Education Committee
Oral evidence: The use of Artificial Intelligence and EdTech in Education, HC 151
Tuesday 10 November 2026
Ordered by the House of Commons to be published on 10 November 2026.
Members present: Helen Hayes (Chair); Jodie Gosling; Darren Paffey; Manuela Perteghella; Peter Swallow.
Questions 123-159
Witnesses
I: Chris Loveday, Interim Chief Operating Officer and Deputy Principal, City of Portsmouth College; Mary Meekings, Head of Equality and Policy, University and College Union; Ben Rowland, Chief Executive Officer, Association of Employment and Learning Providers; Cath Sezen, Director of Education Policy, Association of Colleges.
II: Charlotte Armstrong, Policy Manager, Higher Education Policy Institute; Professor Simon Banks, Vice-Dean, Education, King’s College London; Michael Webb, Director of AI, Jisc; Lewis Wilson, Vice-President for Higher Education, National Union of Students.
Witnesses: Chris Loveday, Mary Meekings, Ben Rowland and Cath Sezen.
Cath Sezen: Good morning, everyone. I am the director of education policy at the Association of Colleges.
Ben Rowland: I am Ben Rowland, the chief executive of the Association of Employment and Learning Providers. Our members are training providers on the frontline of Government programmes, and they deliver about six in 10 of all apprenticeships in England, as well as a number of other skills programmes.
Chair: I should put it on the record that Ben and I have known each other for many years as friends.
Mary Meekings: My name is Mary Meekings. I am the head of equality and policy at the University and College Union.
Chair: Joining us online, we have Chris Loveday.
Chris Loveday: Hi, I am Chris Loveday, the chief operating officer at City of Portsmouth College. I was formerly at Barton Peveril College.
Cath Sezen: I suppose it is still relatively early days, particularly for AI. In total, we have just under 40,000 enrolments in digital qualifications, across 16 to 18-year-olds and adults. About 20,000 of those are at level 3 for 16 to 18-year-olds, to give you an idea of where the bulk of enrolments lie.
Colleges are very keen to be at the cutting edge. We have a lot of sector expertise in college. Staff are utilising AI to support in preparing lessons and making lessons more engaging, and to support students with SEND. I think that that is very important. AI is also having a positive impact on workload. There are questions about how we best support all students towards the future in terms of AI and digital technologies, not just those 40,000 who are doing digital and AI qualifications. That is something that we need to bear in mind.
A last thing on this particular question is just to consider the role of what I am going to call the digiTECs—digital technical excellence colleges. We have a range of colleges of technical excellence, and five are in the digital area. They are focusing on that point about how they can support not only staff, but students to gain the necessary skills, looking particularly at things such as AI. I think that that is very important.
Chris Loveday: AI in particular can have a real impact on education through many lenses. The first one is obviously the workload for teachers, but I think it is important we understand that it can also have a hugely positive impact on the workload of the operational teams and the business services that support educational institutions.
We have a need and an obligation to train and educate the young people at our establishments on the safe and effective use of AI, particularly in and around what is appropriate for academic use, and understanding the challenges around misinformation, disinformation, bias and hallucination. Finally, there is the impact that it will have on teaching and learning and assessment.
There are pros and cons to all of these angles, but the important thing is that we do not lose sight of the fact that, whatever we do, it needs to be differentiated for both the staff and the student, because people will approach this from different levels. Most importantly, we have to continue to provide an opportunity for those who do not wish to engage with the technology as an approach. It is a very challenging environment, but the benefits and wins significantly outweigh the risks and losses at this moment.
Mary Meekings: I am here bringing the perspective of the staff. It is interesting to hear some of the reflections already. They are really diverse: UCU will have members who are designing this tech and are really involved in developing it, members using it at work and embracing it, and members with real concerns about what it means.
For me, that is exemplified by what I am told by FE staff—that it is different in different departments. It is not even college to college where we are seeing different levels of adoption. We have one department really embracing it and encouraging their staff to use it, and then we have a department in the same college rejecting the same bit of tech. So we have to be quite careful about how we talk about that in terms of those diverse ranges.
We are also hearing educators saying that they feel like they are teaching themselves how to use this technology at the moment—it is great to see Chris nodding along with me; it is good to have a shared understanding about that. There are massively different levels of confidence in using the tech, and we have staff telling us that they are using their own money to access some of the tech as well. We really need an urgent look at those points.
The worst-case scenario we have heard about is where humans are almost completely removed—where you have AI being used to develop a learning tool, the student is then using AI to respond to it, and then the marking is being done by AI as well, so you just have computer-to-computer interaction. I don’t think that is anything any of us here would want to see.
I want to pick up on the workload comments. That is not substantiated by evidence at the moment. We have a workload survey coming out and FE workload is through the roof and going up year on year. I completely share the aspiration—we want to direct this tech so it does help with that very real problem—but it is just not what staff are telling us at the moment in terms of their experiences of it. We know staff are using it, so there is clearly an issue there, because workload remains a real issue.
Our central concern, and this is shared by other education unions—we have been working alongside them—is that we need to see AI enhancing rather than undermining the quality of education and the professional autonomy of educators. The relationship between staff and students is key. Whatever happens next, that needs to be retained.
Ben Rowland: I would answer the question in two parts. The first is the impact on institutions; the other is how learners are benefiting from this.
We recently completed a study with one of our partners, Ufi VocTech Trust, on our members’ adoption of AI, which I will share with the Committee. That is obviously quite timely. Reflecting on that, they have made the furthest progress on back-office operations—timetabling, scheduling and so on; next is teaching—preparing teaching materials; and they say learning comes next. Last, in terms of the positive impact, is assessment, for lots of the issues that have already been raised, which we have some numbers on as well.
That is the impact on institutions. Among my members, I think there is a cautious and growing optimism around their ability to use it. They are typically very practical, outward-facing organisations—sometimes quite small—and they are looking to see how they can shave a bit off here, shave a bit off there and improve a bit there, so they are naturally quite experimental. I have to say that they are well supported by the software providers, many of whom they have been partnering with for many years. Lots of the AI innovation is going on in those partners as opposed to in the training organisations.
I think there are two types of learner. One is those who are doing the AI and digital programmes. That is a real beacon of success across the sector because apprenticeships in digital, AI and cyber-security are long-established and are often partnering very deeply with people such as Microsoft and Oracle who provide significant educational resources. We can be quite confident that apprenticeships in the digital space, including modern AI and cyber-security, are successful and mature.
There is a separate question about those learners who are not on digital programmes and how it incorporates into all their learning. We find the same thing that has been hinted at already: where teachers are unsure themselves and do not know how to use AI, it is very hard for them to engage in teaching young people how to use it.
Cath Sezen: I think that is a really interesting question. In terms of T-levels, we are seeing digital T-levels being updated to incorporate more AI. They are also being consolidated, which I think is a good thing because it is about making sure that young people are very clear about the qualification that they are taking and where it is going to lead them.
It is interesting that there is going to be only one digital V-level. You could argue that perhaps there needs to be more than one because it is so wide in terms of the occupations and impact. However, in terms of having a V-level, it does mean that you can mix and match. You could perhaps be doing some business qualifications as your other two V-levels, and then you could do a digital V-level. I think that offers an opportunity because, going back to what we have all been saying, it is not just about those young people who are going into a digital career; we are all going to need digital and AI. That is one of the things that we need to consider.
Realistically, to go into the industry, you probably need to have level 3 and above, so you also need to look at those higher technical pathways at levels 4, 5 and 6. However, when looking at level 2 and below, it is very important that those AI and digital skills are incorporated. I think it is something we have to consider as we move forward.
One of the key challenges that colleges always talk about in this space is the ability to move quickly enough in terms of qualifications. We have quite a regulated landscape here. I am not commenting on whether that is positive or negative, but the regulatory aspect means that sometimes, by the time the qualification is on the table and ready to deliver, there are new technologies that have come on board. One of the things that we need to consider is how we can speed up the process of getting qualifications on the table. When you have a lot of reform going on, that is quite challenging.
Ben Rowland: I think the exciting bit is that AI can really help in both of those. For V-levels, T-levels and anything where the vast majority is in the classroom, creating the experiences out of the classroom is the challenge—we know that is the case for things such as work experience—so simulation becomes really important. Flight simulators have been around for decades and are a crucial part of pilot training. Those simulators have never been around for other occupations because they have been just too expensive, until now. That bit is really exciting. You want to say to a young person, “We’re going to give you some experience—if not quite the experience—of what it’s like to be on a building site or railway track, or in a hospital ward.”
