iSTock/muratdeniz Here’s a question that should keep us up at night: Why do we keep building AI systems that harm people?
It is not for lack of warnings. We know facial recognition misidentifies Black faces [1]. We know predictive policing tools target racialized communities [2]. We know hiring algorithms screen out qualified candidates based on patterns baked into historical discrimination [3]. We have seen the studies, read the headlines, and attended the conferences. And yet, the systems keep getting built. The harms keep happening.
The usual response is to treat these as technical problems requiring technical fixes. Better datasets, we say. More transparency [4]. Improved fairness metrics. These interventions matter—but as I have found in my research on design education, they are not enough. Because the real problem is not downstream in the code. It is upstream in the classroom.
The engineers and data scientists graduating today will shape systems that determine who gets a job interview, who receives medical care, who gets flagged by law enforcement, and who gets approved for a loan.
Think about how we train the people who build AI systems. Whether they are studying computer science, data science, engineering, or design, the curriculum follows a familiar pattern: Master the technical skills, optimize for efficiency, deliver solutions that work [5], [6]. Ethics, if it appears at all, shows up as an elective, a single lecture, or a checkbox on the way to deployment [7]. We are teaching people to ask “Does it work?” before we teach them to ask “Should it exist?”
This is a pedagogical failure with profound consequences. The engineers and data scientists graduating today will shape systems that determine who gets a job interview, who receives medical care, who gets flagged by law enforcement, and who gets approved for a loan. These are not just technical decisions—they are judgments about whose lives matter, whose voices count, whose futures are worth investing in. And we are sending people out to make these judgments with training that treats power, inequality, and structural harm as someone else’s problem [8].
Why Technical Fixes Are Not Enough
When AI systems fail, we often frame the problem as a data issue or an algorithmic oversight. The facial recognition that cannot read darker skin tones? Train it on more diverse faces. The risk assessment tool that over-predicts recidivism for Black defendants? Adjust the statistical weights. These technical interventions assume the problem is in the execution, not the conception.
But scholars like Noble [9], O’Neil [10], and Benjamin [11] have shown that these failures are rarely accidents. They are symptoms of deeper issues: who gets to define what counts as a problem worth solving, whose perspectives inform the design process, and whose experiences are treated as the “default.” When predictive policing tools target Black neighborhoods, that is not a bug—it is the system working exactly as designed, encoding decades of discriminatory policing practices into seemingly neutral mathematics [12].
The technical fix approach also misunderstands how trust works. In engineering and computer science, trust is often treated as a usability problem—something you engineer through transparent interfaces or explainable outputs [13]. But trust is fundamentally relational and contextual [14]. Communities that have been surveilled, excluded, and harmed by technology have every reason to distrust new systems, regardless of how polished the interface looks [15]. Without addressing the historical and structural conditions that shape these relationships, technical interventions produce only the appearance of trustworthiness, not its reality.
Al Systems as Rhetorical Actors
To understand why STEM education needs fundamental reform, we need to recognize what AI systems actually do. They do not just compute—they communicate [16], [17]. They persuade users to trust their outputs, authorize certain actions while foreclosing others, and shape what counts as legitimate knowledge. An algorithm that outputs a risk score is not just calculating—it is making a claim about someone’s future. An interface that requires certain documentation is not just collecting data—it is defining who counts as a legitimate subject. A system that automates decisions is not just increasing efficiency—it is redistributing authority and accountability [18].
Yet most STEM education treats these communicative dimensions as peripheral to the “real” technical architecture. Students learn to optimize algorithms without learning to recognize that every technical choice carries persuasive weight, that systems implicitly define the kinds of subjects they recognize and value, and that code is never neutral [19].
Four Principles for Critical Design Pedagogy
So, what would it mean to redesign STEM education around justice? Based on my research in design education and AI ethics, I propose four pedagogical principles that can guide this transformation.
Privilege Question-Posing Over Solutionism
Traditional engineering and computer science education centers on problem-solving: given a problem, find the most efficient solution. But this approach assumes the problem itself is neutral and given, rather than constructed through particular power relations and priorities.
