Introduction
Optimists tell us artificial intelligence will relieve people of drudgery, leaving more time for family, creativity, and meaningful work. Pessimists predict displacement, surveillance, intensified workloads, and greater inequality. Early workplace evidence supports pieces of both stories. The Gen X pessimism in me remains suspicious of promises that increased productivity will automatically become increased leisure. But optimists and pessimists should be able to agree on one proposition: AI should happen for us, not to us. The problem is that “for us” is often viewed exclusively as a technological claim. “For us” hides a normative claim.
Early workplace evidence complicates the promise that generative AI will simply give workers more free time. In some settings, AI-enabled productivity is associated with work intensification, greater task complexity, monitoring demands, and higher expectations rather than reduced workloads (Högemann et al., 2025; International Labour Organization, 2026). The Gen X pessimism in me suspects that when technology allows us to do more work in less time, institutions will often respond by giving us more work rather than more leisure. Early evidence of AI-related work intensification is not doing much to relieve that suspicion.
Both camps, the pessimist and the optimist, can agree that AI is something that should be happening for us and not to us. AI is part of a long history of technologies that change what humans can do, how work is organized, and how power is distributed. Past technological revolutions (the industrial revolution and the “information age”) brought with them “new” ethical considerations. I placed new in scare quotes because, ultimately, reasoning about ethical issues arising from artificial intelligence is fundamentally the same as ethical reasoning in other domains.
“For Us” Is a Moral Claim
The ‘for us’ hides the normative issues embedded in ethical AI literacy. Birmingham AI’s phrase “AI should happen for us, not to us” sounds almost self-evident until you ask what “for us” means. Does “for us” mean:
- greater GDP?
- higher productivity?
- lower labor costs?
- more leisure?
- better health?
- greater autonomy?
- more democratic control?
- fairer distribution of benefits?
- less drudgery?
- more meaningful work?
Different answers can support radically different AI policies while all claiming to make AI happen “for us.” Choosing among those answers requires judgments about what outcomes society ought to value and how benefits and burdens should be distributed. That is exactly why Ethical AI Literacy is necessary.
Defining “Ethical AI Literacy”
Ethical AI literacy is not a program for maximizing AI adoption. Competent ethical reasoning may justify using AI enthusiastically in one context, limiting it in another, requiring stronger safeguards in a third, and rejecting a particular use altogether. Ethical AI Literacy is
The capacity to recognize, analyze, and reason through moral problems arising from the development, deployment, and use of artificial intelligence, and to make defensible judgments about what humans and institutions should do.
Ethical AI Literacy is not:
• AI literacy
What is this system? What can it do? How do I use it?
• AI safety literacy
What can go wrong? How do I reduce technical or practical risk?
• AI governance literacy
What policies, controls, laws, and accountability structures govern its use?
• Ethical AI Literacy
What should I/we do, why, and who bears responsibility for that judgment?
Ethical AI literacy and AI governance literacy operate at different but overlapping levels. Ethical AI literacy develops the capacity to identify and reason through what ought to be done. Governance translates those judgments into institutional roles, policies, procedures, controls, oversight, and accountability. Ethical reasoning can therefore inform governance, while governance determines how ethical commitments are put into practice at scale. Ethical AI Literacy helps determine whether, when, how, and under what conditions adoption is justified.
Ethical AI Literacy: Four Questions
Ethical AI Literacy can be put into practice across multiple contexts by asking four questions:
1. BELIEF
When should I trust AI?
2. ACTION
When should I use AI?
3. POWER
When is AI being used against me?
4. RESPONSIBILITY
Who is responsible when AI causes harm?
These are not four principles telling people what conclusions to reach. They are four domains in which moral reasoning is required. Belief concerns how a person form beliefs under AI use and addresses issues such as deepfakes, rage bait, AI hallucinations. Action concerns what a person should do and addresses best practices with an individual’s use of AI for oneself. Power concerns issues of autonomy and independence as a result of AI use to influence, classify, evaluate, or exercise power over individuals and groups and addresses issues such in marketing, AI powered scams, criminal justice.
Responsibility addresses who is accountable when AI use causes harms to others (or even oneself) and addresses issues in criminal justice, data center siting, education, data management. Industry-specific examples follow.
Even Rejecting AI Requires Ethical AI Literacy
Ethical AI Literacy makes moral reasoning usable at the level of ordinary decisions without reducing ethics to a checklist. Even people who are opposed to artificial intelligence and want data centers banned still need to be capable of ethically reasoning about artificial intelligence. Unless AI is completely banned some day like in the post-Butlerian Jihad world of Frank Hubert’s Dune series, everyone falls somewhere within the “four questions.” Everyone may still be subject to AI being directed at them or to them.