These technologies exist now and there are a couple to call out. One, which is well established in the college network, is Metaverse Learning. It is providing simulations where people can move around different environments, get feedback on what they are doing, and so on. You can imagine that being easily brought into pre-16. Another one, funded in part by our partner Ufi VocTech Trust, is called Bodyswaps, where you can literally step into someone’s shoes so that you can give feedback and then swap and be the person receiving your own feedback. That really brings it to life in an emotional way: “Gosh, okay, I know how I should be giving feedback, and it’s not like that.”
On maths and English, an interesting thing is going on. Again, a joint member is the Activate Learning group, which is using AI as a maths and English tutor for those who are borderline. This might be an interesting example of taking a human out of the loop, because one of the problems with English and maths for people who have really struggled through school and are having to do resits is feeling that they are being judged by yet another human being. Taking the human out of the loop and allowing a computer to be the person guiding you through what could otherwise be a really awkward interaction—because you are so painfully aware of your shortcomings in English and maths—might be a way of overcoming some of those barriers. There are some really interesting opportunities there.
Jodie Gosling: Chris, I can see you nodding, so we will come to you next.
Chris Loveday: I think this is really interesting, because for me, there are two lenses for looking at this. The first is the curriculum content. Whether it is V-level, T-level or English and maths, does the curriculum prepare the young person—the learner—for the world in which digital technology, including AI, exists, and how does the learner use it within that curriculum? If it is a technical or vocational qualification, they might be able to use some augmented reality and simulation, but also, for a business-related qualification, they might be learning how to use this technology to achieve outcomes within the curriculum.
The second lens is the student experience. For me, this is something we can often lose. I am talking about understanding how a student may engage with the technology, teaching them how to engage with it correctly and, most importantly, not losing the students who do not have access to it, through digital equity.
One of the biggest concerns I have around the evolution of digital and technology, including AI, in education is that the digital poverty gap widens to become a chasm and is no longer a gap. For me, there are four key pillars: the use of digital technology with a full-size qwerty keyboard; access to a device; access to the internet; and access to a safe and secure premium large language model.
We heard Mary refer to staff potentially paying for access to AI systems. The risk is, first, that that is not controlled within the organisation’s tenancy and environment and the data is potentially going into the public domain. I don’t believe young people are aware of those gaps, and if they cannot afford the paid version, they are using the free version. So the risk is this: what intellectual property, what data and what information are they sharing with that free version and how are they exposing themselves?
Those are the two lenses. The first is the curriculum content. The second is the student experience, digital equity and how we bridge that gap.
Mary Meekings: The first thing on the changes to the curriculum and the Government’s agenda is recognising workload. This Committee will be well aware of the system that we have at the moment in further education—a system that is really under strain. I spoke about workload in my last answer, and it is a really big problem, so any changes need to be managed extremely carefully and by really engaging with educators so that they are ready to embrace those and take them forward. I think that is the only way that that is going to work.
The concept of taking humans out of the loop is interesting. I think that in the future we will be screaming to bring humans back into the loop. Again, I think it is about getting a balance and really working that through.
I found the talk about digital equity really interesting, because it could flip the other way. Students could have too much exposure to machines, and then parents and teachers will be looking again for the human experience as the gold stuff in the system. I think we need to be a bit careful about how we talk about that, because we could accidentally create a digital underclass of people who are solely being taught by machines, and that then becomes the problem in our system. I think we need to be really careful about how we embrace that and take it forward.
Clearly, my view is that the way you do that is that you work with the staff and with their unions to work out what is happening on the ground. I think putting those guardrails in place is really important for the next step.
Mary Meekings: I think so. Someone needs to be accountable for it as well. That is the really important thing, so that we could not just say, in a blanket way, that this is how it needs to work in every college, university or school. We have to have that human accountability, so we know that someone is responsible for it.
Ben Rowland: Vocational training is a bit easier. For academic teaching, the tried and tested method for a couple of hundred years has been to have asynchronous written tasks, which AI has completely blown out of the water within a couple of years, whereas vocational training is all about wanting someone to do something. Proving that you can do something is less threatened by that fundamental change to academic learning. It is more about the opportunities side for FE and vocational education.
Cath Sezen: It is really important that any organisation does an audit about what works best where. The most important thing is that you are putting your students and staff at the centre of all this, and then building from there. We have already talked about Bodyswaps and the virtual reality opportunities that offers. That is a great idea.
I totally agree with Mary that the most important thing is having AI and digital technology that augments teachers, rather than replaces them. In terms of the work that is being done at Activate, I totally agree because it offers the opportunity for personalisation, perhaps for a young person who already feels that they have failed, in terms of English and maths, and just wants to sit down and focus on their own.
The key is thinking about what your strategy is as an organisation and how you best support students and staff to make sure that you help everybody reach their potential.
Chris Loveday: To echo Cath’s point and maybe elaborate a little, education is not one size fits all, regardless of qualifications. It is important that institutions recognise their level of preparedness to embrace both technology and artificial intelligence.
At the moment, if you do not have a robust infrastructure or if you do not have reliable wi-fi that can support student use, you can have all the VR headsets in the world, but they become unusable. It is about understanding your institution's position, where its baseline is and where you want to get to, and then moving towards it. You have to ask what problem you are trying to solve with technology to understand if technology is a solution or not. We sometimes lose sight of trying to solve a problem rather than chasing technology.
Cath Sezen: There are probably many things outside the remit of digital and AI that will help young people who are currently NEET. Focusing on the remit of digital and AI, for many young people it is about some of those skills: digital literacy, critical thinking and knowing how to utilise the technology effectively. As I have said, those skills are important not just if you are going to work in cyber-security, but if you are going to work in the health sector. We need to bear those things in mind.
On the other hand, we also need to bear in mind that some entry-level roles may be changing. That is not going to be the case in certain sectors. For example, in hair and beauty and in catering and hospitality you always have those entry roles because you are providing a service. But things are going to change in some sectors. We have seen changes in retail, for example, with things moving online.
We need to prepare young people, which is really important. We need to prepare them for the future, but we also need to take a step back and look at what happens earlier in the education system so that we actually prepare them for the future from a much earlier stage by providing vocational and technical options, which would also include elements of AI and digital. I am pleased to see that the curriculum and assessment review is looking at the importance of those literacies to support young people.
In terms of NEET, there will be some positives, but there are also some challenges that we need to bear in mind.
Ben Rowland: On the slightly broader question of the role of technology in whether we move from one in eight being NEET to one in six, getting a hold on social media is a real problem. A reason for young people being most at risk of not engaging in the labour market is a lack of confidence and self-belief. They are more likely to be hidden at home, staying at home and protecting themselves from the outside world, and social media becomes the thing they do. How we handle social media is a very big part. I am sure Alan Milburn will have something to say on that.
If the Milburn review is to talk about much better outreach to people who are stuck at home and do not know how to get out, as I imagine it might, the quality of information and how they use that information is important. We do not know who these NEETs are. We say there are nearly a million, but there is no list of a million people nicely organised across local authorities—it is a survey-based figure. We need to get hold of information, manipulate it and use it to direct this very scarce resource back to outreach work, where the human should definitely be in the loop. How do we equip those people to go and find young people with the right interventions at the right time, amplified by smart use of technology?
There is a finite and diminishing number of entry-level roles. It is really important that whatever recommendations Alan Milburn makes do not damage employers’ engagement with the apprenticeship system, including for older apprentices. The slightly counterintuitive point is that sometimes the best way to help someone who is NEET get into a job is to help the person who is already in that job and train them up. The idea that somehow apprenticeships should only be for young people is misguided, because older people need to be pulled up. That then creates an entry-level vacancy into which the young person can go.
The final thing I would say is that there will be huge opportunities in AI and digital, particularly around cyber-security, which is the corollary of any good deployment of AI. Our AI and digital apprenticeships are really strong in this country. We should lean in behind that and help them to flourish.
Chris Loveday: Building on what Ben has just said, prevention is better than cure. It is about the power of digital. In relation to the Milburn report, we are currently working on this at my college. We are aware of the factors that drive young people to become not in employment, education or training. Through school data and our college data, we can identify those at greatest risk and proactively target support at those young people on their journey to try to stop them becoming disengaged in education. The earlier we do that, the better.