Critical pedagogy starts by questioning the problem itself [20]. Who defined this as a problem? What assumptions are embedded in that framing? Who benefits if we solve it this way? Who might be harmed? These are not preliminary questions to address before getting to the “real” technical work—they are the real work.
Consider predictive policing. Framed as a problem of “efficient resource allocation,” it invites technical solutions about where to deploy officers. But framed as a problem of “community safety and justice,” different questions emerge: What creates safety? Who gets to define it? Why are police the default intervention? This shift in framing opens up entirely different design possibilities—or suggests that the technical system should not be built at all.
Cultivate Rhetorical Awareness
Teaching rhetorical awareness means helping students understand that every technical choice carries persuasive weight. This is not a call for technical practitioners to become humanists at the expense of their technical training. It is recognizing that the technical is already rhetorical.
Students who learn to see AI systems as communicative actors are better equipped to anticipate how their designs will be read, interpreted, and experienced by diverse users—and to take responsibility for those effects.
Students who learn to see AI systems as communicative actors are better equipped to anticipate how their designs will be read, interpreted, and experienced by diverse users—and to take responsibility for those effects. This means understanding concepts like the “second persona” [21]—how systems implicitly construct their ideal users—and the “null persona” [22]—those rendered invisible or illegible by dominant design choices.
Center Participatory and Community-Engaged Methods
Too often, STEM education treats technical expertise as a sufficient qualification for designing systems that affect real communities. But those most affected by AI systems possess essential knowledge that no amount of technical training can replace: knowledge of how power operates in their lives, how systems of oppression intersect, what genuine accountability looks like.
Participatory design means moving beyond extractive consultation (“We’ll build this, then ask if you like it”) toward genuine co-creation where affected communities shape the project from the start [23]. This is not just an ethical nicety—it produces better systems. When Black mothers who have been targeted by algorithmic child welfare systems help design alternatives, when disabled people co-create accessibility features, when incarcerated people inform criminal justice technologies, the resulting systems are more responsive to actual needs and less likely to reproduce harm [24], [25].
This principle requires humility from technical practitioners: a recognition that technical knowledge, while valuable, is not the only or even the primary form of expertise relevant to sociotechnical systems.
Teach Genealogical Analysis
Drawing on Foucault, genealogical analysis asks: How did this particular technical solution come to seem inevitable? What alternatives were foreclosed? What power relations does this approach naturalize [26]?
For example, facial recognition technology seems like a natural application of computer vision. But a genealogical analysis reveals: the historical entanglement of photography with colonialism and racial classification [27], the privileging of identification and surveillance as problems worth solving, the assumption that automated recognition serves public goods rather than enabling state violence.
Teaching genealogical awareness equips students to ask: What am I taking for granted? Who benefits from framing the problem this way? What different questions might we ask if we started from different assumptions about whose safety, freedom, or flourishing matters?
Bringing It Into Practice
These principles might sound abstract, but they translate into concrete pedagogical practices.
Reframe assignments. Instead of “Build a recommendation system,” try: “A community organization wants to help members discover local resources. What approaches might center on their actual needs rather than maximizing engagement metrics? When might a technical solution not be appropriate?”
Critical case studies. Have students analyze realworld deployments of AI systems—not just the technical architecture, but who commissioned the system, whom it was meant to serve, who was harmed, and what alternatives existed. Treat these not as cautionary tales but as entry points for understanding how technical and social forces interact [28].
Speculative design exercises. Ask students to imagine radically different futures. What would a world centered on care, not efficiency, look like? What would we need to change—technically, socially, institutionally—to get there? These exercises cultivate the imagination needed to see beyond current constraints [29].
Community partnerships. Build partnerships where students work alongside, not for, communities affected by the technologies they are learning to build. This might mean computer science students collaborating with organizers fighting algorithmic bias, or engineering students co-designing with disabled communities rather than designing “for” them. These partnerships should not be service-learning add-ons but core components of technical training.
