An Analysis of Engineering Students' Responses to an AI Ethics Scenario

Alexi Orchard, David RadkeAAAI 2023 · DOI

Introduction

As artificial intelligence (AI) systems become increasingly integrated into critical infrastructure, the industry faces significant challenges regarding algorithmic bias, mass surveillance, and misinformation. While engineering accreditation bodies like ABET and CEAB mandate ethics training, these topics are often taught in a "decontextualized" manner—frequently isolated as separate philosophy courses or brief modules on professional conduct. This research investigates the "ethical maturity" of engineering undergraduates, exploring whether they can bridge the gap between identifying a technical flaw and understanding its broader societal consequences.

Methodology

The study employed a mixed-methods approach involving 53 engineering undergraduates from the University of Waterloo, primarily from Systems, Biomedical, and Mechatronics disciplines.

The AI Bias Vignette

Participants were presented with a hypothetical "final project" scenario: a facial recognition model achieving 95% overall accuracy but failing significantly on individuals with darker skin tones. The scenario noted that the intended population was mostly represented in the test set, and that reporting the high aggregate accuracy would likely result in a top grade. Students provided qualitative responses explaining their intended actions and the factors they considered most important.

Curricular Context

To contextualize the survey findings, the researchers also analyzed five years of "Capstone" project abstracts (2018–2022) to quantify the prevalence of AI within the engineering curriculum.

Results: The Technical-Ethical Disconnect

The findings reveal a stark contrast between students' technical proficiency and their sociotechnical literacy. While the vast majority of students could identify the model's failure as a technical bug, very few recognized it as a systemic ethical issue.

Figure 1: Distribution of themes identified in student responses to the AI bias scenario.

As shown in Figure 1, 77.4% of students successfully identified the model's weakness (the accuracy gap). However, only 17.0% explicitly connected this failure to broader societal implications such as racism or discrimination. Most students (28.3%) focused on technical mitigation, with 73.3% of that group attributing the issue solely to insufficient dataset distribution. This suggests a "technical fix" reflex where bias is viewed as a data problem to be solved with more samples rather than an inherent ethical failure in design.

Context: The Ubiquity of AI in Engineering

The urgency of these findings is underscored by the rapid integration of AI into student projects. The analysis of Capstone metadata shows that AI is no longer a niche elective but a core component of modern engineering design.

Figure 2: Percentage of AI-related Capstone projects by major from 2018 to 2022.

Figure 2 illustrates that AI-related projects have trended upward across all studied majors, peaking at an average of 60% in 2022. This high exposure to AI development makes the lack of integrated ethical reasoning particularly concerning, as students are increasingly likely to deploy these technologies in professional settings.

Discussion: Barriers to Ethical Engagement

Qualitative analysis revealed several key insights into why students struggle to engage with ethics:

  1. The "Junior" Barrier: Students frequently cited a power imbalance experienced during co-op (internship) terms. Many felt that as junior employees, they lacked the standing to raise ethical concerns, fearing that senior team members would be dismissive.
  2. Ethics as a "Side Dish": Participants reported that ethics rarely feels applicable to their technical coursework. One student noted, "My other courses are technical and there hasn’t been a need to include any ethical reasoning."
  3. Sociotechnical Illiteracy: While students are "AI literate" in a mathematical sense, they lack the framework to see how a skewed dataset translates into real-world harm.

Conclusion and Future Directions

The study highlights a critical need for a sociotechnical pedagogical model. Ethics should not be a separate lecture but a standard part of the "debugging" process in every technical lab. Future work will focus on "scaffolding" ethical outcomes throughout the four-year degree, ensuring that students view ethical considerations as fundamental to their technical workflow rather than an optional supplement. By integrating ethics into the core of engineering education, institutions can better prepare students to navigate the complex societal impacts of the AI systems they build.

Summary updated

BibTeX
@inproceedings{orchard2023analysis,
  author = {Alexi Orchard and David Radke},
  title = {{An Analysis of Engineering Students' Responses to an AI Ethics Scenario}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2023},
  doi = {10.1609/aaai.v37i13.26880},
}
Figure