At the moment, one of the benefits of this technology is that we are able to use live data, rather than historical data, with the triggers contextualised to our organisation and our catchment. Portsmouth is one of the highest-risk areas for potential NEETs. We are proactively trying to intervene earlier to support learners to stop them becoming disengaged and leaving the sector. Through trend analysis of student cohorts, we understand that students who arrive from a certain socioeconomic position or a certain postcode with an average GCSE score and doing a certain type of subject typically do not succeed or typically need more intervention. By using that type of trend analysis of historical data, we can target the student, intervene meaningfully earlier and allocate the resources directly to the right people at the right time. That is what we are working towards. That is something where technology will excel.
The role of technology in keeping young people in education through practical use is very different from identifying them before they become disengaged and trying to re-engage them, identifying the barriers to their learning and helping to break them down so that they want to stay in education, employment or training. It is essential that this technology helps to guide the strategy rather than the operational changes for each institution to target those young people.
Mary Meekings: That targeting sounds great. I am mindful that we are talking about so many people. We are umbrellaing the term “NEET”, which many of us do not really like—maybe we can think of something else to categorise it. The targeting is needed. Again, it is about how we direct the technology, rather than the technology itself being the solution.
I am also really mindful of the pace of change. If you have not been in education, employment or training for a couple of years, the tech is pulling away from you. Cath spoke about the importance of having this earlier in the education system, but I would also point to later in the education system, which means looking at our adult education system. We know there are grave concerns about the cuts that have been made to adult education, but surely that is going to be a key part of the retraining tools.
That would be a plea to the Committee: really look at the role of adult education and work on that. Again, I reiterate the role of educators in this. It is great if we can have that targeted use and the tools to support it, but human intervention is going to be critical.
Mary Meekings: I go back to what we were saying earlier. There should not be a sense that having access to the technology equals benefiting from the technology. There is a risk that we create a digital underclass of people who almost have too much exposure to the technology, so we need to be careful about determining it by saying, “The amount of technology equals the amount of benefit.” I think it is about trying to get that balance right.
Ben Rowland: It has always been the way with technology that you need a certain degree of confidence and self-belief that you can learn a new thing. Young people who have had the privilege or benefit of being in that position are the ones who are most able to take advantage, and the inverse is also true. It has always stretched the opportunity and capability divide. If you want people to be competent and comfortable with AI, it is about their underpinning education and the self-belief they have in themselves generally, not just about AI.
Cath Sezen: For me, it goes back to something Chris said earlier about equity of access. It is really important that we fund our organisations effectively so that they can update and provide those resources. Education should be your point of access, particularly if you do not necessarily have good wi-fi or the resources at home. Lots of young people will have a phone, but that is not necessarily what is needed when you are trying to teach people how to interact in a digital and AI world. I think it goes back to the point about equity and providing facilities for young people, notwithstanding the point that you obviously need to make sure that they are very well aware of how to use that technology appropriately.
Chris Loveday: The point about too much exposure to technology is absolutely valid, but having led a secondary school during the covid pandemic—where our reaction was to fast-track technology to people and then train them to use it in a way that was meaningful—I can tell you that technology is here and is prevalent in education. The digital divide is one that really poses a risk. Typically, the students who are excluded from education, employment and training are those from a lower socioeconomic background who typically do not have access to the technologies.
To the point that it could create a digital underclass, I agree that too much technology is wrong, but we cannot govern or regulate that at the moment. What we can control is the fact that we know that a lot of education is delivered online—whether through a platform or a digital solution like Teams, or using Google for submitting homework.
We must give students the access, literacy skills and confidence that make it accessible and convenient. For example, if you say to young people, “Okay, you don’t have internet at home. If you stay until 6 o’clock every night of the week, we will help you at the school,” they are not necessarily going to engage with that solution, so the risk is being perpetuated by the lack of technology availability and the lack of digital equity.
To the point about what happens if we achieve digital equity, it is important that we do not go too far and lose the thing that we inherently crave as individuals, which is human contact. I think there is a balance to be found, but the risk of digital inequity, for me, is far more prevalent and far more dangerous than the risk of a digital underclass with too much technology.
I am sure we have all heard parents and teachers being really positive about the potential of AI to support the learning of young people with special educational needs. There is also a lot of concern that there might be an overreliance on it. It would be good to have your expert view on how we get that balance right.
Chris Loveday: For me, it has to be operational-level again—based on the organisation. We built our own large language model for staff and students that was multilingual by design. It was accessible by design, with voice to text and text to voice. It had light and dark modes, and a colour-blind mode. We designed the tool to be accessible, but it was always designed to support the teaching and learning in the classroom—it was there to support the teaching and learning, not replace it, and the young people still needed a TA, for example, to support them.
The risk is that education is at a bit of a precipice, where recruiting staff to roles like TAs and LSAs is more challenging. Typically, the entry requirements are lower, yet these people provide the one-to-one support, helping young people with great need to navigate a curriculum that potentially the LSA or TA themselves do not understand.
Technology can play a role in supporting teaching but, speaking to the point raised previously, the risk is becoming dependent on it. It is important that the organisation, the institution and the individual have judgment. They get to know the young person and they get to regulate the use and the method of use, which is where guardrails are so important.
I advocate the solution of building your own model so that you can build in guardrails, so that it does not become overfamiliar and so it does not help you build a relationship or a bond with the technology, and that it remains precisely what technology should be—an enabler and facilitator, not a replacement for humanity.
This is where technology is now a risk. It is about making sure those guardrails are in place and that human oversight remains in the loop at all times, so that we do not see young people being adversely affected by relationships with a technology or dependence on a technology.
Chris Loveday: It is very early in adoption at the moment. What is interesting is that the more a grown-up tells a young person, “This is a great technology”, the less they want to use it, so not promoting it works. Like anything, it is about human engagement. If a young person is curious, interested and finds the benefit, then they engage, but for others I think there is a real apprehension.
A lot of our SEND students are not technology experts, so to give them more technology without human intervention is a fear. It is about the member of staff sitting with them and going, “Look, this is how we can use this technology.”
Where I have seen the technology excel is in English as an additional language. With people who do not speak English as their first language, the ability to converse in their natural language and have it converted to English, or from English to their natural language, has really helped EAL students to excel. Also, those with visual impairment have benefited, but it has to be a visual impairment that light or dark modes can support. Like anything, it has its place, but it is very early days to be able to say meaningfully what impact it is having.
Mary Meekings: I think we all agree that the guardrails that we are talking about are potentially not quite in place. That is what we need, and it is certainly what teachers are looking for.
From an equalities perspective, it is always quite concerning if people are being categorised and monitored and having decisions made about them. I think we would all want humans to be involved in those judgments and perspectives, rather than handing that over to the machines.
Of course we want digital inclusion, and we want everyone to have access to those things; I don’t want to suggest otherwise. The point we are trying to make is that we do not want to pass over our responsibility and accountability to those machines. That really is where Government can help by stepping in, looking at the balance and putting guardrails in place. Particularly with SEND students, we have a real responsibility to make sure that people are not further disadvantaged by judgments that might be made by the—I don’t want to say “technology”, actually; I specifically mean the AI.
Ben Rowland: One of the areas that is more advanced in the use of AI is creating tailored teaching materials. There is a solution called TeacherMatic, which a growing number of colleges and ITPs are using. That really helps teachers. Late at night, it might be, “It’s 10 o’clock and I need some fresh Quizlet cards now,” and it will help them to do that. But it can also be useful for saying, “Five of my students have needed particular extra help—can you remodel these questions or this material so that it suits them?” or “Can you just check that this is okay to work with for them?” In many cases, it says, “Yes, this is fine,” so they go in the next morning confident that it is going to work. One of my members in Southampton, Yuzu, is making extensive use of that technology.
There is another, quite tedious admin bit of this; I will write up the detail for the Committee afterwards. There is money set aside by the Department for extra learner support, but claiming it can be really painful, particularly for independent training providers. They have to interface with multiple local authorities in small numbers, so they are far down the list of the local authorities’ priorities. They often do not get it, so they then tend not to ask for it, which means that resources set aside to help them to help SEND students are not making their way to the frontline. I would like to submit a bit more information on that afterwards.
Ben Rowland: No, I don’t think it should be. I think it is easier for organisations. If you are an individual member of staff saying, “How do I make my job easier and more efficient?”, you can control that more. The moment you get into a world of saying, “Actually, this impacts other people and I depend on their input,” I think that is what slows people down.
Cath Sezen: We have to bear it in mind that SEND is a broad range. One of the great things about digital and AI is that a small thing can make a huge difference, such as with fonts. When we have written exams, it can be problematic to get the right font or the right colour background for the young person, but on screen that is easy. Again, it goes back to starting with the young person—or the adult—and thinking, “What do they need? What is best going to support them in their learning?”