Even contemporary AI opponents, including those who identify with a renewed Luddite tradition or advocate stopping AI infrastructure development, require Ethical AI Literacy. Refusing to use AI does not prevent AI from being used to influence, classify, surveil, persuade, or make decisions about them. Everyone will be exposed to generative AI rage bait, state-sponsored shitposting, deepfakes, phishing scams. Anyone could be the subject of AI use in the following industry specific examples; sometimes involuntarily.
Industry-specific Applications
Below is an overview of artificial intelligence use within specific industries and samples of questions that fall within the Ethical AI Literacy four questions spectrum.
HR: When and how can AI appropriately screen applicants avoiding discrimination?
Healthcare: When should clinicians trust AI analysis of medical images and AI generated healthcare recommendations?
Marketing: When does AI persuasion become manipulation or deception?
Sales: When does AI generated depiction of products or services alter consumer expectations and choices?
Legal: What must lawyers independently verify?
Management: What decisions cannot responsibly be delegated?
Nonprofits: When may AI appropriately influence decisions about who receives scarce services, funding, or organizational attention?
Government: When may governments use AI to make or support decisions affecting people’s rights, benefits, liberty, or access to public services?
Education: How can AI reinforce my learning rather than be a substitute for my learning?
AI infrastructure: When the benefits of AI infrastructure are broadly distributed, how should governments weigh environmental, economic, and social burdens concentrated in particular communities?
All of the industry-specific questions relate to at least one of the considerations in:
- how AI use affects the beliefs I form;
- how I should act when AI informs or influences my decisions;
- how someone else’s use of AI affects my autonomy and independence; and
- who bears responsibility for the consequences of AI
What Ethical AI Literacy Failure Looks Like
Artificial intelligence does not need to malfunction for its use to create an ethical problem. Sometimes a system performs exactly the function assigned to it while the underlying human decision about whether, where, or how to use it remains ethically defective. Several recent controversies illustrate different failures of Ethical AI Literacy.
Algorithmic Risk Assessment: COMPAS
COMPAS was used in parts of the American criminal-justice system to estimate the likelihood that defendants would reoffend. A widely discussed ProPublica investigation found substantial racial differences in prediction errors. Black defendants who did not subsequently reoffend were more likely than white defendants who did not reoffend to have been classified as higher risk, while white defendants who did subsequently reoffend were more likely than Black defendants who reoffended to have been classified as lower risk (Angwin et al., 2016).
The ethical failure is not adequately captured by asking whether COMPAS was “accurate.” Different kinds of errors impose different harms. A false positive can contribute to treating someone as more dangerous than warranted; a false negative can understate a genuine risk. Someone must decide which errors are more tolerable, what role an algorithm should play in decisions affecting liberty, and whether disparities across groups are morally acceptable.
The Ethical AI Literacy question therefore becomes:
When an algorithm helps exercise state power over a person’s liberty, who should determine which errors society is willing to tolerate?
That is principally a power and responsibility problem.
Automated Employment Screening: Mobley v. Workday
In Mobley v. Workday, plaintiffs allege that Workday’s algorithmic applicant-screening systems discriminated against applicants on protected characteristics including race, age, disability, and gender. The litigation remains ongoing in 2026, and these allegations should not be presented as judicial findings of discrimination. The court has nevertheless allowed significant disparate-impact claims to continue through multiple rounds of pleading (Mobley v. Workday, Inc., 2026).
The case exposes a recurring ethical problem created by technological delegation. Employers may argue that they did not design the algorithm; technology vendors may respond that employers chose to deploy it. Neither response alone settles responsibility.
Purchasing a decision system does not purchase a transfer of moral accountability. The Ethical AI Literacy questions are:
What decisions may an employer delegate to an algorithm? What must the employer know about the system before relying on it? Who is responsible when a vendor-designed system is used to make decisions about another person’s livelihood?
This is primarily an action, power, and responsibility problem.
Vanderbilt’s AI-Written Message of Empathy
After the 2023 Michigan State University shooting, Vanderbilt University’s Peabody College sent students a message about community, care, and empathy that had been drafted with ChatGPT. The disclosure that ChatGPT had been used generated criticism from students, and Peabody’s dean subsequently described the use as poor judgment and emphasized the need for human connection during tragedy (Wu, 2023).
