My students have been turning to artificial intelligence (AI) for help with problems that, not long ago, they would have brought to me, a teaching assistant or a classmate.
At first, like many academics, I saw this as an assessment and academic integrity problem. If ChatGPT could produce an answer, how could we make sure students were doing the work themselves? How could we design assessments that AI could not complete?
The more I have experimented with AI in my teaching, however, the less convinced I am that this is the right question.
My research has also given me reasons to rethink. One study examining ChatGPT’s performance in solving chemical engineering problems found that about 95 per cent of its responses to questions requiring knowledge recall were correct but accuracy fell to about 41 per cent when questions related to complex, open-ended problems that required professional engineering judgement. Some of the tool’s incorrect answers were sufficiently plausible that identifying them required genuine engineering knowledge.
My earlier research also showed that student characteristics affect performance in a digital learning environment. It found that students benefit from digital tools differently, and how they engage with technology can influence its educational value. This shows that access to a powerful tool does not guarantee effective learning.
I have therefore moved away from thinking about how to AI-proof learning. Instead, I am interested in “AI-authentic”: learning that recognises that students use AI and ensures they develop the knowledge, judgement and skills to do so effectively. I have been experimenting with a simple four-stage approach:
Do → augment → challenge → defend
The order matters. AI comes after the student’s thinking, not before it.
1. Do: start with the engineering problem
I first ask students to have a go themselves. This does not necessarily mean completing a long calculation. It might mean defining the problem, identifying the system boundary, stating assumptions, deciding what information is needed or explaining which engineering principles they think are relevant.
The point is to get students to engage with the problem before an AI tool starts suggesting answers.
This is important because it is very easy to accept an AI-generated approach when you have not developed your own.
2. Augment: use AI as a tutor, not an answer machine
Once students have tried, I allow them to bring AI into the process but I encourage them not to start with: “Solve this problem for me.”
Instead, I ask them to give the AI tool a more limited role. “Act as an engineering tutor. Do not give me the final answer. Review my proposed approach, identify inappropriate assumptions, ask questions about my reasoning and suggest alternative approaches.”
In this situation, the AI is no longer simply producing an answer. It is being used to question and extend the student’s thinking.
3. Challenge: ask students to assess the AI output
At this stage, students compare the tool’s suggestions with their original work. I ask them questions such as:
- What did AI improve?
- What did it get wrong?
- Did it make different assumptions?
- Are the units and calculations consistent?
- Does the answer make physical sense?
- What would you need to check before using this professionally?
Sometimes I deliberately give students an AI-generated solution containing an error and ask them to find it. This task is a good test of engineering skills, because a student who does not understand a material balance cannot reliably identify an incorrect AI-generated material balance. The ability to write a clever prompt does not compensate for lack of understanding of the underlying principles.
This is one reason I am increasingly reluctant to think of AI literacy as separate from disciplinary knowledge.
4. Defend: make students explain and justify their decisions
The final stage is to bring other students into the process. Students share their work with another group whose members act as engineering reviewers. The reviewing group challenges assumptions, calculations and recommendations. The original group then must respond and defend its decisions.
In professional practice, colleagues review engineering work. Assumptions are questioned. Calculations are checked. Recommendations are challenged.
Why shouldn’t students practise that before they graduate?
What skills do future engineers need?
If AI can provide explanations, generate calculations and suggest possible solutions, we need to think carefully about where our teaching adds the most value.
For engineering, I would put that value in several areas:
- problem framing: understanding what needs to be solved
- technical reasoning: choosing appropriate engineering principles and methods
- verification: checking assumptions, evidence, calculations, units and physical plausibility
- professional judgement: making decisions when the information is incomplete or when there is no single obvious answer
- communication and collaboration: explaining, questioning and defending technical decisions
- responsibility: understanding that engineering decisions have consequences beyond getting the calculations right.
None of this means that engineering fundamentals become less important. In fact, I think the opposite is true. The better students understand the fundamentals, the better positioned they are to use AI critically.
The changing role of the educator
AI does not make an educator’s role less important. It allows us to focus more deliberately on the things that are difficult to automate: creating meaningful problems, challenging assumptions, facilitating discussion, giving feedback and helping students develop judgement.
That, ultimately, is what I think we should be preparing engineers to do: not simply to use AI, but to think, question, judge and take responsibility when doing so.
Salman Shahid is senior lecturer in chemical engineering and discipline head of education at the University of Manchester.
If you would like advice and insight from academics and university staff delivered direct to your inbox each week, sign up for the Campus newsletter.
comment