I would like to draw one other thing to your attention, which is a slight curveball but is related to accessibility. That is the lack of information that colleges get about young people with SEND, unless they have an education, health and care plan—
Peter Swallow: And even sometimes when they do have one.
Cath Sezen: Yes, I was going to say; they are not always entirely up to date. But we do not get a lot of information for the vast number of young people. In colleges now, 29% of our 16 to 18-year-olds have SEND needs. If there were a digital solution for how that information could be transferred—obviously, with guardrails—we would be much better prepared to support those young people from day one or even during their enrolment process. That is really important and is perhaps something to consider as part of the work that you are looking at.
Chair: Could I just flag to members and witnesses that we are running quite behind? Given the progress we want to make and the number of topics we want to ask about, I make a plea for brevity in answers and questions, which will help us to get through as much as we can.
Chris Loveday: If you don’t know how to use a washing machine, you don’t get very clean clothes. It is exactly the same principle with this technology. At the moment, teacher workload can be adversely affected through trying to adopt AI without differentiated training for the teacher or support and clear guidance on its appropriate use and the way it should be used. I think that that differentiation does not exist in a lot of institutions.
On the point raised previously, a lot of the self-deemed experts are there because they have piloted or trialled the use of technology rather than having a knowledge and understanding of it. I have seen a positive impact on teacher workload, but that comes from significant engagement from the teachers in wanting to learn and develop, and from us as an institution being clear with our teachers about what they can use it for and how they can use it successfully, having done a lot of the prep work in the background ourselves before giving that technology over. It is important to remember that if you put bad in, you will get bad out. AI perpetuates that; it just accelerates the process. That is important to understand.
On the point about SEND, if you can upload the lesson resource that you have written for your class and say, “Differentiate it for these five students with these needs,” and if you are clear in your prompt, you will get a very good output that will really help you. But that comes down to your ability to understand the technology, what it does and how to use it. It can and does reduce teacher workload, but it is about how we train the people to use it and the effectiveness of the system we give them.
Mary Meekings: I am very grateful to Ben for raising the point about our members working at 10 o’clock at night preparing their lessons for the next day. That is far too normalised in our sector, but it is very real.
I will challenge this notion about workload. I do not think we have the evidence yet that this reduction is coming through the system, when you speak to teachers or educators in FE. We have just done a workload survey as a union, and it is going up every single year. I do not think it is on the teachers either, in terms of how well they are or are not using the technology. There is far more to do in the system to support them to realise some of these gains that we have been talking about this morning. I have heard that MPs’ offices are struggling from an influx of communication. That is happening to our educators as well: they have far more to respond to because people are able to produce things at their fingertips.
I am not sure we will get on to this point, but we also wanted to talk about how work is assessed. A lot of challenges are coming down the track and are already baked into the system.
Workload is through the roof. It is going up almost as quickly as the use of AI, so we do not have that correlation between AI and bringing workload down. That urgently needs looking at and these guardrails urgently need putting in place.
Ben Rowland: What we are experiencing in education, as described by Chris and Mary, is replicated in the wider economy. Studies from employers across the world show that workloads are massively going up because of AI—with a hope that at some point we will all figure out how to use the washing machine, as Chris says, and it will start to come down, but not yet.
Cath Sezen: I think it can reduce workload, but it needs to be seen within the wider strategy for the college, including supporting training and giving people confidence.
Chris Loveday: I am not advocating AI creating the lesson plan; the lesson plan is still created by the teacher, but it is differentiated for the individual needs of the other students. Being able to create six variations of the same lesson is a strength that can come from AI.
Although AI can play a big role in marking and assessment, I am concerned that it should not play that role, because you often learn more about your students through marking than through anything else. That, to me, is a concern. We do not want a dystopian world in which AI produces the answer, AI marks the answer and humans do not ever interact, so there is a challenge there.
Every digital revolution sees us lose some skills, so it is about which skills we prioritise as a sector and how we find the balance to keep the ones we want. If planning lessons is a skill that we want to keep, we plan the original lesson and use technology to help us differentiate it. It is about understanding that.
On Mary’s point around workload, there is no consistent approach to the adoption of AI, the training or the platform, so you will not see a universal decline in workload. I can tell you that in institutions where I think it is being adopted well, we are starting to see signs of workload decreasing. I do not think you can take a one-size-fits-all approach here.
Chris Loveday: We need to start assessing some of the softer skills that humans require, like critical thinking and presentation skills. Those things are essential now. We used to assess them many moons ago; we have lost that a bit, and it is important that it comes back. Ironically, there is a lot of uproar around students still sitting in rows hand-writing answers to questions, but I think the answer to the AI revolution might be to sit people in rows in a timed environment where they hand-write answers. Assessment is a challenge, but we need to start assessing the skills, including the softer skills that Skills England’s employability work has recently flagged. That is an essential part of the evolution of assessment.
Cath Sezen: Assessment presents us with a number of challenges. We hear about plagiarism and malpractice in coursework. We have to grasp it by the neck and think differently about how we assess. Particularly in vocational and technical, the most important thing is that you can do the thing, rather than simple knowledge recall and writing about it. We need to be thinking much more about practical assessment. A lot of people we talk to in colleges feel that sometimes expertise is being taken away now, because we are having more external assessment. Perhaps this is an opportunity for a complete rethink of how our assessment operates.
Ben Rowland: Assessment is the area that appears to be furthest behind when it comes to AI’s impact. In a recent survey we did, 80% said that it had improved operations, 70% said that it had improved teaching and 66% said that it had improved learning, but only half said that it had made an impact on assessment so far. There is still a bit of a disconnect between what goes on with Ofqual and what happens on the frontline, with awarding organisations in the middle. They are capable and willing to experiment with more innovative forms of assessment, but they feel that Ofqual is not quite with them, and yet Ofqual seems to think that it is on top of things. There is a bit of a disconnect and a bit more of a discussion to be had.
On what we should be assessing people for post 16, I am part of a small group of people who have come together to look at work readiness, which is slightly different from Skills England’s list of skills. Rather than talking about communication, teamwork or critical thinking skills, this is about whether they have the understanding and attributes to be ready to engage in the world of work. Sometimes it is higher-level, and sometimes it is about simpler things like whether someone understands that they need to take instructions, that their fellow workers depend on what they do and that turning up on time is important. There is a bit of a gap in the system in terms of what work readiness means, and I am part of a group trying to fill that gap. We are involving hundreds of employers in defining that, all of whom are saying, “All the specific skills we can probably teach. It is about whether they’ve got the raw material right.” I think there is something interesting there.
Ben Rowland: There are lots of great examples where it is being used, where people kind of get it instinctively and are able to deploy it, but perhaps a lack of understanding about why it is good and why it works in the way it does prevents them from getting the most out of it. Crucially, it prevents them from communicating to others about why this technology might be good. Of course, the biggest thing about simulation is that it recreates an emotional state inside the person, which is the crucial thing that you cannot get from anything else.
Mary Meekings: Picking up on what Cath was saying, a redesign of assessment is clearly needed. Educators are telling us that it already feels as though there is an arms race: they are trying to keep up with what the students are using to submit their assessments and trying to respond to that. That is a very real and live issue. Our plea would be for that to be led by educators. Yes, there might be things that we can hark back to—we can look at what we did in the old days to help us understand how we might use technology, including AI—but I do not think it is either/or. I think it is about working with teachers to rethink and redesign this.
Cath Sezen: I honestly think that it is quite difficult, with the way assessment is now set up, because we have removed a lot of the coursework and a lot of it is done under controlled conditions. You were asking about V-levels earlier, for example. Those will all be done under controlled conditions. The challenge is in finding that balance where you have control but are not overburdening, by which I mean overburdening organisations, students and staff. Perhaps this offers the opportunity to sit back and ask whether we should think slightly differently.
Chris Loveday: I was fortunate enough to support the Chiltern Learning Trust in producing the materials for the DFE. What is crystal clear is that the sector, technologically, is in a very damaged state. We all have different systems, different infrastructures and different approaches.
The Windows refresh recently caused uproar within the sector, because people had not maintained their estate for the last 10 years, and then they want to adopt a technology like AI. Most people were looking to the DFE guidelines to give them definitive black and white answers: “This is what you can do, this is the platform you should use and this is how you should use it.” The DFE is not in a position to give that, because of the fragmented nature.
