Using AI in engineering education: a balancing act, driven by clear purpose

Synthesis: Based on a questionnaire of 100 higher-education engineering students and a critical literature review, examines how students use and perceive LLMs. Students value LLMs for writing support, conceptual clarification, coding assistance, and brainstorming, but express concerns about inaccuracies, bias, overreliance, and academic integrity. Analyzes two dominant metaphors — Large Language Models (LLMs) as 'oracle' and 'tutor' — showing they cultivate expectations exceeding actual capabilities. Argues students' attachment to LLM efficiency reflects 'cruel optimism' — benefits depend on skills still being developed.

Key Findings

• In a questionnaire of 100 higher-education students, predominantly from engineering-related fields, LLMs were most valued for writing support, conceptual clarification, coding assistance, and brainstorming.
• Students simultaneously expressed concerns about inaccuracies, bias, overreliance, academic integrity, and the burden of verification imposed by AI-generated output.
• Two dominant metaphors — the LLM as "oracle" and as "tutor" — cultivate expectations of authority, expertise, and personalized learning that exceed what probabilistic text generators can actually deliver.
• Students' attachment to the promises of efficiency and personalized support reflects a form of "cruel optimism": the perceived benefits depend on the very skills, vigilance, and expertise that students are still developing.
• Student respondents who offered suggestions (N=27) called for assessment redesign — in-person essays, grading shifted from memorization to analysis, oral exams, and presentations with personalized questions — pointing beyond the reproducibility of knowledge.
• The chapter argues for a purpose-driven, context-sensitive approach to AI integration in engineering education, emphasizing critical AI literacy, reflective assessment design, pedagogical caution, and consideration of broader ethical and environmental impacts.

Assessment Design

The chapter treats assessment as a key lever, reviewing frameworks such as the AAA (Against, Avoid, Adopt) approach, which keeps lower-order assessment tasks supervised when AI can complete them (e.g., recalling facts, explaining concepts) and relies on contextualization in current affairs, personal experiences, and in-class engagement for higher-order skills. These recommendations align with the students' own suggestions for closed, supervised, or oral formats, and signal a shift in the instructor's role from policing AI use toward reflectively adapting assessment for AI's possible (mis)use.

What this means for practice

• Instructors. Gate every adoption decision on purpose: ask why GenAI is needed for this course or assignment and what added value it delivers before adding it, which is the chapter's own proposed approach.
• Instructors. Make AI Literacy and verification capacity prerequisites rather than extras, because the efficiency gains students prize are only realized by learners who already have the Critical Thinking and domain expertise to judge AI output — the skills they are there to build.
• Learners. Stop treating the Large Language Models (LLMs) as an oracle or a tutor: it is a probabilistic text generator whose fluent authority exceeds its reliability, and the verification burden lands on you and on your knowledge of the domain.
• Instructors. Redesign assessment toward contextualization, in-class engagement, and oral or supervised formats for lower-order tasks, which is both the AAA framework reviewed in the chapter and what students themselves proposed (N=27).
• Faculty developers. Support staff in shifting the instructor's role from policing AI use toward reflectively adapting assessment and pedagogy for AI's possible misuse.

Limitations

• The evidence is a questionnaire of 100 higher-education students conducted online in 2023–24; the author states she does not "pretend to generalize from the questionnaire alone" and describes its size and timespan as limitations.
• The sample is self-selected and international, dominated by engineering fields (52% of respondents), so the use patterns reported are not representative of all students.
• 19 of the 100 respondents had never used LLMs for their studies, so the reported use patterns rest on the remaining 81 users, and the questionnaire collected no learning-outcome data.
• The chapter is a conceptual and literature-based analysis, not a controlled study: it documents perceptions of value and risk and cannot establish an effect of LLM use on learning.

Citation

Olya Kudina (2026). Using AI in engineering education: a balancing act, driven by clear purpose. The Routledge Handbook of the Philosophy of Engineering, 2nd ed.