As part of our training in Artificial Intelligence, a number of staff in the Faculty of Creative Arts and Humanities at Liverpool Hope University were given the opportunity to undertake the King’s College, London, FutureLearn course on AI in Education. It’s a subject about which I have very mixed feelings, so here are my reflections on the third week of the course.

Week three of the course moved on to think about some of the things that have been bothering me right from the word go – ethics, sustainability and bias. After an interesting overview video, I had to watch an AI voiced video on AI and copyright, which is something I’m really interested in, but the AI voice was really difficult to listen to. I didn’t do too badly on the BBC’s AI or genuine Picasso quiz that I completed when we were looking at issues around AI and copyright. 7/10 is apparently a ‘strong’ score, and the activity did help to highlight some of the issues about where creativity lies and whether it infringes copyright if you can create something apparently original, based on a prompt that draws so heavily on other people’s work, especially considering how much data is itself unethically sourced.

You might expect me to have strong feelings about AI and creativity. I am a trained singer, married to a trained composer, with three incredibly creative children who have all chosen to study arts based courses at university – music, set design and wildlife media. Quite apart from the ethics of these models basing their output on many millions of data points which have in many cases been stolen from the original creators, there is the question of what creativity is.

The OED defines the verb create as ‘To bring into being, cause to exist; esp. to produce where nothing was before.’ On that basis, I guess you can accept that GenAI is creative, and so are the people that use it to create things. But I think most of us would assume creativity to be more than just an AI prompt. It’s not just about creating things. Many of us could attempt to write a song or a poem, or paint an artwork, but it doesn’t necessarily mean that the resulting creations would be particularly good, and it tends, I think, to be by that sort of measure that we decide whether or not someone is creative – do they have the skill or the flair to touch us in some way with what they create? The OED defines art as ‘The expression or application of creative skill and imagination, typically in a visual form such as painting, drawing, or sculpture, producing works to be appreciated primarily for their beauty or emotional power’. Can a large language model generate emotion? Or do you need to be human to do that? It certainly can’t create from a genuine emotion. And we’d like to think that we can tell when something comes from the heart. I suspect that the reality is, we can’t.

Early on in this course, I was discussing these matters with my daughter, who has extremely strong opinions on GenAI of all sorts, not just in relation to art. When I completed the activity on AI songwriting, she sent me these links to instagram posts in which Alex Harvey sings about what human music does that AI music doesn’t. And if, as the FutureLearn course suggests, students feel cheated when they discover that their educators haven’t read and graded their assignments themselves but have asked AI to do so (in carefully curated chatbots that are locally controlled so that student data isn’t put at risk, I hasten to add), then many of us, I think, still think that ‘art’ should be fundamentally human.

But then, there is the other side of the argument. While looking into AI for this course, I remembered that there is a voice on this that I trust – and I trust it because I know it from way back when. I’m happy to say that I knew Jon Whittle long before he was an internationally renowned expert on AI, and back in the days when he bemoaned the fact that he appeared to be destined forever to play the part of the slightly wimpy character (my how things change!) in whatever performance he was in – he was Buttons to my Cinderella. And in his substack, Jon has some very different things to say about creativity, AI and creatives’ backlash to its use. And I have to respect his position – it’s not as if he isn’t someone with a healthy interest in and respect for the arts. I sometimes think it is the bane of my life that I can see and understand so many different points of view… it would be so much easier just to disagree completely, but maybe he has a point.

I also found the course’s advice about AI hallucination slightly less than helpful. The course acknowledges it exists, but rather downplays it. It is becoming less common, it is true, but it’s not enough just to say that we should fact check important information. Not everyone has the skills to realise what they need to check. And the models present everything with an air of authority. As an example, I recently asked ChatGPT about recent trends in the historiography of music during the English Reformation. It’s response was to give me a run down of major debates since the beginning of the twentieth century. When I asked for names to go with the arguments, it listed Whigs and revisionists, and finally a very short list of names from the 2010s to 2020s with no detail. There was a fundamental error in my prompt – ‘recent’ not ‘current’ – and when I challenged it on the fact that there was nothing from the last 15 years, it pointed out my mistake, then identified some very specific work from individual people that wasn’t really allied to wider trends in the scholarship. But the thing is, I have enough expert knowledge to realise that. Many of my students do not.