People who are leading institutions—I count myself as one of those—are not experts in this sector. They have to really dedicate themselves to upskilling and learning to be able to make meaningful guidance. There are colleges up and down the country that will be putting intellectual property into the free version of ChatGPT and breaching GDPR by inadvertently uploading a classroom lesson list to AI and asking it to give some attendance targets. These mistakes will be happening at operational levels, and leaders and governors will not be aware. I have supported schools and colleges up and down the country, and I have had to flag, “You cannot do that.” Most do not have a meaningful AI policy, even if the policy is, “Do not use it.” Staff do not know what they can and cannot do if the policy is not clear. I want to be really transparent: I think the governance of this is the biggest risk to the education sector.
Chris Loveday: I do not believe it should lie with the suppliers, because the reality is that they are trying to sell us a product. Typically, the product is quite expensive and will go up year on year, which creates a barrier to exit. Once they capture us as educational institutions, we are very bad at leaving, which means that they have a customer for life in many cases. So it does not sit with them; it sits with the institution.
I do not think many institutions have data protection officers. I have done the training and I still do not consider myself an expert, but I have to make a judgment call. Further support and guidance to the sector is needed on what is and is not a safe or effective system to use.
Ben Rowland: Across independent training providers, there is a correlation with scale. The smaller they are, the less comfortable they are to invest, and the bigger they are, the more so. You would expect that but there is a different dynamic, which is a real plus for the system. A large number of my larger independent training provider members are owned or part-owned by private equity, which bring in the standards and disciplines that they expect across their portfolio of companies, many of which are not education companies. They can say, “This is how we do it.” They have a presence with some of the other big technology players, which means that they can ensure that the smaller organisations—some of which are my members—are really in the right place when it comes to discipline, regulation and governance. It is a bright spot in that regard.
Cath Sezen: AI governance is absolutely the No. 1 thing that we are asked about, particularly by governors in colleges. We produced a road map last year in response to feedback from colleges. We worked in partnership with Ufi and Jisc, and one of the areas that we looked at was leadership culture and governance. That road map provides a baseline. Everybody is working at different levels, and it is really difficult to control. I described this space as like a plate of spaghetti. You don’t necessarily need to control it, but putting in the guardrails and creating the system is really important.
We are keen to do more work. We all need to work together to ensure that we are not overlapping and that we do not have different people doing the same thing in different spaces. I know it is really tricky because it is such a big space, but we are keen to work with Ufi, Jisc and others to support the sector to move forward more coherently so that every young person, adult and member of staff gets a similar experience. I think that is really important.
Chair: We will go back to Peter for the final question.
Ben Rowland: The first thing I would say is that it is encouraging that they are aware of that—if only voters of all ages were equally concerned about mis and disinformation. It is not straightforward, particularly for an independent training provider or a college delivering an apprenticeship, when time is so scarce with the person they are supporting, but there are parts of the programme that they are required to do—British values, Prevent and so on. They are always looking for new and interesting ways to provide a hook for that. For younger learners, the vote potentially coming in could be that hook. There is a way in which you could do it, but given the pressure on time, it is hard for them to do.
Cath Sezen: I agree. There is a pressure on time, but this is so important. If we are giving people this opportunity, they need to be well informed. It goes back to the whole point about use of digital and AI technologies. It is about critical thinking, information literacy and digital literacy, and responsible use. It has to be a priority. In many respects, we are perhaps now catching up on AI and digital technologies, but Pandora’s box is already open. In this case, perhaps we have the opportunity to make sure that we are better preparing for this particular initiative, making sure that it is positive and that all young people are supported. In that case, you need some sort of entitlement or an agreed approach.
Mary Meekings: This really speaks to the importance of our educators and why we need them so much. It is very much about the skills we have talked about, such as being able to analyse things, check your sources and understand what you are looking at and reading, which can really only come through educators. I echo that I do not think we should single out young people for this. It is needed across society. Access to those human teachers is important for everyone.
Mary Meekings: I think they have the skills; I don’t think they have the time.
Chris Loveday: It is really important that we go back to the point of educating young people—or anybody, actually—on the safe and effective use of AI, understanding how you essentially sense-check the output to make sure that it is valid and consistent. If you can educate them to do that in life, they will automatically apply it to mis and disinformation.
I echo the sentiments about older generations. I do not mean to be ageist, but for anyone who is not a young person in education now, many of us have the same gaps. We still do not understand how to differentiate mis and disinformation that has been AI-generated. Gone are the days where the human images had six fingers. It is now very believable. How do we educate everybody in general to stop believing that something they saw on TikTok is real? That, essentially, is a societal problem we need to address, starting in colleges with younger people.
Chair: Thank you all very much indeed; that was fascinating. There were a couple of mentions of further information that you might provide to the Committee afterwards. We would certainly welcome that, if you have further thoughts or more detail that we should be aware of. Thank you very much for taking the time to be with us this morning.
Witnesses: Charlotte Armstrong, Professor Simon Banks, Michael Webb and Lewis Wilson.
Michael Webb: Hi, I am Michael Webb, director of AI at Jisc.
Charlotte Armstrong: I am Charlotte Armstrong, policy manager at the Higher Education Policy Institute.
Professor Banks: Hello, I am Simon Banks. I am vice-dean for education in the faculty of natural, mathematical and engineering sciences at King’s College London.
Lewis Wilson: Hi, I am Lewis, vice-president for higher education at the National Union of Students.
Lewis Wilson: I think the crux of the question is why students are using it so much.
Chair: The question is really about the implications of that use. We know that the use is very wide; we can maybe get into a discussion later about why that is the case. What impact is it having on students’ ability to properly understand their subject? To what extent are students using AI to leap over a hurdle of assessment—“Can I get my essay done this week, and can I get a decent mark for it?”—and to what extent is it playing a role in whether they emerge at the end of their three or four-year degree course with a deep understanding of the subject that they set out to study?
Lewis Wilson: Sure. I think students are using it in a multitude of ways. It is not just, “Write me 3,000 words on this essay title so that I can get a 2:1.” Universities are having to make cuts because of the financial situation, and those cuts tend to affect the frontline members of staff that students interact with the most. When a student cannot access a writing skills lesson or an office hour with a university lecturer, they are pushed towards using AI, whether that is, “Help me draft a skeleton argument for this essay”, “Can you think of some synonyms for sentence starters to outline this point?” or “Can you help me with a mark scheme so that I might be able to achieve that first?”
To bring you back to the question, what impact is it having on students’ learning? I am not asking why they are doing it, but what impact is it having? In an age of AI, are students emerging at the end of their three-year course with a better understanding of physics, geography, history, business studies or whatever they set out to study, or is it compromising the depth of the knowledge that they are getting from their degree? That is what we want to explore.
Lewis Wilson: I do not think it necessarily leads to skills erosion, in so far as it amplifies a student’s ability to focus on other areas. However, I do think it comes back to that time piece: students lack time and resource, so they have to take the shortcut of AI. That might lead to a long-term skills erosion in the workforce.
Professor Banks: How they are using it and what the impact is depends very much on the individual student. We are seeing a positive impact on students’ learning where they are engaging with it appropriately and using it as a tool to enhance their learning. Students who are perhaps more AI-literate, or who have had some support and training in the appropriate use of AI, can use it as a force multiplier: they complement what they get from their studies and enhance the cognitive process.
With other students, there is a risk that they are using it in a way that is effectively—I think we are all familiar with the term “cognitive offloading”—outsourcing the very purpose of the learning, which is the engagement and the thought process, for a whole host of reasons. I would characterise the sector as being very much in transition. If you asked these questions two years ago or last year, you would have had very different answers. If you ask them again in two years’ time, the evolution of the tools is such that the way in which they complement learning will change significantly, as will the risks they pose.
Regarding the impact at the end of that journey through university, are students going out less equipped? Are they going out having had a less rich experience? That entirely depends on the way in which they engaged with those tools while they were there.
Professor Banks: I think it is a mix of all the above. Of course that is nothing new, in the sense that there have always been opportunities for students, in a range of different ways, and students avail themselves of those opportunities or not depending on their investment in their learning and their personal circumstances. I am sure we may pick up on questions about equity and so on.
There is obviously also the question of how the university designs its teaching and assessment, and designs the support that it offers students and staff, because that frames the context in which the students make those decisions.
Charlotte Armstrong: To give a brief bit of context, HEPI produces its generative AI survey, which surveys thousands of students on their AI use. It has been running for three years and that is where a lot of my data will come from.
