Elsevier Medical Education presents an interview with Pippa Furey, Senior Lecturer in Paramedic Practice & Emergency Care at Nottingham Trent University and National AI Lead (Education) at the Royal College of Paramedics, on how educators can integrate AI responsibly in emergency healthcare education.
From your experience as an educator, what has changed most in the way paramedicine is taught since the introduction of generative AI?
I think that what’s changed most is not what we teach, but how openly we talk about knowledge itself and organic learning. Since generative AI became more widely available, we’ve had to explicitly address some concepts that have previously sat beneath the surface in education that we’ve known about and alluded to, but we maybe haven’t had as explicit topic types that we’ve needed to engage with so willingly. So, things like cognitive offloading, bias, and what counts for that organic learning and understanding, and how do we guarantee that people know what they actually know?
If I think about it in paramedic terms, a useful analogy from my training was about access to automated equipment. Let’s take a blood pressure measurement, for example. We were taught how to do it manually and we were also taught how to do it with an automated blood pressure machine. But we didn’t stop teaching manual blood pressures when the technology came out because you still need to understand the underlying physiology, and what’s happening and how to interpret the results. Also, if your equipment doesn’t work or you’ve got a patient who shouldn’t have an automatic blood pressure taken, you still need that knowledge there. With the automated blood pressure, we were redistributing the cognitive effort. AI is similar – it can generate information quickly, but if you haven’t got that foundational knowledge, then the output is just text on the screen and what you make of that is going to be very different. So, education has shifted from simply presenting information to helping students interrogate it. I think now we’re really asking, where does this come from? Can I trust it? What does it mean clinically? And that shift feels really fundamental.
"AI can produce plausible responses instantly, so assessment has to move upstream, I would say, toward that reasoning and judgment justification of what you might want to know."
Pippa Furey, Senior Lecturer, Nottingham Trent University | Elsevier Faculty Hub, June 2026
How can educators effectively foster critical thinking in students who now have access to tools capable of generating complex answers instantly?
We need to redesign tasks so that success is no longer just arriving at the correct answer but also demonstrating how we know that answer is correct. AI can produce plausible responses instantly, so assessment has to move upstream, I would say, toward that reasoning and judgment justification of what you might want to know. That might mean asking students to critique an AI-generated answer, compare sources, or explain their clinical decision-making step-by-step. In essence, we want to move away from that "Do you know the answer?" to "Can you evaluate whether an answer you know is safe and appropriate?", which aligns much more closely with, for me, real-world paramedic practice. Also, being able to justify your decisions and not just having the decision as the output and the sole thing that we’re relying on is absolutely something that we should be moving towards with this critical thinking. So, if anything, it gives us the opportunity to really deepen that natural understanding and really interrogate whether knowledge is where it needs to be. But we’ve got to build that in. We can’t expect on assessment day that students can do that if we haven’t set tasks throughout the academic year and educated them on what we need them to know and how we expect them to know something.
Given your role in curriculum development, what concrete changes should be introduced today to integrate AI in a meaningful and sustainable way?
Curriculum development is hard because institutions have got their own timelines and ideas and therefore, there are goalposts and deadlines that are hard to move to. For example, in my experience, we have to make academic changes for the next year by January of the previous academic year, and a lot can change in the AI world in that time. So, I think it’s about working out what you can change and where you can change it. You might not be able to make a big change within your curriculum assessment design, but you can make a smaller change to your delivered content, and that content can be bite-size and drip fed and spiraled throughout the degree. It’s all about embedding that digital and AI literacy from the beginning and that’s something that we sometimes feel is assumed in our learners. That they turn up and they have a phone and a laptop and they know how to use them. But that can be completely incorrect, and it overlooks digital exclusion and varying levels of confidence. So, you’re not only learning new information in the classroom about your subject matter, but then you’re being expected to suddenly learn new information about how a technology works. If we’re asking you to take part in quizzes online or a voting scale or a word cloud, and if you’ve never used that technology before and you don’t understand how it works, then we haven’t educated you in that either.
