In this episode, we're going to be asking whether our current AI systems are capable of creating something entirely original. The AIcademia podcast is a weekly show helping educators like you leverage AI in your everyday practice. I'm your host, Andy Fisher, and thanks for joining me.

Some of my happiest memories of my own days at school were double periods spent in the art studio as part of my A level studies. While I wouldn’t describe myself as a good artist, I’ve kept my hand in and often sketch, and I’ve worked semi-professionally as a photographer which formed part of my degree course back in the mid 90s. I write poetry, I’ve dabbled in printmaking and have been known to sing in the shower. The point being that I would like to think of myself as a creative person and I’m at my most engaged when I’m innovating.

So when I started experimenting with generative AI a couple of years ago now, I had really mixed feelings about the process. On the one hand, it was exciting to see that in just a few seconds I could create something that was superior to anything I had the technical ability to generate. But on the other hand, it felt somewhat like I was cheating and I didn't have the same degree of investment in the output as I did with something I've crafted with my own hands. Perhaps you feel the same?

For a while, I've been wanting to put a podcast episode together exploring this question around artificial intelligence and creativity. It’s such a huge topic that I struggled for a while to find a shape and direction but I’ve landed on four topics I'd like to focus on today. 

The first is the question of what we mean by creativity? Can a machine be truly creative? 

The second is to try and determine if there is a difference between human creativity and the potential for machines to create? 

Thirdly, we're going to be exploring the question of whether AI threatens to displace human creativity? 

And then finally, I'm going to end by looking at some practical ways that we teachers and our students can harness AI to boost our creativity in our classrooms and lessons.


So let's begin by looking at what we mean by creativity. I’ve read books and blogposts about creativity for a long time now. I’m interested in creative breakthroughs, ways in which we can create the conditions for our creative to flow and how we can overcome the barriers to having new and original ideas. What I keep coming across again and again is that creativity seems to be the ability to generate something which is new, but also something which is useful. 

It's a combination of innovation and utility. The sources I’ve read tend to present creativity as a process of breaking free from conventional patterns to arrive at unexpected connections or solutions. It’s often described as the drive to explore the unknown and to transcend barriers. 

It’s a 10 letter word with at least as many different definitions and that is part of the challenge of determining if an AI system is capable of meeting these same benchmarks for being creative.

As I’ve read more widely around the strategies employed by the most creative people in society to generate original ideas at will, two distinct patterns or workflows emerge. 

One approach says that to be creative, you just need to put the work in, you need to prime the pump. To have a good idea, you need to have lots of ideas. You have to be fearless in iterating again and again, and then be ruthless in throwing out all but the best ideas that emerge. This is very similar to the process of brute forcing a solution using AI systems. 

One of the breakthrough moments in AI advancement came in 1997 when Gary Kasparov, chess grandmaster, lost to IBM's Deep Blue, a chess engine with access to huge data sets which was able to analyse thousands of positions before selecting its next move. It was able to brute force its way to a win by doing what great chess players do (that is, thinking multiple moves ahead based on pre-ingested board positions) and it did it better. 

Similarly, the Nobel prize for Chemistry last year was awarded to the developers of AlphaFold AI which is a 3D protein folding simulator which has the potential to revolutionise new drug discovery. By predicting the shape of protein structures and their potential binding sites, it can anticipate candidates deserving of further medical research. In fact the first new class of antibiotics to come to market in over 60 years which is effective against MRSA was discovered using AI systems. They are so successful because these models can run millions of simulations in the time it would take a human researcher to test one viable candidate in a laboratory and it can do so at a fraction of the cost. We can say, then, definitively, that artificial intelligence has created a new medicine with real world value - but is this all there is to the mysteries of creativity?

What about that other process by which humans create - this is not the result of working your craft day in day out but involves those strange Intuitive leaps, those moments of insight, those eureka experiences which leave our best neuroscientists baffled. We suspect that it involves subconscious processing of some kind but our experience is that a solution to a problem pops into our imagination, fully formed. We might be out for a brisk walk, soaking in a bath or doodling in a margin when boom - the muse lands in our lap and were gifted with a sparkling brand new idea.