One question in that survey asks what effect the student thinks AI is having on their student experience. We found that 49% of students say it is making their student experience better, 16% say it is making it worse, and 35% say it is having no significant impact at all. When you go into the two groups who say it is making it better or worse and try to find out why they give that answer, the students who say it is making their experience better tend to draw on things such as their appreciation for getting access to 24-hour support. They feel that they can use AI to ask questions that they might be a bit embarrassed to ask a lecturer because they feel that the content has been gone over and they should understand it already. Some students say that it is making their experience better because they can get their work done quicker and have more time to do other things.
The interesting thing about the students who find that it is making their experience worse is the sense of guilt that pervades their answers. We had a lot of responses along the lines of “I feel like this is making me stupider,” “I feel like I am not using my brain,” and “I feel like I am not learning any more.” If you cross-reference their answers throughout the survey, you see that those are students who are using AI quite frequently, so there is a real consciousness among that 16% of students that what they are doing is negatively affecting their degree outcome in terms of their actual learning—though not necessarily their grade—yet they are choosing to do it anyway. That is often because of the speed and ease that it gives them.
However, when we look at how students are using AI explicitly, very few are just generating an essay and inputting it without editing it—that is 12% of students, although that has gone up over the three years of the survey. Students are more likely to do it for things such as summarising content, helping with literature reviews and asking in a collaborative process way. We find that AI is used more frequently as a study aid than as a way to complete the work, although of course some students are doing that.
Michael Webb: Our experience maps very much to what the other panellists say. An extra bit I want to add is that in Jisc over the last three or four years we have done what we call student perception workshops, where we get a bunch of students together and get them to talk about AI. They said early that they recognised the impact that it was having on their learning. Slightly differently to what Charlotte is saying, they said to us that they were pulling back on their use intentionally, which we took as a really positive sign.
There are lots of fantastic individual examples of a positive impact on the learning experience. I was at a session where a lecturer in green chemistry showed how they built interactive simulations really quickly with AI tools, and that was really helping students to learn more about the topic. He was teaching them how to use AI to select the right experiment. That meant that they spent more time hands on, on the right thing. There are lots of positive examples, but they are very specific and patchy.
Charlotte Armstrong: There are a couple of different things here; we found it is really subject-dependent. Students who are studying arts and humanities subjects are broadly less confident that AI will be useful for their future careers, but they are also less confident that their teaching staff are equipping them with the skills they will need. It is interesting because they are falling down on both gaps there. When you look at the other subjects, broadly 70% of students across the board say they think it is useful and it will be essential for their future careers, whereas for arts and humanities, you are down at around 50% or 60%.
It is very much on an institution-by-institution basis. We have seen over the three years we have run the survey that Russell Group institutions have made some very rapid progress over the last year in adopting and encouraging AI use and giving clearer guidelines around what students can use. But as was said previously in the FE session, even when institutions have across-the-board strategies, there are often still differentiations between subjects and even individual lecturers or tutors.
Anecdotally, I have heard of a student who was told by their tutor that they were not allowed to use AI to complete their assessment, while another student on the same course in the same year was told that they could. This kind of confusion for students is leading to that feeling of guilt. When we find out why students avoid using AI, the most common response is anxiety about being accused of cheating. Again, that lack of clarity is feeding into the concern and legitimate emotions that students feel about their AI use.
Professor Banks: Picking up on Charlotte’s point, she evidenced some data from HEPI. At King’s we have the artificial intelligence institute, which came together with the policy institute to conduct a survey that has a wealth of interesting information about perceptions of the future of work. Picking up on your question, 60% of university students who were interviewed as part of that survey felt that universities were well placed to provide those skills for their future careers, but only 36% thought they were doing so at the moment.
Are universities doing what is needed at the moment to provide the skills and the training? The answer is that it depends on the university, where they are on that journey and how they are working with their students. Again, that will be a crucial part of that movement across. It is not just about how students use AI to study. There is an ecosystem here: how are students using AI to study, how are staff using AI to enhance the learning progress and how are students learning the skills that will be useful for their future careers. They are all very closely related, but there are differences in what is required to deliver the skills in relation to each of those things.
Universities are doing a huge amount at the moment. I am rarely in a meeting that is not in some way focused on the impact of AI on what we do and how we do it, but the solutions we design will change. We have just heard some really interesting things from colleagues in further education. The students we have at the moment have not come through with that earlier-years exposure to AI, and the training in AI skills. The students who will arrive with us in the years to come will have had much more of that. That will mean we will have to shift, as that wave comes through, until we hopefully reach something approaching a steady state. That steady state will have to be adaptable to the changes in AI practice as we go through.
What are we doing? There are all sorts of things going on in terms of supporting students with upskilling. Science and engineering are a great training ground for this. We have colleagues working in engineering who are using AI tools that are absolutely deeply embedded in their teaching. The students are using these tools, for example, as part of the design cycle to work through: they have a range of specific tools targeted at each part of that cycle, not just generic LLMs and “How are we doing it?”
Professor Banks: Absolutely. The answer that I gave before was specifically to the question of how we are helping to upskill people in relation to AI. Really usefully, the whole AI debate is causing a lot of deep thought around what the purpose of education is and what the value is for somebody coming through a university education. It is really helping to focus attention on the fact that the output from a university education is not the degree that is awarded but the person who emerges at the end, their set of skills and the way that they have developed through those three or four years, or one year if it is postgraduate. That causes a shift in focus towards asking where the added value is and how we are supporting students with that.
It has to be very much rooted in a disciplinary or interdisciplinary knowledge base. We do not want to just teach skills without rooting them in the knowledge associated with the discipline; those two go hand in hand. You will have AI skills coming through, and you will have skills that have variously been described as soft, transferable and so on. They will relate, but you will have to have an education that builds in that mixed economy. Certainly, at King’s for example, we are looking to a future where we embed AI skills training in the disciplines but also root that knowledge base in the analogue skills so that the cognitive process is not lost.
Lewis Wilson: Charlotte touched on it very briefly. It is that game of cat and mouse where universities’ policy is too inflexible and does not keep pace with how rapidly AI is changing, and where the way that students are using it is unfortunately always outpacing lecturers. The real problem is that universities are under-resourced to be able to deliver and prepare students properly for the skills that they need when they enter the workforce because there is just no money in higher education.
Michael Webb: I see a very similar picture, particularly to Simon’s description of the work at King’s, across the sector. When you say, “Are universities doing it?”, it is more complex. The intention across the university sector is very much to prepare students for that AI workplace, and top-level mechanisms to make that happen are in place pretty much universally, but the capacity to deliver on those is much more variable. These tools cost money and staff will be under different pressures in terms of workload and so on, so by the time it reaches students, their experience will be very mixed.
Michael Webb: I will start with Jisc’s part to play in that. Our expertise is around infrastructure, so network connectivity and cyber-security, both of which are hugely impacted by AI. You cannot have escaped seeing all the cyber issues caused by AI in the news. We need to have a collective, sector-level response to support the institutions. Obviously, cutting-edge AI research requires fast, reliable connectivity, which, again, we have in the UK and need to continue to invest in. We have the core. We need to continue investing in it.
Charlotte Armstrong: The main point that I want to make about research is that it is a live discussion in institutions. A lot of the response so far has been so focused on students and responding to assessment and things, so that research has often not been the first point of concern because there has been an element of firefighting as the first response.
On research, the difficulty is that research is frequently very collaborative by its very nature—not only collaborative across institutions, but across nations. When it comes to regulating rules around how much AI use is acceptable—the methods but also the intellectual property and data security around that—it becomes increasingly difficulty the more partners that you have involved.
That is one of the massive stumbling blocks that institutions are facing at the moment when they are deciding how they should react to AI in research.
Charlotte Armstrong: I do not know. It is not something that we have done much investigation into. Professor Banks may be more equipped to answer that.
Peter Swallow: We will go to Professor Banks, then.
Professor Banks: It is discipline-dependent. It is having an enormous impact on research. It is having a potentially very positive impact in terms of providing a way to accelerate some of the ideation process in generating and testing out ideas.
In some areas, it is also just having a very direct application. One can see all sorts of practical benefits in the sciences for bypassing some of the more labour-intensive aspects of the chain that runs through a research project. But it is crucial that that “human in the loop” element is there, and that there is human ownership of decision making, authorship and so on.
The area in which it is causing the most concern at the moment seems to be mathematics, for obvious reasons. Recently, there was the publication of over 700 proofs of mathematical theorems that have been out in the ether for sometimes decades without much progress, or with incremental progress.