Nothing about this case establishes that AI should never assist with institutional communication. The ethical problem concerns the purpose of the activity. If the purpose is merely to produce grammatically competent sentences, AI assistance may be unobjectionable. But if part of the purpose of the communication is to demonstrate genuine human attention, care, and presence, then outsourcing the communicative act can undermine the very value the message is supposed to express.
This distinction can be framed as:
ASSISTANCE → SUBSTITUTION → ABDICATION
The ethical question is not simply, “Did AI write this?” The ethical question is:
Did AI help a person fulfill a human obligation, or was it used to avoid exercising the very human judgment or engagement the situation required?
That is primarily an action problem.
Accusing Students of AI-Assisted Cheating
A 2023 incident at Texas A&M University-Commerce provides a useful reversal of the usual student-AI ethics debate. An instructor used ChatGPT itself in an attempt to determine whether students had used ChatGPT to produce assignments. Some students were temporarily assigned incomplete grades while the matter was investigated; several were later exonerated. The university confirmed that no students ultimately failed the course or were prevented from graduating because of the episode (Texas A&M University-Commerce, 2023). Reporting also documented students producing version histories and other evidence to demonstrate that they had written their work themselves (Verma, 2023). Universities have subsequently cautioned instructors against treating AI-detection systems as definitive evidence because false positives can lead to unjust accusations and undermine student trust (University of Iowa, 2024).
This case shows why Ethical AI Literacy applies to people judging AI use as much as to people using AI. A legitimate concern about academic integrity does not justify an unreliable method of determining guilt. The relevant question becomes:
What level of evidence should be required before an institution uses an AI-related inference to accuse someone of misconduct?
That is simultaneously a belief, power, and responsibility problem.
Uber Self-Driving Vehicle Fatality
In 2018, an Uber autonomous test vehicle struck and killed a pedestrian in Tempe, Arizona. The National Transportation Safety Board found that the safety operator was distracted and could likely have avoided the collision, but it also identified broader institutional failures. Uber’s automated driving system did not include an adequate safety risk assessment process, emergency braking was disabled while the vehicle was operating in automated mode, and the company’s safety culture was inadequate. The NTSB also identified insufficient state oversight as a contributing factor (National Transportation Safety Board, 2019).
The ethical lesson is that assigning responsibility to the safety operator does not eliminate the responsibilities of the company that designed and tested the system or the regulators who permitted its operation. AI-related harms can involve several actors with different degrees of knowledge, control, and obligation. The relevant Ethical AI Literacy question is:
When several people and institutions contribute to an AI-related harm, how should responsibility be distributed among those who knew about the risks, controlled the system, and had the power to prevent the harm?
This is primarily a responsibility problem and supports your principle that shared responsibility does not mean equal responsibility.
A Common Pattern
These cases look very different: criminal sentencing, employment, university communication, and academic integrity. Yet the same underlying failure appears repeatedly. Someone treats the presence of an AI system as though it settles a question that still requires human moral judgment.
COMPAS does not determine which errors society should tolerate.
Workday does not determine how employers should distribute responsibility for employment decisions.
ChatGPT does not determine whether an expression of empathy should be delegated.
An AI system’s judgment about whether a student used AI does not determine the evidentiary threshold for accusing that student of misconduct.
The Uber vehicle case does not determine how responsibility should be distributed among the safety operator, company, system designers, and government regulators.
AI systems can generate predictions, classifications, recommendations, images and video, or text. They cannot relieve humans and institutions of the obligation to determine when those outputs should be
trusted, acted upon, imposed upon others, or used to assign responsibility. That is the gap Ethical AI Literacy is intended to address.
From AI Adoption to Human Agency
Discussions about AI literacy often begin with adoption: Who is using AI? Who is falling behind? What skills do workers need? How can organizations integrate AI effectively? These are important questions, but adoption alone is an incomplete measure of whether artificial intelligence is happening “for us.”
A person can become highly proficient at using AI while gradually surrendering important forms of judgment to it. An organization can successfully automate a process while making it harder for the people affected by that process to understand, challenge, or appeal its decisions. A government can deploy an AI system efficiently while shifting power away from citizens who have little knowledge of how the system evaluates them. Increased AI adoption and increased human agency are therefore not necessarily the same thing.
Human agency requires more than the ability to operate a technological tool. It requires the capacity to form beliefs, make choices, evaluate reasons, exercise judgment, and remain meaningfully responsible for one’s actions. Artificial intelligence can strengthen those capacities when it provides information, exposes alternatives, reduces unnecessary labor, or helps people reason through complex problems. It can weaken them when people defer to outputs they do not understand, delegate judgments they ought to make themselves, become unable to distinguish persuasion from manipulation, or lose meaningful opportunities to challenge decisions made about them.