Integration needs to be intentional, but also not everywhere. We don’t need every task to be AI-driven. I think it needs to be appropriate for the job that we’re doing, for the tasks that we need. Also, thinking about the environmental impact if we put an AI task into every module, into every lecture, then actually, what knock-on effects we’re having on the planet is a big consideration. Paramedicine, for me, is inherently an analogue profession in most contexts. It’s the back of an ambulance, it’s a patient’s home, and it’s human connection and conversation. So, we need to be thinking about where AI enhances that and where does it break down some of the real strengths within the profession. We must avoid producing a workforce that’s overly dependent on digital support. Therefore, if I was to summarize, I think it’s early structured exposure to AI literacy and digital literacy, clear guidance on when and how to use the tools, and maintaining some tech-free spaces so that we have analog tasks. Balance is key.
Your research highlights important ethical challenges around AI. How should these dilemmas be addressed within healthcare education programmes?
Yes, the ethical dilemmas are really prevalent, but I wouldn’t want to write about AI without talking about them. I also appreciate that the ethical dilemmas have remained consistent whilst the user interfaces of AI have been updated and changed over the past couple of years. Therefore, when we talk about progression with AI, we talk a lot about what it can do and what it can’t do, but we don’t talk about the changes in these ethical dilemmas because they’re pretty constant. So, the risk of bias, over-reliance, hallucination, just to name a few. AI systems are built by humans and humans are biased in who they are and who they become. Bias is unavoidable and I think it would be incorrect to suggest that we could build a system by humans that wasn’t biased. We all come from different backgrounds, and we all have our own lived experience that’s going to make our narrative. That’s absolutely the same for these data sets and the people that are deciding what it’s trained upon. So, the responsibility is really for the educators to bring that bias to the forefront, encouraging students to question their own biases, their own perspectives and their own lived experiences – who’s represented and who isn’t? And then, when we think about AI systems – who’s represented in that data and who isn’t? When we’re discussing using a large language model, is it a westernized-centric English-speaking model? Has it been trained on mainly data from the internet? Who is going to be missed by that data? One practical approach, I think, is about positionality statements. We see those in research where people talk openly about their own lived experience. We can ask students to reflect on their own standpoint there and extend their thinking on what would an AI’s positionality statement look like.
There are also broader concerns, like I mentioned, with over-reliance, environmental impact and also unequal access. I think it’s important to occasionally step away from technology entirely – and how uncomfortable does it feel? When we think about over-reliance, most of my teaching slides are on PowerPoint or a similar system and I feel very comfortable if I’ve got them there to remind me of what I’m going to say. If we now take that away, and if I was asked to lecture in front of a group of 200 people without any lecture slides, I think I’d find it quite uncomfortable. So, if we’ve got students that are constantly using AI and then suddenly, we take that AI away and ask them to do a task, they’re going to feel uncomfortable as well. There’s constant AI integration into everyday user interfaces. And yet, we’re going to maybe enhance this even more, whereas I think we need to have that balance and think about our own over-reliance and our own positions within the ethical dilemmas.
"I’ve seen some really great suggested uses by students. The problem is we need to foster an environment where they feel comfortable to talk about how they’re using AI and also educate use versus misuse, and that’s really difficult if it seems like something that people shouldn’t be using."
Pippa Furey, Senior Lecturer, Nottingham Trent University | Elsevier Faculty Hub, June 2026
Beyond the current enthusiasm surrounding AI, what are the real pedagogical implications of generative AI in healthcare education? What is genuinely working, and what is not?
Genuinely working, I think, is AI as a thinking partner and not a shortcut. When we’ve got students that are using it, not in any way as a supervisor or to do the work for them, it’s about what they are getting from it. I’ve seen some really great suggested uses by students. The problem is we need to foster an environment where they feel comfortable to talk about how they’re using AI and also educate use versus misuse, and that’s really difficult if it seems like something that people shouldn’t be using. And there’s a lot of shadow AI use, where people are using AI without saying, "Actually, I formulated this document with AI," or "My slides were enhanced by AI," or "Actually, I asked AI for some idea creation." If we’re not open about it, students won’t be open back because a lot of the narrative is around academic integrity and the risk to cheating and creating work that’s not their own.