An LLM solved or was purported to have solved—because there has been challenge on some of them—such a large number in such a short time. That has caused a great deal of thought and discussion among the mathematical community about what the implications are for what maths research is. That is particularly focusing on questions around the PhD level, which is in the sweet spot.
Of course, that then extends out into the question of which discipline is next. Where is the language of a particular discipline likely to be under threat, as opposed to aided by this?
There is a huge amount going on. If anything, the aspect that we are probably most in need of at the moment is a discussion around what regulation of AI looks like in that context, and what the UK’s role is in terms of how it positions itself relative to that idea of a regulatory framework in relation to research.
Lewis Wilson: I will just say that PGR students sometimes feel the pressure or cannot use AI because of the researcher that they are working most closely with. It is about student choice and allowing them to either use it or opt out of not using it.
Charlotte Armstrong: As was said in the previous session, there are varied levels of enthusiasm and adoption among staff. One recommendation we made, as part of our survey, was the importance of integrating AI into the curriculum. That cannot simply be about teaching students; it also needs to be about equipping staff with the skills to teach students how to use AI. There needs to be a conscious effort to allow staff the time and space to have that training, so that they can assist the students most effectively.
There has been some discussion, slightly adjacent to the education sector, around AI potentially being able to teach students, which concerns me. We recently made a 20-year compilation of our academic experience survey. That dataset contains 206,512 responses from students over 20 years. The most important driver of student experience is relationships with academic staff and the quality of the teaching they receive. The idea that AI might mean that we do not need staff to interact with students as much is incredibly concerning and needs to be pushed back on as hard as possible.
The other thing I want to raise briefly is some research done by Queen Mary University of London, alongside Erasmus University Rotterdam, University of Brazil, Campinas, and McMaster University, Toronto. Medical students were provided with virtual case studies. Some were simply provided with the cases; some were given AI feedback; some were given feedback just from lecturers; and some were given feedback from both AI tools and lecturers.
The students who had received feedback from both human lecturers and AI tools performed best in the assessments. The students who had received feedback simply from AI tools performed the worst, but were most confident in their answers, which is really concerning. That is just one study, but it shows the importance of having a combination, taking and using the benefits of AI but ensuring that human feedback and interaction remain a key part of the process, so that we can be critical of what AI produces.
Professor Banks: Inevitably, staff will have to be upskilled across the board. We are not talking about a world in which only science and engineering graduates need these skills and therefore need to learn them at university; it will be across the disciplines. There will be a change in baseline expectation about awareness, exposure and so on in the skills for staff. Will it change requirements for the number of staff? I would be surprised; I have not seen evidence of that at the moment.
In my experience, staff tend to use AI to do new and different things to help enhance the experience of students, and to provide additional resources and ways of doing something. They are, therefore, effectively taking their time. They might introduce an efficiency over here, and then build on that to do something different and new over here. They have not reduced the overall time. If anything, a lot of us probably spend a bit more time.
An interesting study that came out of Harvard recently looked at using AI tutors as part of a human-led course. That would be humans teaching the course but using a specifically designed AI tutor. The results from that look very promising. There is that partnership, seeing AI as an assistant that can help and accelerate students’ learning. I do think that there is a question around course administration.
To come back to the further education part of this morning’s discussion, I think a conversation is needed about the application to marking, and the way in which trust is going to change as people’s understanding of how the models work, what they do and what their role in marking might look like in future changes. We need to be open to that. A very good point was made about marking being a way to understand your students; that is less relevant if you are marking anonymously, which is the most common way in the university sector. There are ways that we have not yet explored in detail, partly because of lack of trust and partly because of lack of evidence base.
Lewis Wilson: Definitely. I have heard of one horror case, which I will briefly outline because it is so egregious. I was speaking recently to a maths student who had been to a lecture, and the lecturer had made all the slides through Grok’s AI—X, formerly Twitter. At the end of the lecture, all the students were a bit confused and wanted to ask some questions, and the lecturer said, “I’m too busy. Ask the questions to Grok instead.” That speaks to the wider point I made earlier about the lack of resourcing in the higher education sector. It definitely concerns students, especially when frontline staff are being cut from universities.
Another thing that we have touched on briefly was replacement or enhancement with AI tools, rather than frontline members of staff. At the moment I am seeing across the sector the introduction of chatbots so that students have a place to ask questions and all that sort of stuff, whether about university processes or the course. I am really concerned about the legal need around reasonable adjustments. The University of Bristol v. Abrahart case showed that if a student says to any part of the university, “I need a reasonable adjustment—I’m disabled,” the university needs to see that through and make sure that reasonable adjustment is met. What is the obligation on the chatbot to disclose that to the university and make sure that that reasonable adjustment is made? A lot of students are worried about other pitfalls and that they might start slipping through the cracks. It is a really scary prospect.
Michael Webb: I will start with the marking side, because we focused on that a lot last year, exactly for the reasons Simon gave. There are potentially benefits, because it is time-consuming, but also because human marking is not perfect, which we sometimes forget in this debate. It is one of the areas where people fear a loss of their professional identity, as marking is very much at the core of what they do.
Interestingly, we found that, at the moment, the mechanics around checking the stuff outweigh the time benefits for high-stakes stuff anyway, so it became a secondary issue. It would have become a core thing if we had found out that it saved 50% of the time—the loss of professional identity would then be something we need to talk more about—but we actually think it is now really only for lower-stakes stuff.
It is similar to things like the lesson plan debate we heard about earlier. We need to understand that the use of AI is very varied. We tend to think of it as, for example, putting in a prompt and a lesson plan coming out. That would clearly mean you are losing your craft, and people are pushing back on that. Actually, there are all sorts of ways that you can use it. I particularly hate using word processors to format stuff—it takes me ages. I just want my AI tool to collect my thoughts, so I tell it how to do it and hand over the formatting. To me, I am keeping all my craft, and it is taking away the bit that I do not like and that does not add value. The training we do at Jisc for lots of staff says, “Use AI for the bits that you don’t value doing and keep the other stuff for yourself.” We can then keep the craft.
Lewis Wilson: It definitely varies from lecturer to lecturer, from department to department and from course to course. Students doing PPE, which is effectively three different courses, see such variation across different schools in how they are being taught and in what is allowed and is fair usage of AI. It really comes down to the whole academic freedom debate—so many lecturers are anti AI and so many are pro it. It needs to come back to the point that students should be allowed to have the choice of whether they use it, rather than necessarily others.
Michael Webb: I want to pick up on that point as well, because this is one of our latest hot topics. We have monthly practitioner meet-ups, and the right response to people's ethical objections to using AI is a live issue. A staff member might say, “I wish you to use AI in this assignment because you will gain these skills.” What should the response be if students say they have an ethical objection to AI? The same goes for if a staff member is required to deliver it to meet the learning aims, but they object—how do we deal with it? That has added to the complexity of very uneven adoption and skills, so unfortunately this space is getting more complex as the public debate becomes more polarised.
Michael Webb: There are different things that need to happen. First, we all need more training on some of the details so that our objections are well placed and well informed. Some of them come from misinformation, so it would be better if there was more education.
If we put aside the ethical discussion and move on to the fundamentals of how to use it, more and continual training needed is absolutely needed. That is the real problem: AI is moving so fast that the training you received last week might not be so useful this week. It is about investment in the initial training and then keeping everyone’s skills up to date so they can continually apply it in whatever they do.
Charlotte Armstrong: I agree with everything that has been said. I want to reiterate the last two points. Staff and students who do not want to use AI are currently getting lost slightly in the discussion around how it is being used. We have seen a rapid rise in the number of people who are concerned about AI use, particularly for environmental reasons, over the past three years—that has completely shot up in our responses. I would imagine—we do not know yet; our next survey is about to go out into the field—that that will continue to rise with. My main things to highlight are the need to think about those who disagree, our response to that, and the importance of training. I really agree with what was just said.
Professor Banks: It is very difficult to quantify the level of variation. It seems clear that there is a great deal of variation across institutions and across subject areas, but actually putting numbers on that would be a challenge.
On the training aspect, there are some really good training tools out there already—Jisc, for example, has recently published some, and there are others. Sometimes there might be a lack of awareness of what is available. Sometimes there is a sense of not being able to see the wood for the trees, in that there is so much. There are so many guidance documents and training tools, and they could be brought down, for a practitioner, to the level of somebody who wants just to upskill and get a basic fundamental level of knowledge to be able to move forward and do their job.
The other thing is the slight fear of missing out. Do I understand this big scary world of AI and what it means for me? Is it easier to set it to one side, focus on the known, and hope that this will clarify over time and I will join it then? For both staff and students there can be a sense of feeling, “I ought to be doing something with this. There’s a real opportunity over here, but I don’t quite see my way in.”