The four questions of Ethical AI Literacy are therefore also questions about human agency.
BELIEF asks whether I retain responsibility for determining what I have sufficient reason to believe.
ACTION asks whether AI is assisting my judgment or replacing judgment that should remain mine.
POWER asks whether another person’s or institution’s use of AI changes my ability to understand, contest, or independently respond to decisions and influences directed at me.
RESPONSIBILITY asks whether the delegation of a task or decision to AI has obscured the human and institutional obligations that remain.
Ethical AI Literacy does not require preserving every task humans currently perform. Delegating tedious calculation, transcription, scheduling, information retrieval, or other forms of labor may expand human agency by allowing people to devote greater attention to activities they value more. The relevant question is not whether AI has replaced some human activity. It is what kind of human activity has been delegated, what value that activity served, and what capacities or responsibilities may be lost through the delegation.
This distinction matters for the aspiration that AI should happen “for us, not to us.” If successful AI adoption is measured only by productivity, efficiency, capability, or the number of people using AI, then adoption itself has become the goal. Ethical AI Literacy introduces another criterion: Does the way we develop, deploy, and use AI preserve or expand people’s capacity to understand, choose, challenge, and take responsibility for the decisions shaping their lives? In short form, how does AI preserve human autonomy? AI adoption asks whether people can use artificial intelligence. Ethical AI
Literacy asks whether they can do so without surrendering the human agency that makes the technology worth using in the first place.
Conclusion
“For us” hides a normative claim. “…AI happens for us, not to us” introduces a normative criterion for successful adoption. AI literacy teaches people what AI can do and how to use it. Ethical AI Literacy teaches people how to decide when it should be used, when it should be trusted, when it is exercising power over them, and who remains responsible for its consequences. If the goal is for AI to happen “for us, not to us,” both forms of literacy are necessary.
Birmingham AI’s aspiration that artificial intelligence should happen “for us, not to us” identifies the right goal. But the phrase “for us” cannot be defined by AI itself. It requires human judgments about benefits and harms, autonomy and power, fairness and opportunity, acceptable risk, responsibility, and the kind of society technological adoption should help create.
AI literacy helps people participate in an AI-mediated world. Ethical AI Literacy empowers people to retain a voice in deciding what that world should become.
References
Angwin, J., Larson, J., Mattu, S., & Kirchner, L. (2016, May 23). Machine bias. ProPublica.
Brynjolfsson, E., Li, D., & Raymond, L. R. (2025). Generative AI at work. The Quarterly Journal of Economics, 140(2), 889–942.
Dillon, E. W., Jaffe, S., Immorlica, N., & Stanton, C. T. (2025). Shifting work patterns with generative AI (NBER Working Paper No. 33795). National Bureau of Economic Research.
Geninatti Cossatin, A., Ferrero, F., Ardissono, L., & Mauro, N. (2026). The autonomy equation: How agentic AI reshapes trust and workload in routine productivity applications. Information Processing & Management, 63(5), 104681.
Högemann, M., Hein, L., Britsche, J.-O., & Thomas, O. (2025). Technostress and generative AI in the workplace: A qualitative analysis of young professionals. Frontiers in Artificial Intelligence, 8, 1728881.
International Labour Organization. (2026, April 30). AI-driven intrusive surveillance and loss of autonomy at work linked to psychosocial risks for employees.
Karunakaran, A., Kellogg, K., & Wiesenfeld, B. M. (2026). Experimentalist intensification governance: Managing worker negative consequences associated with generative AI innovation work. SSRN.
Mobley v. Workday, Inc., No. 3:23-cv-00770-RFL (N.D. Cal. 2026).
National Transportation Safety Board. (2019). Collision between vehicle controlled by developmental automated driving system and pedestrian, Tempe, Arizona, March 18, 2018 (NTSB/HAR-19/03).
Northeastern University. (2026, September 2). What can the Luddites (and Ned Ludd) teach us about AI?
PauseAI. (2026). PauseAI proposal.
Sanders, B. (2026, February 23). Yes. We need a moratorium on data center construction. United States Senate.
Texas A&M University-Commerce. (2023, May 17). Texas A&M University-Commerce addresses concerns about ChatGPT in Ag classroom.
University of Iowa Office of Teaching, Learning, and Technology. (2024, September 30). The case against AI detectors.
Wu, D. (2023, February 21). Vanderbilt apologizes for using ChatGPT to write message on MSU shooting. The Washington Post.