How can we utilise AI in higher education?

• This blog was kindly authored for HEPI by Professor Peter D Haynes FREng, Vice-Provost (Education and Student Experience), Imperial College London.
• HEPI is hosting two upcoming webinars: on the topic of climate change and student mental health on the 18th October with Student Minds and the UPP Foundation, and the relationship between teaching and research in UK universities on the 6th November with the University of Bristol. Registration is now open: you can register on the links above.

This autumn will see a strong government focus on artificial intelligence (AI) with the UK hosting the inaugural AI safety summit in November. In education, generative AI tools have both excited and alarmed the world over the past year, with tools like ChatGPT reportedly reaching over 100 million users within three months of its public launch. At Imperial College London, we recognise that generative AI is poised to transform many of our graduates’ careers, and we must therefore adapt our educational offering to provide them with the skills that will be in demand. Our staff and students are working together to realise the significant opportunities offered by generative AI and to address the evident challenges. Our initial approaches have covered a range of education activity, from assessment and feedback to entrepreneurship support and the wider student experience.

Providing instant personalised feedback to students

AI has significant potential to improve the university learning experience by enabling increasingly personalised and real-time feedback to students, particularly for many formative assessments. In fact, a team of staff and students have together developed a new web platform to deliver automated feedback on mathematical problems, with positive results. The platform – Lambda Feedback – currently hosts nine modules across eight departments, and is being tested by over 1,000 students.

The long-term goal of Lambda Feedback is to provide highly personalised feedback at the point of coursework submission, allowing students to receive constructive feedback instantly. This, in turn, can increase the time spent on more valuable discussions between staff and students during contact hours.

AI in student innovation

Imperial has a strong track record of nurturing student entrepreneurship by supporting our current students and alumni to develop their bright ideas into businesses of the future. The Imperial Enterprise Lab, is a dedicated entrepreneurship training centre, inspiring and supporting the next generation of student innovators and entrepreneurs, and AI will play a key role in its future.

For example, the Imperial College Business School and Department of Computing have together introduced an “AI Ventures” elective to help students launch new enterprises using or building AI. Students work in teams to workshop a business strategy and product idea during the module. This curriculum was adapted in January 2023 to explicitly incorporate generative AI as a ‘thought partner’ in creating the components of their business plans. Student prompts were restructured to encourage them to apply critical thinking and analysis to the outputs produced by Large Language Models.

Redesigning assessment

The emergence of generative AI requires us to ensure that automatically generated text or data is not used unethically in coursework. Imperial’s decision to eschew the use of AI detection tools has been vindicated by the high rates of both false positives and false negatives for AI content classifiers and their susceptibility to obfuscation attacks. This means it is vitally important that university communities understand which assessments are vulnerable to being compromised by unauthorised generative AI content. The Imperial College Business School IDEA Lab has proposed a ‘Generative AI stress test’ for all their taught module assessments, which is designed not only to evaluate and mitigate potential vulnerabilities in their assessments, but also to provide insights for staff into how students are using AI.

The stress tests comprise a four-phase process (Analyse, Review, Collaborate, Adapt), gathering insights that aim to help support students and staff to use generative AI tools effectively and appropriately, and to foster collaboration across the faculty to drive innovative teaching methods. The focal point of the stress test comprises a meeting between a faculty member and member of staff from the IDEA Lab, in which past indicative assessments are tested against generative AI tools Chat GPT4, Bard and Claude 2 using six criteria: accuracy, clarity, relevance, compliance, referencing and ease of use. Members of staff from across the whole university gathered for a workshop in August to put their own assessments through this process, and a number commented on how rapidly the capabilities of AI tools were developing. In my own case, a mathematical problem about heat diffusion dressed up as a wordy question about the cooking time for turkeys of different weights caused ChatGPT to identify the relevant equations rather than quoting from a recipe book. While its flawed attempt to manipulate the algebra created a risk of food poisoning, the resulting effort would have achieved a pass mark and I reflected that students could learn a lot from critiquing it.

Realising the potential of AI in education

Championing the responsible and ethical use of AI tools in education is central to realising a future in which this emerging technology is utilised positively from the outset. The recently developed Russell Group principles on the use of generative AI tools in education provide a guide for universities to support students and adapt teaching and assessment to promote ethical use of generative AI. Sharing the latest cutting-edge technologies across the sector is also key – Imperial has recently partnered with the London School of Economics, to host a two-day symposium on generative AI and the knowledge economy, exploring potentially transformative applications across education, economics, finance, law and regulation.

In their own position paper, the Department for Education (DfE) recognised the potential of generative AI to support excellent teaching, but also caution against potential malpractice. To truly maximise the potential for generative AI in a responsible way, we must consider how to provide a robust array of training and professional development opportunities for educators, ranging from workshops (like the AI stress test) and online courses to forums where best practices for using AI in education are shared and discussed. Alongside these efforts, further research into the application of AI in education must be championed and supported, and the Jisc national centre for AI in tertiary education is providing welcome leadership in this space. Above all, government, students and universities must work together to deepen our understanding of AI’s impact and effectiveness in the education system, to ensure we equip the current and future workforce with the skills to reap the benefits of AI for our society.