Professor Banks: Universities have a crucial role in certifying the students who leave them, but it is important to remember that we are certifying the human being. Ideally, you want to be able to say, “This person has this level of knowledge, this set of skills,” and so on, and you want to be able to do that with confidence and to be able to stand behind that certification.
I think there is a wide and growing acceptance of the fact that some of the traditional assessment practices in higher education are seriously threatened by the realities of what generative AI is now able to do. That is necessarily going to cause a really significant rethink of the way we assess. It is already happening—those discussions are already under way—and some of the changes are moving forward.
For my part, it comes down to a question of the balance of risk. There has always been some risk associated with remote assessments, but the ability to cheat—for want of a better word—the integrity of those assessments and the balance of their weight overall has been managed and acceptable. I think that balance is shifting. We are going to have to bring a lot more assessment directly back in, and we need to do that in a way that does not lose the genuine benefits of those remote assessments—the authenticity, the skills and so on. It is going to be a very mixed portfolio, in which there will absolutely be a place for examinations, for orals and for practical classes. It will be a suite.
The other aspect to highlight is that that has to be done in a way that designs in the inclusivity from the start. You cannot retrofit that afterwards. You also cannot redesign assessment without redesigning your teaching to go along with it. The two are not separate; they are inherently linked. You have to think about it end to end: if we are assessing differently over here, how are we teaching differently?
Lewis Wilson: On academic integrity and assessment, I want to touch on how academic misconduct is being flagged somewhat disproportionately for international students who do not necessarily have English as their first language. They sometimes therefore use translation tools, or even tools that are recommended by the Government, like Grammarly, which integrate AI without them necessarily realising it, so they are being flagged unfairly. They also perhaps use practice essays to model their essays, without realising that the sentence starters or the structure mirrors an AI’s way of speaking, meaning they get flagged for AI. I think AI statistics are definitely overinflated, particularly for international students.
I do agree with Simon about needing to develop and change, but I do not think the answer is necessarily more technology to try to catch out people who are misusing AI. When I was a student, something that I saw at my institution, the University of Sussex, was the remote proctoring of exams. An AI camera is filming you the entire time and monitoring your movements, what screen you are looking at and all sorts of different stuff. The concern around your personal data being sent off to one of these big American firms is a different concern, but that can affect disabled students who maybe have tics, and it can lead to the false flagging of misconduct.
It comes back to the wider issue of assessment design needing to be collaborative, and speaking not only to employers but to students to make sure that assessments meet the workforce needs and look like students want them to look like. Students want to learn, they want to do well, they want to get firsts or 2:1s and do well in their assessments, and they would like to have a say in how assessments are designed so that they can get the best out of their education.
Charlotte Armstrong: We found that 65% of students say that the way that their institution has assessed them has changed in response to AI, so we are definitely seeing shifts in institutions. Anecdotally, we have some students who said, in terms of the qualitative responses to that survey, that they felt like assessment had been made harder in response. However, I would caution you that this was quite a small number in comparison to the survey and obviously we do not know—they are judging from their own personal opinion and view on how they found the assessment, so I would not say that conclusively.
I think the main thing to think about with assessments here is the fact that employers want students to be able to use AI. AI will be a key part of much of the workforce; students recognise that. So, when we are considering the role of institutions and higher education in preparing students for employment in the future, being able to say to an employer, “Actually, my ability to use AI skills is certified by my institution,” is something that could be really valuable for those students.
The line that we have taken in some of our recommendations from our report is the importance of thinking about what an assessment is for—what are you trying to find out that the student can do? And if you are not allowing a student to use AI in an assessment, it should be physically impossible for them to use AI in the assessment. If it is possible for a student to use AI in the assessment, then why not say that AI is allowed and have that as part of the assessment criteria and be aware of that? Particularly if you have integrated a degree of training and teaching as part of the course and particularly subject-dependent AI skills as part of the course, that is something you can assess as part of your coursework and as part of an education institution.
It is about just stepping back and thinking, constructing AI-aware assessments, and factoring in whether and how AI impacts what you are trying to find out about what the student understands.
Michael Webb: I agree with that; I will just add some detailed points. First of all, on the technology landscape I think we have to remember that it is changing rapidly again and—unfortunately—that is going to make this whole area even more complex. We have agentic AI now, which can control your computer and use all the applications that you might normally use. It is no longer just about writing essays. We have wearables coming along very quickly, which are going to make supervised assessment a lot more challenging. With glasses today, we can spot them; the next generations will not be able to, so it is going to get more complex.
In our armoury, broadly we have supervised education, which we talked about and which today is safe; it is going to be challenging tomorrow, so let’s not put all our eggs in those baskets. We have then got a mixture of advice and spotting, which is really the predominant method at the moment. It is an interesting one. This is where the students are told how to use AI and the people marking, either with technology or through some guidance, try and make judgments as to whether there is cheating or not. This is probably the most challenging area, to be perfectly honest. There is a lot of pushback against AI text detection tools. The problem is that people are even worse at spotting it than AI. That one needs to move away.
The emerging technical solutions now are, first, around process monitoring. You write your essay in a particular tool that records how you do it and creates an evidence trail, or just by saying to students, “Look, show me your workings, basically, as you go.” And then there are knowledge checks at the end; oral exams and we are increasingly seeing technology come in, whereby you submit your essay and then you will be asked a few questions. You have to record your spontaneous responses, which, if it wasn’t your work, is quite hard. There are a range of things coming along in this space.
Professor Banks: The go-to answer for that is usually anything involving writing computer code at the moment. LLMs tend to do an extremely good job of it. I have spent a lot of my time over the years writing code, and I now use AI for vibe coding. It does a fantastic job, so we really have to rethink there.
At the other end of the disciplinary spectrum, potentially—not my area, so others may have more expert views—there is the essay simply as a method of assessment. I think there are mixed views on where we stand with the value and the integrity of the essay. What seems clear is that it is an evolving picture, and LLMs are only going to get better as we go.
The last thing to say is on dissertations. You can add an oral component to a dissertation. You can also change the way in which you supervise that research project, and build in discussion, reflection and so on as you go. There are checks and balances there, but fundamentally the issue is about how well the LLM or generative AI can tackle a particular discipline at the moment.
Michael Webb: The research, even a year or so ago, showed that AI could pass pretty much all subjects where you just had to submit written work, and things have moved on quickly. Anything where you give students advice and then rely on somebody checking that they have followed it is complicated at the moment.
Michael Webb: What is different about AI from other things? We have always given fairly broad leeway to institutions to set their academic regulations as they see fit, so long as they fit with things like the OfS guidance, and I do not think AI is any different. I think anything we attempt to dictate across the board will be either so general that it is meaningless or so specific that it starts affecting how individual courses can be delivered.
Charlotte Armstrong: It is important that institutions retain their independence through any process. There are a number of institutions explicitly saying that they want guidance; they are looking for somebody to tell them what to do. If you ask them who that should be, you get a lot of different answers. Some people think it should be mission groups. Some would point more towards OfS and some point towards DFE, so there is a lot of disagreement, but the idea of guidance is welcomed by a lot of parts of the sector. Because of the speed with which things change, as we have already discussed, anything that was implemented, at least from my perspective should be around values and principles, rather than an explicit, “You must do X; you must not do Y.”
A really interesting example of that is what the Higher Education Authority in Ireland has been doing recently. It has produced a document that is value and principles-led on the ethical concerns and the importance of ensuring that there are not EDI implications around AI adoption. I would push something in that direction, rather than say that the Government should come in and very explicitly say, “Institutions must do this.” That would clash hard against that independence, which is so important.
Professor Banks: I would largely echo everything that Charlotte has just said. I very briefly add that there are institutions, such as King’s where we have things such as the AI Institute. We have people whose day job is to develop AI and AI tools and think very deeply about its role in society and education, and so on. There will be institutions where that expertise may not be so prevalent. Across the sector as a whole, there is clearly some value in having guidance and principles-led frameworks for helping institutions to develop their own policies, rather than some central and regulatory framework that sits on top and imposes things on universities.
Lewis Wilson: I would largely agree with what has been said. If there is a national framework, it needs to come with proper resourcing to make sure that no students are left behind.
Chair: That brings us more or less to the end of our session. If there are points of detail or further thoughts that you have not able to get across to us today, please write to us; we would welcome that very much. Thank you for taking the time to be with us this morning.