Detroit, was arrested in front of his wife and daughters. Police had used a facial recognition system to match him to surveillance footage from a shoplifting incident. The system said it was him. The match was wrong.

Williams spent 30 hours in police custody before investigators, reviewing the evidence, acknowledged the identification was incorrect and released him. He became the first documented case in the United States of a wrongful arrest based on facial recognition AI. The ACLU took his case. He sued the Detroit Police Department.

Williams later wrote: "It's bad enough that you're being accused of something you didn't do. It's even worse knowing the accusation came from a machine that had already been shown to be least accurate for people who look like me." He was referring to exactly the research Joy Buolamwini had published two years earlier — the research that showed facial recognition made the most errors on darker-skinned faces. The problem had been documented. The technology was still deployed. The arrest still happened.

The Gap Between Knowing and Changing

Williams's story makes the stakes concrete. This wasn't a theoretical problem or a research paper. A man was arrested by a machine known to be unreliable for people with his skin tone, and nobody in the system stopped it. Not because nobody knew about the bias — Joy Buolamwini's research was published in 2018 and covered widely. But knowing about a problem in AI isn't the same as fixing it or being protected from it.

This gap — between documented problems and actual change — is where the work of accountability happens. And accountability in AI happens at several levels at once: individual, organizational, and institutional/legal.

Algorithmic accountability The practice of requiring AI systems to be transparent, auditable, and correctable — and holding the humans who build and deploy them responsible for the harms they cause.

Williams's case eventually contributed to real policy change. In 2021, Detroit amended its policy on facial recognition, requiring that AI identifications be treated as investigative leads only — not as sufficient basis for arrest. Several U.S. cities, including San Francisco, Boston, and Portland, banned government use of facial recognition entirely. These changes happened because Williams was willing to speak publicly, because the ACLU litigated, and because journalists and researchers kept the story alive. None of it happened automatically.

What Individuals Can Actually Do

There's a familiar frustration with "what you can do" sections: they usually end with "recycle more" or "talk to your friends" — advice that sounds like action but doesn't change systems. This isn't that. The actions that have actually moved AI accountability forward involve specific, trackable behaviors.

Name what you see. When you notice AI producing outputs that seem biased — a search result that stereotypes a group, a recommendation feed that never shows certain voices, a writing tool that flags vernacular as incorrect — it's worth naming it specifically, not just noticing it privately. Researchers at Google, academic institutions, and advocacy organizations have all been alerted to specific AI failures by regular users who described what they saw in enough detail to investigate.

Ask for transparency. When an institution tells you "the algorithm decided" — in school, in a hiring context, in a police interaction — you are entitled to ask what algorithm, trained on what data, with what documented error rates. This is increasingly a legal right in the European Union under the AI Act (2024) and the GDPR, and in several U.S. states. Asking the question — even if you don't get an answer — creates a record that the question was raised.

Support the researchers. Organizations like the Algorithmic Justice League (founded by Joy Buolamwini), the Distributed AI Research Institute (DAIR, founded by Timnit Gebru), and Masakhane do the work of documenting AI bias and advocating for affected communities. Following their work, sharing it, and — eventually — contributing to it or supporting it politically are concrete actions.

You Now See What Most People Miss

Most people encounter AI bias as something abstract — a problem "out there" in tech companies. You now see that it shows up in specific decisions about specific people: who gets hired, who gets arrested, whose health needs get assessed, whose voice gets heard. That specificity is what makes accountability possible. You can't hold a system accountable for being "generally biased" — you can hold it accountable for a specific identifiable error that harmed a specific person.

Institutional and Policy Level: What's Actually Being Done

This is where you see what accountability looks like at scale — and where the stakes are explicitly policy-level decisions being made right now.

In March 2024, the European Union's AI Act became law — the world's first comprehensive regulatory framework for AI. It categorizes AI systems by risk level. High-risk uses — including AI in criminal justice, employment, education, and critical infrastructure — face strict transparency and auditing requirements. Real-time biometric surveillance (like facial recognition in public) is largely prohibited. Companies deploying high-risk AI must document their training data, demonstrate their systems don't discriminate, and register with a public database.

In the United States, the approach has been more fragmented. The Biden administration issued an executive order on AI in October 2023 covering safety, security, and equity concerns. The FTC has brought enforcement actions against companies for deceptive AI claims. Several states — California, Illinois, Colorado — have passed laws requiring impact assessments for high-risk AI in employment and housing. But there is no single federal AI law equivalent to the EU's.

This matters because the regulatory landscape shapes what companies are required to disclose, what harms are legally redressable, and what communities have standing to demand accountability. A community harmed by a biased AI in a country with no AI regulations has significantly fewer options than the same community in a jurisdiction with transparency requirements.

Ethical Question — No Clean Answer

The EU AI Act bans most real-time facial recognition in public. Proponents say this protects people from documented wrongful arrest and surveillance. Critics say it makes it harder to find missing children, track terrorists, and solve serious crimes. How should democratic societies weigh an individual's right not to be wrongly identified by a biased system against the public interest in solving crimes? And who should make that trade-off — courts, legislatures, tech companies, or citizens?