Particularly, I think AI is effective for prompting that idea generation to get the ball rolling. We’re an increasingly neurodivergent diagnosed population and therefore, we’ve got much more of an understanding now about how people’s brains are working and how much of the population is affected. In my own student population, we have a significant student body with ADHD and when I speak to those students, they talk about how they’ve got this list of tasks to do and they are essentially just so stuck with how to begin, because it seems like a never-ending list of tasks that are an end-product. So, it’s something like ‘create a dissertation’, and it’s like, well, how do you create a dissertation? Whilst, yes, we have lecture content, seminars, and workshops to teach them how to create a dissertation, but actually, what does that mean for them and how do they make it into their own language? AI can massively help with getting the ball rolling. It shouldn’t create their content for them, but it absolutely can give them ideas on how to begin something and how to start when they don’t know how to begin.
I think it’s really good for feedback and informative learning. For example, if you’re about to have a multiple choice quiz and you want to test your own knowledge using AI. However, you need to verify those sources because if you tell it an answer and you say, "I think this is correct," it will possibly tell you it is correct when it’s not. It’s one thing to say, "Create a multiple-choice questionnaire that I can fill in." It’s another thing to then verify what the AI thinks is the correct answer and create learning there.
Then, I think as well, helping students explore those multiple perspectives and finding voices that aren’t their own and getting out of the echo chamber. We all, in life, surround ourselves by people that are like us and reflect our values and actually, AI gives us the opportunity to find some different voices in that. But we also need to be aware about the things I mentioned before about the bias and whose voices are not being heard.
Where it doesn’t work, I think, is replacing that thinking element. If students are using AI to generate answers without engaging, then learning becomes really superficial. Therefore, if we’re going to create a task that’s AI-enabled, we need to make it really clear to the student that it’s not just about generating an output – it’s about what that output actually means. So, pedagogically, the implication is that AI amplifies the existing teaching design. It can make teaching more accessible to students that struggle to engage. But strong pedagogy will become stronger with it, but weak pedagogy becomes more visible, I think, and we really can widen that gap.
You have explored the role of AI in digital inequality. What responsibilities do educators have to ensure equitable access and use of these technologies in the classroom?
Educators can’t solve structural inequity and inequality alone. That’s system wide. But they can make access and expectations more transparent. Universities need to support widening participation, and we see initiatives within that already. At my university, we have something called Success for All. So, that’s where we talk about people’s different backgrounds and their educational background, but also their socioeconomic background and how this is going to impact their education and how do we create an equitable space for all parties. It now needs to be extended to digital access and capability. One thing is digital access and whether you’ve got a device that can log on to the internet, and what your internet speed is. Then the other thing is about capability. You might have the best Wi-Fi in the world and the best new device in the world, but actually, do you have the digital literacy capability to be able to make the most of it? That’s creating a whole new area of inequity.
Educators should clearly map out what tools will be used in their sessions and when. I think it’s got to be really clear and spoken to students that, "Okay, in the next couple of weeks, we’ve got X amount of lectures and on the fourth lecture, we’re expecting to use this technology." So, we’re giving people a heads up that this is coming and saying, "Talk to me," or "Here’s the resource you may need to access before you’re expected to know this technology inside and out." Build time into teaching to teach them those tools. An example is, in our dissertation module we talk about a tool for screening research for your literature review. So, you’ve done your Boolean search terms, you’ve got your papers, and you need to now screen them with your inclusion exclusion criteria. We use a tool that keeps those citations and stores which ones you’ve accepted and which ones you haven’t. We spend two hours teaching the students how to use that tool because we know that it’s going to massively enhance their dissertation projects, their capability, their capacity, and the cognitive load. But also, it prepares them for the future if they want to do a project like this again so they know that they’ve got those tools and they can absolutely use them.
