In November 2023, a coalition of civil society organizations submitted formal comments to the European Parliament during the final negotiations of the EU AI Act. Among them was a group of disability rights advocates who argued that AI-powered hiring tools systematically screened out candidates who had non-linear career histories — a pattern disproportionately affecting people with chronic illness. Their testimony changed specific provisions of the final law. These were not engineers. They were people who had been affected, organized, and spoken.
A stakeholder is anyone with a legitimate interest in how a system behaves. In the context of AI, that is nearly everyone alive today — and certainly everyone reading this course. You interact with AI recommendation systems, AI-assisted medical diagnostics, AI-driven hiring screens, AI content moderation, and AI-curated news. Each of those systems was designed by people making choices, and each choice is contestable.
The question is not whether you have a stake. You do. The question is whether you exercise it — through the products you choose, the employers you hold accountable, the representatives you contact, the communities you participate in, and the work you do every day.
When Reddit announced pricing changes that would effectively kill third-party apps used by accessibility communities, hundreds of subreddits went dark in a coordinated protest in June 2023. Reddit reversed several policies. The episode demonstrated that users — as a collective stakeholder — can exert genuine pressure on technology platforms even without technical expertise or regulatory power.
Market behavior: Every time you choose or decline a product, you send a signal. Apple's 2021 App Tracking Transparency update — which required apps to ask permission before tracking — was partly driven by demonstrated consumer preference for privacy. Within a year, opt-in rates for tracking hovered around 25%, forcing the entire ad-tech industry to restructure.
Civic participation: AI governance is increasingly a legislative and regulatory matter. The EU AI Act, the US Executive Order on AI Safety (October 2023), and the UK AI Safety Institute were all shaped by public comment periods, advocacy organizations, and elected representatives responding to constituent pressure. These are not closed rooms.
Organizational voice: Where you work matters. Employees at Google, Microsoft, Amazon, and dozens of other technology companies have publicly objected to specific AI applications — sometimes successfully. Google's Project Maven (military drone targeting) was partially cancelled in 2018 after more than 4,000 employees signed a petition. That is stakeholder action from inside.
Passive acceptance is itself a choice. When users accept default privacy settings without reading them, when employees deploy AI tools without questioning their provenance, when citizens skip public comment periods on technology regulations — these are all choices that cede influence to whoever is paying the most attention.
Research on the diffusion of responsibility in social psychology applies here. When everyone assumes someone else is watching the AI systems that affect public life, the effective number of watchdogs approaches zero. Awareness of this dynamic is the first step to countering it.
You do not need to be an AI researcher to be an AI citizen. Literacy — understanding what AI can and cannot do, how it is governed, and where accountability lies — is itself a form of power. This course is part of building that power.
In this lab you'll have a structured conversation about the AI systems that already affect your life — at work, in your community, or in daily decisions — and explore which channels of influence are most accessible to you. The AI will help you think through your specific situation.
In February 2024, a British Columbia Civil Resolution Tribunal ruled against Air Canada after its AI chatbot told passenger Jake Moffatt that he could apply for a bereavement fare discount retroactively — advice that was false. Air Canada argued the chatbot was a "separate legal entity" responsible for its own statements. The tribunal rejected this, ruling that Air Canada was responsible for all information on its website, AI-generated or not. The passenger won. The company paid.
The case became a landmark on AI accountability — and a reminder that when AI gives you wrong information, someone is still responsible. Usually the organization that deployed it.
AI language models, including the most capable ones available today, hallucinate — they generate confident-sounding text that is factually incorrect. A 2023 study by researchers at Stanford found that AI legal research tools produced hallucinated citations in roughly one in four responses. In medicine, finance, and law, this is not a minor problem.
The practical rule is simple: for any AI output that will inform a real-world decision — a medical choice, a financial commitment, a legal action — verify the key claims with authoritative sources before acting. Treat AI outputs as a first draft, not a final answer.
In May 2023, New York attorney Steven Schwartz submitted a legal brief containing citations to six cases that did not exist — generated by ChatGPT. The court fined the lawyers involved $5,000. ChatGPT, when asked, had confirmed the fake cases were real. The episode triggered a wave of court rules requiring disclosure of AI use in legal filings.
