Artificial Intelligence–in the form of ChatGPT and its other large language model (LLM) cousins–has become available to most of us. 55% of the adult Danish population have used them, and only 8% claim to never have heard about them (September 2025 [1]). Among the age-group 18-29 years, 84% use LLMs!
In the UK, 92% of undergraduate students report using Generative AI tools, and 88% have used them in assessment to e.g. explain concepts, summarize articles, suggest research ideas or structure thoughts (February 2025 [2]). Students report both saving time and improving quality of their work as a result, and 63% think using Generative AI is essential yet only 36% have received training in AI skills, and there are signs of a digital divide in Generative AI use. Perhaps DTU is not much different from this perspective?
In May 2022, the Danish government expected digitalization and new tools to free up 10.000 full-time equivalent jobs in the public sector. With the advent of Generative AI the expectation now in 2025 is that it could free up 30.000 full-time equivalent jobs towards 2035 [3,4].
Generative AI tools will be ubiquitous–but to which degree are they useful, appropriate and responsible tools?
In academic or engineering contexts, they are perhaps both challenging and useful at the same time. Students could use them to quickly solve assignments that can be hard to differentiate from an original work. As a result, learning may suffer and we could end up graduating engineers that lack essential skills. Researchers could use them to quickly provide an overview of a new field, but it may not be reflected in a deeper cognitive insight and the satisfying acquisition of new conceptual knowledge.
In my own field, Computer Science, an emerging phenomena was humorously coined “vibe coding” by Andrej Karpathy (of OpenAI) in February 2025 [5]. Vibe coding is when the “programmer” doesn’t really write code, but acts as a director instructing an LLM to write and edit code, blindly accepting all suggested changes, and asking it to fix any errors itself. It may work surprisingly well and even untrained “programmers” can rapidly generate at least a functioning prototype.
Generative AI may initially save us time. But then, where is the satisfaction of engineering an effortful but beautiful solution, or finding an elegant mathematical proof instead of relying on a brute force exhaustive search, or achieving a new eye-opening insight that can lead to new discoveries? Perhaps we should even be more concerned on a basic level regarding the robustness of products created using Generative AI? Would you rely on a bridge over troubled waters if it were designed by “vibe engineering”?
Worryingly, one recent study found that a “higher confidence in GenAI is associated with less critical thinking” and “GenAI shifts the nature of critical thinking toward information verification, response integration, and task stewardship” [6]. Similarly, another recent study found “a significant negative correlation between the frequent use of AI tools and critical thinking abilities” [7].
Hence, it remains crucial to teach students basic engineering and problem solving skills as well as critical thinking, yet at the same time we must make sure Generative AI is embraced where useful and where it can be used responsibly [8,9] and we need to demonstrate how and where in our teaching. Furthermore, some students need to learn how such systems should be engineered to become useful and valuable for others.
In the words of Carl T. Bergstrom and Jevin D. West [8] “Ultimately, educators have to figure out how to get students to master basic skills, and also teach them how to use AI responsibly in their work.” And according to Michael Grove [9] “We must design assessments and learning experiences that either incorporate AI intentionally or make its misuse educationally irrelevant”.
Yet, often the discussion on Generative AI in academical settings revolve around cheating and lack of academic integrity. While those are important topics that may seem overwhelming or cause uncertainty, we need to move the discussion further. The use of Generative AI needs systematic, structured and coordinated thinking within academic disciplines as well as on program level [9], acknowledging we’re in uncharted waters!
Generative AI excels at thinking fast more than inviting to slow and reflected thinking [10]. Could we (re)design Generative AI systems or how we use them to instead cause us to slow down and focus less on problem solving and more on problem understanding [11]? Should we consequently redesign our assessment methods to emphasize conceptual understanding? And how do we adapt the way we teach to also demonstrate where and how Generative AI tools are useful and can be used responsibly, and when they cannot?
This will change our role as engineers and educators. Some of the effortful things we do now can be automated and done faster (and perhaps better) with Generative AI. But Generative AI can also augment the human-in-the-loop and allow us to improve quality [13] and build better and more advanced solutions.
It is in our hands to shape the direction. We need to move beyond “hype and dispair” [13]. As a university, we train the next generation of engineers that will be designing and using these systems. We impact policy makers. We provide foundational knowledge to society. Informed strategic decisions must shape the future to build human aligned systems [14], not the tech industry [15] alone.
For me, the human-in-the-loop remains important. Machines are built to serve humanity, not the other way round. I still enjoy finding elegant solutions to complex problems–when it happens (admittedly rarely)–even if Generative AI sometimes could have found a quick fix for me. I try to broaden the way I use Generative AI, both in research and in teaching, e.g. by asking for another perspective or a counterpoint to my own ideas instead of asking for an answer. And I love to learn and be inspired by others too, and to be part of developing new Generative AI solutions and systems!
How do we ensure that Generative AI becomes a vehicle for deeper conceptual understanding rather than just finding faster solutions? How can we inspire the next generation to remain curious, creative and critical in the age of Generative AI? How should we best have this conversation at DTU? What changes should we prioritize in our programs and curriculum? And what does it mean to your learning/teaching experience and to mine?
Per Bækgaard, Head of Study, Human-Centered AI MSc/Chairperson of the DTU AI Working Group
03.10.25
[1] Skjelborg, M.W.D. and Saaby, F.R.E. (2025): Over halvdelen af danskerne har brugt ChatGPT eller andre sprogmodeller. Dansk Erhverv.
[2] Freeman, J (2025). Student Generative AI Survey 2025. HEPI and Kortext.
[3] Olifent, L (2025). Se regeringens AI-planer: Her kan der spares flest lønkroner med kunstig intelligens. Ingeniøren, September 2nd, 2025.
[4] Fribo, A (2025). Regeringens AI-mål er håb og drømme forklædt som strategi. Ingeniøren, September 4th, 2025.
[5] Edwards, B. (2025). Will the future of software development run on vibes? Ars Technica, March 6th, 2025.
[6] Lee, H. P. H., Sarkar, A., Tankelevitch, L., Drosos, I., Rintel, S., Banks, R., & Wilson, N. (2025). The Impact of Generative AI on Critical Thinking: Self-Reported Reductions in Cognitive Effort and Confidence Effects From a Survey of Knowledge Workers.
[7] Gerlich, M. (2025). AI Tools in Society: Impacts on Cognitive Offloading and the Future of Critical Thinking. Societies, 15(1), 6.
[8] Bergstrom, C.T. and West, J.D. (2025). Modern Day Oracles or Bullshit machines? How to thrive in a ChatGPT world.
[9] Grove, M. (2025). It’s time we moved the generative AI conversation on. Higher Education Policy Institute.
[10] Kahneman, D. (2011). Thinking, fast and slow. Farrar, Straus and Giroux.
[11] Dalsgaard, P. (2025). Reflective Friction in Generative AI: Designing for Slow Thinking in Fast Systems. The Aarhus 2025 Workshop on The Role of Design in the Future of Human-AI Interaction.
[12] Jochim, J., & Lenz-Kesekamp, V. K. (2025). Teaching and testing in the era of text-generative AI: exploring the needs of students and teachers. Information and learning sciences, 126(1/2), 149-169.
[13] Zysman, J., & Nitzberg, M. (2024). Generative AI and the future of work: Augmentation or automation? Working Paper, Weizenbaum Series 38, Berlin: Weizenbaum Institute.
[14] Christian, B. (2020). The alignment problem: Machine learning and human values. WW Norton & Company.
[15] Vazquez, K (2025). Eccentrics and visionaries: The 15 tech bros who rule the world. El Pais, July 12th, 2025.