The communities doing the most concrete work on this — Masakhane for African languages, the Algorithmic Justice League for facial recognition, Data for Black Lives for health equity — operate at the intersection of research and advocacy. They publish findings, pressure companies, and testify to legislatures. Some of the most important AI accountability work of the last decade has been done by researchers who were themselves affected by the systems they studied.

That matters for you specifically: the person most likely to notice a gap in AI performance is the person for whom the gap exists. Expertise in this field doesn't require a computer science degree. It requires the ability to see the problem, describe it precisely, and connect it to the larger pattern. That's something you can do right now, with what you've learned in this module.

Lesson 4 Quiz

What We Do About It · 5 questions
1. Robert Williams was wrongfully arrested in Detroit in 2020. What makes his case significant beyond the individual injustice?
Correct. Williams's case was the first documented instance of its kind in the U.S., and his willingness to speak publicly — combined with ACLU litigation — drove real policy changes in Detroit and contributed to municipal facial recognition bans elsewhere.
The lesson emphasizes two things: it was the first documented case of its kind, and it led to concrete policy changes. That's what makes it significant beyond the individual harm.
2. The EU AI Act, which became law in 2024, classifies AI use in criminal justice, employment, and education as:
Correct. High-risk AI uses face the most demanding requirements under the Act — documentation, auditing, public registration, and demonstration that the systems don't discriminate.
The EU AI Act uses a risk-based framework. Uses in criminal justice, employment, and education are classified as high-risk, requiring transparency, auditing, and demonstration of non-discrimination.
3. When an institution tells you "the algorithm made the decision," what does the lesson say you are entitled to ask?
Exactly. The lesson identifies these three specific questions as the ones that matter — and notes that in the EU and several U.S. states, asking them is increasingly a legal right, not just a reasonable request.
The lesson is specific: you can ask about the algorithm's identity, its training data, and its documented error rates. In the EU and several U.S. states, this is becoming a legal right.
4. A city is considering deploying facial recognition at a public transit hub. A community group wants to challenge this. Which approach, based on the lesson, has actually produced policy change?
Correct. The lesson traces exactly this pattern: Williams's documented case, ACLU litigation, journalist coverage, and sustained public pressure produced specific, measurable policy outcomes. That process is replicable.
The lesson traces the specific pathway that produced actual change — documented cases, legal action, public pressure, policy response. That's the model that worked in Detroit, San Francisco, Boston, and Portland.
5. The lesson argues that expertise in AI accountability doesn't require a computer science degree because:
Exactly. Joy Buolamwini noticed the problem when she noticed it happening to her own face. The lesson closes by making explicit that this kind of situated observation — combined with the ability to describe and connect the pattern — is itself a form of expertise that doesn't require a computer science background.
The lesson's closing argument is precise: the person the gap affects is most likely to see it, and noticing it, describing it clearly, and connecting it to the larger pattern is itself expertise. That's reachable without a technical degree.

Lab 4: The Accountability Advocate

Build an accountability argument for a real AI harm · 3+ exchanges to complete

Your Role: Advocate

A city council is deciding whether to allow the local police department to use a facial recognition AI system for identifying suspects from CCTV footage. You've been asked to present the accountability case — either for strict regulation with mandatory transparency, or for an outright ban. Your lab partner will play a skeptical council member who needs specific, concrete arguments, not general claims about bias.

You'll need to draw on what you've learned across this whole module: training data bias, representation gaps, the Williams case, what actual policy changes have looked like, and what rights people have under existing law. Be specific. Vague arguments don't win council votes.

Start here: "I'm arguing for [regulation/ban] of facial recognition AI in our police department, and here's my strongest specific argument..." — or challenge the premise of the decision itself.
Council Member AI
Lab 4
I'm on the city council. I've heard general complaints about AI bias before — I need specific evidence, specific harms, specific policy models that have worked. You have five minutes. Make your case. What's your strongest argument?