Not all AI tools are designed primarily to help you. Recommendation algorithms at TikTok, YouTube, and Meta are optimized for engagement — which often means maximizing time-on-platform rather than user wellbeing. A 2021 internal Facebook study, later revealed by whistleblower Frances Haugen, found that the platform's own research showed Instagram was harmful to a significant minority of teen girls — and the company chose not to act on that research at the time.
Understanding what an AI system is optimized for changes how you should use it. If a recommendation engine is optimized for engagement, you should expect it to amplify content that provokes strong reactions, regardless of whether that content is accurate or good for you.
Automation bias is the tendency of humans to over-rely on automated systems, even when those systems are clearly wrong. Studies of aviation accidents have linked it to crashes where pilots failed to override autopilot errors. The same phenomenon appears in AI-assisted medical diagnosis: a 2019 study published in Nature Medicine found that radiologists shown AI predictions were significantly less likely to catch AI errors than radiologists working without AI assistance.
The antidote is not to avoid AI, but to maintain active engagement — forming your own judgment before consulting the AI output, rather than anchoring to it. This is sometimes called "human-in-the-loop" practice at the individual level.
Stop before acting on AI output. Investigate the source or logic of the claim. Find better, authoritative sources. Trace claims back to their origin. This framework, developed originally for media literacy by Mike Caulfield, applies directly to AI-generated content.
In this lab, you'll practice the skills of critical AI consumption. Ask the assistant about a topic you know something about — your field of work, a historical event, a technical subject. Then probe the output: ask for sources, check for hallucinations, and test the limits of its confidence.
In May 2022, Clearview AI settled with the American Civil Liberties Union, agreeing to stop selling its facial recognition database to most private companies in the US — the first major legal restriction on a commercial facial recognition firm. The settlement came after years of investigative journalism, advocacy by privacy organizations, and state-level legislative campaigns that built the political will for enforcement. No single action did it. The cumulative pressure did.
AI policy is not made only in Washington or Brussels. It is made at multiple levels simultaneously, and many of those levels are highly accessible to engaged citizens:
Federal regulation: The US National Institute of Standards and Technology (NIST) published its AI Risk Management Framework in January 2023 after an extended public comment period that received hundreds of submissions from individuals, civil society groups, and academic institutions. Those comments shaped the final document.
State and local legislation: Illinois' Biometric Information Privacy Act (BIPA), passed in 2008, became the most powerful biometric data protection law in the US — predating federal action by over a decade. It was driven by a coalition of labor unions, privacy advocates, and community organizations. By 2024, BIPA had generated over $1 billion in settlements against companies that collected facial and fingerprint data without consent.
International standards: The ISO/IEC 42001 AI management system standard, published in 2023, was developed through a process that included national standards bodies from dozens of countries, many of which accepted public input.
The European Commission's 2020–2021 consultation on AI regulation received over 1,200 responses from individuals, civil society organizations, companies, and governments. Provisions on prohibited AI applications — including social scoring systems and real-time facial recognition in public spaces — were significantly shaped by civil society input. Individual submissions were publicly logged.
Public comment periods: Every major US federal rule goes through a notice-and-comment period. Regulations.gov is the portal. Comments from individuals carry weight, especially when they are specific, technically informed, and represent perspectives the agency hasn't heard. Boilerplate form letters are counted but rarely influential; personal testimony with concrete experience is different.
Contacting elected representatives: Congressional offices track constituent contact on specific issues. A legislator who receives 50 calls from constituents about AI hiring discrimination is more likely to prioritize it than one who receives none. The same applies to state legislators, who often have more direct influence over tech regulation than federal representatives.
Joining organized coalitions: Organizations like the Electronic Frontier Foundation, AI Now Institute, Algorithmic Justice League, and Access Now do sustained policy work on AI issues. Membership, volunteering, or financial support amplifies individual capacity. The Algorithmic Justice League's work, founded by Joy Buolamwini at MIT, directly contributed to IBM, Amazon, and Microsoft pausing sales of facial recognition to law enforcement in 2020.
If your professional expertise intersects with AI — in healthcare, education, law, social work, or any field where AI is being deployed — your testimony carries particular weight in regulatory proceedings. Congressional hearings on AI regularly include clinicians, educators, and workers describing lived experience with algorithmic systems. These voices are actively sought and often underprovided.