This is what it came up with for readers of my blog:

A busy Renaissance village square during a carnival, crowded with dozens of figures engaged in everyday life and festivity. In the foreground, several disabled beggars wearing white cloaks move on crutches or crawl on the ground, while others sit in carts or are assisted by companions. A small white-and-brown dog stands near two pecking birds. To the left, a man rides a brown horse through the crowd. Across the centre, villagers gather around tables, carry baskets, converse, and watch the festivities. On the right, people lean from a tavern window while others buy food from a market stall. Stone houses, towers, and narrow streets recede into the background, where more crowds continue the celebration. The muted earth tones and densely packed figures create a vivid contrast between carnival merriment and the presence of poverty and disability, highlighting the social diversity of the scene.

And this is what it came up with for a powerpoint for my first year undergraduates:

Sixteenth-century village carnival scene showing a crowded marketplace where disabled beggars and townspeople mingle. Several beggars using crutches, wheelboards, or carts occupy the foreground, while villagers eat, trade, dance, and socialise around them. The painting juxtaposes festive celebration with poverty and disability, illustrating how disabled people were visible participants in public life and often associated with begging during the early modern period.

It actually told me that the alt-text for the powerpoint needed to ‘prioritise the information students need to understand the slide rather than describing every visible detail.’ The problem is, I might not be using it to teach about poverty and disability. Spoiler alert, I don’t. I use it to teach about popular culture. Again, this could be refined by me giving a more tightly focussed prompt, but the fact is that on the surface this seems a plausible enough interpretation… It really highlights the need for high levels of AI literacy if we are going to use these tools.

Then there’s the issue of sustainability, which I’ve already talked about back in week 1. All of this leaves me feeling deeply uncomfortable about incorporating AI into my teaching on a regular basis. I can see big advantages in improving accessibility for students with additional needs, where it is essentially being used to level the playing field. But as a regular part of teaching, where students are being encouraged to use it as part of what they do to research and write history? I think that here, a lot of the benefits are outweighed by the risks that students don’t have the expert knowledge they need to be able to tell when AI is producing poor quality outcomes, and that they will fail to develop the high level cognitive skills that are produced by reading and writing for yourself.

The final activity was based around AI and the future of work. One thing that the course suggested was that ‘To assess how AI shapes an occupation, it is helpful to think of AI systems as task centres and then ask which tasks of a particular occupation can be automated with the current technologies’. But there is a lot more to it than this. This Adapt Insider podcast, again featuring Jon (largely because he’s someone I trust not to just say what he thinks people want to hear), raises a lot of interesting points about what the benefits might be to business in using AI. One of the big things he points out is that productivity – or saving people time – shouldn’t be the main driver of adopting AI, because just because you save an employee a certain amount of time by using AI doesn’t mean that the time saved is well spent, or even if it is, whether it might even lead to burnout. Instead, he advises thinking about what the purpose of your business is and how AI might advance that purpose. It’s less for efficiency than effectiveness.

There are so many other issues that I have barely touched upon here: the fact that training models are largely based in US English sources; that poor quality outcomes from AI feed back into the pool of ‘knowledge’ that is being used to create future responses; that the LLMs embed western colonial mindsets in their responses; heck, even things as simple as the fact that if you ask AI to come up with an image of a doctor and a nurse, they will both be white, but the nurse will be female and the doctor will be male. This stuff is embedded, and attempts to override it can produce utterly bizarre results. So while I’ve found the course interesting, and it’s given me plenty to reflect on, it’s left me not much further on than I was when I started. It’s provided one or two ideas about where AI might be a help in my academic life, but left me still unable to decide whether it’s really worth all the hype, or whether the challenges it poses to education are greater than the possible gains.

One thing it really has brought home to me, though. I am not an expert. There is a lot here that I don’t really understand, and I can only keep learning and reflecting in the hope that I make the best choices I can for me and my students…