Module 4 Test

Communities, Bias, and Who Gets Heard · 15 questions · Pass at 80%
1. Amazon's résumé-screening AI penalized women's organizations on résumés because:
Correct. Historical bias: the AI learned from who had been hired in the past — mostly men — and treated that pattern as a predictor for future success.
No deliberate rule was coded. The bias emerged from training data that overrepresented male hires, making maleness an implicit success signal.
2. "Representation bias" in AI means:
Correct. When groups are underrepresented in training data, the AI simply has less to learn from about them — producing worse performance, not deliberate prejudice.
Representation bias is about data gaps, not deliberate misrepresentation. Underrepresented groups in training data result in AI that works poorly for those groups.
3. Joy Buolamwini's "Gender Shades" research (2018) found that facial recognition systems were most accurate for:
Correct. Lighter-skinned men had near-perfect accuracy while darker-skinned women had error rates up to 35% — directly reflecting which groups were most represented in training data.
Buolamwini found the highest accuracy for lighter-skinned men (near 99%) and the lowest for darker-skinned women (error rates up to 35%).
4. A study published in the journal Science in 2019 found that a widely-used U.S. health algorithm underestimated Black patients' health needs because:
Correct. This is a precise example of historical bias: the algorithm used cost as a proxy for need, but cost reflected past discriminatory access to care, not actual health need.
The algorithm used health care cost as a proxy for health need. Since Black patients had historically received less care (and thus been charged less), the algorithm systematically underestimated their needs.
5. Why is consistent AI bias considered more dangerous than inconsistent human bias in high-stakes decisions?
Correct. Scale and consistency are the key factors. An AI doesn't have good or bad days — it repeats the same pattern millions of times, simultaneously, across every institution that deploys it.
Scale and consistency are the key points. A biased algorithm makes the same error every single time across millions of decisions — human bias is variable and affects far fewer cases.
6. Engagement optimization in social media recommendation algorithms tends to promote extreme content because:
Correct. The algorithm has no ideology — it follows engagement signals. Extreme and outrageous content generates more engagement, so the algorithm learns to recommend it more.
Nobody programs in a preference for extremism. The algorithm learns from engagement data — and extreme content tends to get more engagement. It's an emergent property of the optimization goal.
7. Frances Haugen's disclosure of Facebook's internal research revealed that company researchers had documented:
Correct. This is significant precisely because it shows the company had internal documentation of the problem before it became public — knowledge of the harm did not automatically produce change.
Haugen's disclosure revealed internal research showing the algorithm amplified divisive content and pushed users toward extremes — research the company possessed before it became public knowledge.
8. According to the lesson on language and AI, what percentage of internet content is estimated to be in English — despite English being the native language of about 5% of the world?
Correct. English makes up roughly 46% of internet content while being native to only about 5% of the world's population — creating a massive training data imbalance for AI language models.
The lesson cites approximately 46% — a massive overrepresentation relative to actual native speakers, which directly shapes what AI language models know well versus poorly.
9. The Māori community of New Zealand's 2023 position statement on AI and te reo Māori raised which key concern?
Correct. This case raises a deeper question: even if collecting language data could improve AI performance, communities have the right to decide whether and how their cultural knowledge is used for commercial AI training.
The statement raised the question of consent and cultural representation: training AI on a community's language without consent can produce systems that misrepresent that culture and strip language from its context.
10. A student uses an AI chatbot to research the history of a community whose knowledge has been primarily oral and has very little written digital documentation. The lesson suggests the student should:
Correct. This is the practical application of the lesson's insight: AI confidence reflects data density, not truth. For communities with limited digital text documentation, AI is less reliable and human sources within the community are more trustworthy.
The lesson is explicit: AI performs less reliably for communities underrepresented in training data. For topics with sparse or oral-tradition knowledge bases, human sources from within the community are more reliable.
11. Robert Williams's wrongful arrest in Detroit in 2020 demonstrated that:
Correct. Buolamwini's research had been published and covered widely before Williams's arrest. Knowing about bias didn't stop deployment. His case shows that accountability requires the full cycle: documentation, legal action, sustained public attention, and policy response.
The key point of Williams's case is that research documenting the bias already existed — and the system was still deployed. Accountability required more than published research: it required litigation, publicity, and political pressure.
12. The EU AI Act (2024) treats real-time biometric surveillance in public spaces as:
Correct. The EU AI Act largely prohibits real-time biometric surveillance in public — a direct policy response to documented harms from facial recognition systems like the one in the Williams case.
Real-time biometric surveillance in public is one of the EU AI Act's most strictly treated categories — largely prohibited, with only narrow exceptions defined in the law.
13. TIKTOK acknowledged in 2020 that LGBTQ+ content was being filtered or hidden due to:
Correct. TikTok acknowledged the filtering was due to an overly broad safety system — another example of algorithmic suppression happening without deliberate intent, but with real consequences for a marginalized community's voice.
TikTok attributed the filtering to an overly broad safety system — not deliberate discrimination. This illustrates that algorithmic suppression of marginalized voices can happen without anyone intending it.
14. Which organization, founded by Timnit Gebru after she left Google, focuses specifically on AI bias research and advocacy?
Correct. DAIR was founded by Timnit Gebru after her departure from Google. The Algorithmic Justice League was founded by Joy Buolamwini; Masakhane focuses on African language AI; Data for Black Lives works on health equity.
DAIR (Distributed AI Research Institute) was founded by Timnit Gebru. The Algorithmic Justice League was founded by Joy Buolamwini. Each organization focuses on a related but distinct area of AI accountability work.
15. A researcher from a community directly affected by a biased AI system is best positioned to notice the problem because:
Exactly. This is the module's closing argument: situated experience is a form of expertise. Joy Buolamwini noticed the problem when she saw it on her own face. That observation, precisely described, changed the field. The lesson applies to you directly.
The lesson closes by making this explicit: experiencing the gap, describing it precisely, and connecting it to a pattern is itself expertise. Buolamwini's work started when she noticed the system worked poorly on her own face — that situated observation drove major industry change.