I think we also need to avoid assuming familiarity and fairness amongst our student cohort and not take the student we see in front of us at face value. We need to really think about who they are, what they’re capable of and ask them. Then, once they’re used to using those tools, we can use them to enhance learning. It’s not the tool itself that creates the exclusion. I think it’s the lack of structured introduction and ability to access the tools fairly.
AI “hallucinations” present clear risks. How should students be trained to identify and manage these inaccuracies?
Firstly, I tend to try and acknowledge from my medical background that "hallucination" is a really imperfect term. It humanizes a process that’s fundamentally statistical, but it’s useful for shorthand. Yet, it doesn’t fit the model and the robot element of large language models. But students need two things – they need that awareness and a method. Awareness that AI outputs can be confidently incorrect. I think it really shines a light on something that we probably needed to reorientate ourselves to, if I can say it like that. I remember when I started learning at university, we talked about how to know whether the academic paper you’re reading is unbiased or if it’s relevant and that we must look at recent publications. The problem with AI is a lot of our publications about it comes from blog posts in the media before we get any kind of big data that we can really trust in the conventional ways. And that’s quite difficult because then, which newspaper do you trust and which blog post is reliable?
I think we need to think about something I heard a couple of years ago being referred to as ‘BREAD’. So that’s Bias, the Relevance of a project, the Evidence involved in it, who’s the Author, and what Date was it published. We’ve always needed to think about the motivator of publication, who’s benefiting from putting these things out there and who is not. It’s not new. It’s something we’ve always done with sources, but it translates really well to AI outputs and it’s really great for starting that conversation.
I think we’ve really put on the forefront this notion of, don’t believe everything you read, don’t believe everything everyone tells you. I, as a lecturer, have a power within the room to tell people information and for them to feel like they should believe me, but what’s my background? How do they know that what I know is true? Therefore, sometimes when I’m teaching, I’ll give a bit of a history of who I am, my background, and why I know something to be true, without giving my CV. I don’t think we need to be doing that but I do think we need to be showing our own credibility, probably more so now that students are so willing to question it.
Looking ahead, what are the three key competencies that healthcare educators need to develop in order to remain effective in an AI-driven environment?
I’m all for the prevalent uses of AI and the positive inquiry versions of AI so, this is really something I enjoy talking about. For number one, it’s got to be about conceptual understanding over tool familiarity. Educators say to me, "I’m worried about not being able to keep up," and I put my hands up, I would not say that I’m up to date with how every tool works, how every tool user interface is functioning currently, and the new releases. But I know what’s going on in general, and I know what kind of things are out there. I think that worry of keeping up with every single platform makes it that you will never be able to keep up and it will feel like such a huge task that’s going to add risk to your own knowledge. So, don’t worry about keeping up with every new platform. The underlying issues of the bias, the reliability, and the over-reliance, they’re all remaining consistent regardless of which AI model you want to use, which user interface is stipulated by your employer and also, what the task is. Whatever task you’re trying to achieve, you shouldn’t just use one AI model to achieve them all, because some will be better at certain things than others. If we take a public model like ChatGPT, for example – it’s not created for research. So, if there’s an AI model out there that is created for research and it is in your methodology and in your ethics form, then absolutely, that’s the right tool for you to be using.
Number two, I would say, is to think about critical digital pedagogy – the ability to design learning that integrates AI meaningfully rather than reactively. We shouldn’t feel like we need to add a tick box to say, "Yes, we’re using AI in our programs and therefore, today, we’re going to do this task," which isn’t meaningfully embedded in some kind of way. So, really working out what would be enhanced by AI, what wouldn’t be enhanced by AI, and where we can create outputs that are genuine for the learner. For example, instead of providing a case study for your assessment, can the students generate a case study and then critique how correct the case study is? So, the assessment being their critique of the case study instead.
And I think finally, just confidence to engage and experiment. There’s no need to keep up, and that can become an absolute barrier. You just need to start, test, adapt, and see what works. An AI output should probably never be just copied and pasted and left as it is. It should absolutely be engaged with and refined. And when you finally get the output you want, ask AI what you should have used as your initial prompt, and then you’ll educate yourself on prompt generation.
Leave a Reply