Policy change is slow. The Clearview AI restriction took four years of sustained advocacy. BIPA took over a decade to produce major enforcement. The EU AI Act took five years from inception to passage. Individual actions matter most when they are part of sustained, organized efforts — but they do matter.
In this lab you'll work with the AI assistant to draft a substantive public comment on an AI policy issue that matters to you. The assistant will help you identify the right issue, frame your lived experience, and structure your comment for maximum impact. The goal is a draft you could actually submit.
In December 2020, Timnit Gebru — then co-lead of Google's Ethical AI team — was fired following a dispute over a research paper on the risks of large language models. The paper, which she co-authored, raised concerns about environmental cost, bias, and the homogenization of language generated by AI systems trained primarily on English-language internet text. Google disputed whether the paper met internal review standards; Gebru and supporters saw it as suppression of legitimate safety research. Whatever one concludes about the specifics, the episode clarified that technical AI work and institutional pressures are not separable. It prompted a wave of academic and industry discussion about researcher independence that continues today.
AI literacy is not a destination — it is a practice. The field moves quickly enough that knowledge from two years ago may be significantly outdated. The most effective approach combines stable conceptual foundations (how do these systems learn? what do they optimize? where do errors come from?) with ongoing engagement with current developments.
Reliable sources for ongoing learning include: the AI Now Institute's annual reports, the Partnership on AI's resources, academic preprint servers like arXiv (particularly the cs.AI and cs.LG sections), and investigative journalism from outlets like MIT Technology Review, The Markup, and Wired. These sources vary in technical depth but all engage with AI's real-world effects rather than its marketing.
Princeton researchers Arvind Narayanan and Sayash Kapoor published "AI Snake Oil" in 2023 — a systematic critique of inflated AI claims across healthcare, criminal justice, and education. Their work documented dozens of products marketed as AI-powered that had no meaningful predictive validity. The book became a widely-used resource for practitioners, journalists, and policymakers trying to distinguish genuine AI capability from marketing. Developing the capacity to make that distinction yourself is a core literacy skill.
Using AI tools in professional contexts raises genuine ethical questions that most workplaces have not yet fully resolved. When you use AI to draft a document, are you being transparent about that? When AI assists your decision about a person — a hire, a loan, a medical treatment — are you maintaining appropriate accountability for the outcome? When your employer deploys AI systems that affect workers or customers, do you have an obligation to raise concerns you observe?
There are no universal answers, but there are emerging norms. The ACM Code of Ethics for computing professionals, updated in 2018, addresses AI specifically — including obligations to consider the impact of systems on affected people, to be transparent about automated decision-making, and to not implement systems known to be harmful. These norms apply broadly, not just to software engineers.
AI tools can subtly erode intellectual independence if used uncritically. When you outsource your thinking to AI for routine tasks, you may gradually lose the skills, confidence, and habits of independent analysis. Research on calculator use in education shows that over-reliance can reduce numeracy — the same dynamic may apply to writing, reasoning, and creativity.
A sustainable personal practice involves being intentional about when you use AI and why. Using AI to accelerate tasks you already understand well is different from using AI to substitute for understanding you have not yet developed. The former is a tool; the latter is a crutch that may weaken your judgment over time.
The next few years will be among the most consequential in the history of AI governance. Regulatory frameworks being built now in the US, EU, China, UK, and India will shape how AI systems are deployed for decades. The workforce transitions underway will determine who benefits and who is harmed by automation. The scientific breakthroughs being pursued — in AI safety, interpretability, and alignment — will affect what AI systems are capable of and whether those capabilities are safe.
You are living in the period when the choices matter most and when the outcomes are least determined. That is uncomfortable — and it is also an invitation. The people who engage thoughtfully with these questions, bring their professional expertise to bear on them, and participate actively in the institutions that govern them will have more influence than those who don't.
You have completed this course. You have some understanding of how AI works, what it can and cannot do, how it is governed, where it succeeds and fails, and what your role in it might be. That understanding is not fixed — it should grow, be challenged, and be revised. The commitment to staying engaged is, itself, the most important thing you can carry forward.
In this final lab, you'll work with the AI assistant to develop your own personal framework for AI engagement: how you'll use AI tools responsibly, how you'll stay informed, and what specific civic or professional action you're willing to commit to. The goal is something concrete and yours.