Cross-disciplinary ethics modules. Create modules that bring together engineering, computer science, design, and humanities students. Have them analyze real cases of algorithmic harm, not to assign blame but to trace how technical decisions intersected with social context to produce harmful outcomes—and to imagine what different educational preparation might have enabled practitioners to do differently.
It is worth noting that focusing solely on individual moral development or personal character— teaching students to be “good” engineers—while necessary, is insufficient.
Centering marginalized voices. Most importantly, center the voices and perspectives of those most impacted by algorithmic systems. This means not just reading about algorithmic bias, but learning from Black scholars, organizers, and technologists who have been leading the work of algorithmic justice [30]. It means treating lived experience with surveillance, discrimination, and exclusion as a form of expertise that should inform technical education.
The designers, engineers, and computer scientists we teach today will shape the systems of tomorrow–we need to teach them to center justice, not just solve problems.
Beyond Individual Ethics to Structural Change
It is worth noting that focusing solely on individual moral development or personal character—teaching students to be “good” engineers—while necessary, is insufficient. The structural conditions within which technical work occurs matter enormously.
Organizations reward speed and efficiency, not ethical deliberation [31]. Institutional incentives favor solutions that can be patented and monetized, not those that genuinely serve marginalized communities. Regulatory frameworks lag decades behind technological capabilities. Individual practitioners, no matter how well-trained in critical thinking, will struggle to make ethical choices in environments that punish them for doing so.
This is why pedagogical reform must be accompanied by institutional change: restructuring academic departments to reward interdisciplinary collaboration, reforming hiring practices to value critical awareness alongside technical skills, creating industry partnerships that prioritize social impact over profit maximization, and developing policy frameworks that hold organizations accountable for the harms their systems produce [32].
Path Forward
Transforming STEM education will not happen overnight, and it will not be easy. It requires institutional commitment, curricular restructuring, and sustained investment. Universities need to hire faculty who can teach critical approaches to technology. Departments need to reconsider what counts as technical rigor and broaden it to include ethical reasoning [33]. Industry partners need to value critical awareness alongside technical skills when hiring and promoting.
But the urgency could not be clearer. AI systems are rapidly reshaping social life, often in ways that amplify existing inequalities. Every cohort of engineers and computer scientists we graduate without critical training is another cohort building systems that will harm people—not out of malice, but out of a pedagogy that taught them to see only technical problems requiring technical solutions.
The question is not whether STEM education will change. It is whether it will change quickly enough and deeply enough to meet the ethical challenges of our AI-mediated present. Will we continue producing technically skilled practitioners who lack the tools to grapple with power and justice? Or will we cultivate a generation of technologists who understand that building responsible AI systems requires not just technical excellence, but ethical commitment, critical awareness, and genuine accountability to those most affected by their work?
The choice is ours to make. But we need to make it now, in the classroom, where the habits of mind and professional commitments of future technologists are formed. Justice cannot be an elective. It must be foundational. The designers, engineers, and computer scientists we teach today will shape the systems of tomorrow–we need to teach them to center justice, not just solve problems.
- “We are teaching people to ask ’Does it work?’ before we teach them to ask ’Should it exist?”’
- “When predictive policing tools target Black neighborhoods, that’s not a bug-it’s the system working exactly as designed, encoding decades of discriminatory policing practices into seemingly neutral mathematics.”
- “An algorithm that outputs a risk score isn’t just calculating—it’s making a claim about someone’s future.”
- “Justice cannot be an elective. It must be foundational. The designers, engineers, and computer scientists we teach today will shape the systems of tomorrow.”
- “The choice is ours to make. But we need to make it now, in the classroom, where the habits of mind and professional commitments of future technologists are formed.”
Author Information
Kem-Laurin Lubin specializes in artificial intelligence and computational rhetoric, with research spanning design pedagogy, human-centered systems, and AI ethics. Her work examines how rhetorical frameworks and design heuristics can guide responsible AI development, particularly in socially, and culturally sensitive contexts. Lubin has a PhD from the University of Waterloo, Waterloo, ON, Canada. Email: k4lubin@uwaterloo.ca.
________
To read the full version of this article, including references, click HERE.
_______
JOIN SSIT