There are lots of examples in history of this kind of creative breakthrough - we can even create in our sleep. Elias Howe's invention of the sewing machine in the mid-1800s came from a dream where he was being chased by natives who had holes in the tips of their spear. This led him to understand that the thread could be introduced into the tip, rather than the head of the needle. 

Similarly, Auguste Kekulé in 1861 was struggling to try and understand the structure of benzene when he fell asleep and dreamed of a snake biting its own tail and realised on waking that it was a ring structure.

Surely these kinds of moments are uniquely human? Perhaps even divinely gifted. After all the root of the word inspiration - spiratus -means ‘to be breathed into’ - to be filled with life and ideas by the gods.

Well perhaps not! We only need to look as far as another game where a machine beat a human being to challenge this paradigm. In 2016 AlphaGo, born from the same Google lab which would produce AlphaFold a few years later, beat grandmaster Lee Sedol with its famous move 37. 

If you're not familiar with Go, it's sometimes referred to as Chinese checkers. It's a highly complex game. There are more board positions than there are atoms in the observable universe and when AlphaGo performed move 37, the observers of that game described it as ‘beautiful’. They said that it overturned hundreds of years of received wisdom, that it was counter-intuitive, indeed creative. Sedol was so shaken by the move that he left the auditorium where the match was taking place for 20 minutes and those overseeing the AlphaGo model were sure that it had made an error. The brilliance of the move only emerged later in the game as it capitalised on the board position that pivotal moment created. Was this just another example of brute force or could it be more aptly described as an example of an emergent property which was not foreseen by those who programmed the system? An original idea? A silicon based flash of inspiration.

Some interesting research around creativity was carried out by Humboldt University Berlin in conjunction with the University of Essex on something called the AUT or Alternative Uses Test last year. The test used LLMs to evaluate divergent thinking.

This involved asking people and LLMs to come up with as many different uses for a mundane object as they could in a constrained time period. So subjects might have been asked to list as many alternative uses for a paperclip, a paintbrush or an ice cube tray in a minute.

Let’s have a little fun and carry out an N=1 experiment. I’ll give you 60 seconds to come up with as many creative and interesting uses for a roll of dental floss and then we’ll compare your answers to Chat GPT. Ready? And Go…

How did you do? Did you come up with 5? 10? Did you start running out of steam half way through or did your best ideas emerge towards the end of the process? What’s your best alternative use for dental floss? How would you rate your output?

When I put this challenge to ChatGPT 4.0, it spat out a list of 50 answers in under 10 seconds. And before we hang our heads in shame as failed apes with an overinflated sense of our own potential, we should acknowledge that this kind of creative task is exactly playing to an AI’s strength. It is trained on massive data sets so it has access to a broader range of information and associations than we have at our disposal. It doesn’t feel tired, it doesn’t get distracted, it doesn’t experience self doubt and it doesn’t critique its output and then self censor for fear of seeming stupid. It churned out 50 ideas - half of them weren’t very interesting in my opinion, a few repeated but then there were some really interesting ideas too. 

It suggested dental floss could serve as a replacement for a shoelace, a fishing line, a clothesline, a way of securing packages, a bracelet, a picture hanging wire, a way of slicing soft cheese, a hair tie, a candle wick, a fire starter, a miniature rope for doll houses, puppet strings, a string for a musical instrument, a fake spider's web for Halloween and a miniature zip line. Not bad in 10 seconds.

Does THIS then prove that Chat GPT4 is creative? After all, divergent thinking is one of the hallmarks of creativity. Emily Bender would counter with a heart-felt ‘no’. In a 2021 paper on large language models, she introduced the term ‘stochastic parroting’ to explain away this illusion of creative thinking. The term ‘stochastic’ means probabilistic or based on randomness while ‘parroting’ suggests repeating something without real comprehension. In other words a GPT model might look like it is reasoning or understanding but behind the scenes it is just using probability algorithms to predict the most likely sequence of words based on the patterns in its training data.

Bender contends that AI, at least the models we have today, are tethered to imitation, Chat GPT cannot innovate so much as interpolate. It’s a neat party trick but that’s all it is. It all comes down to intent.

I can remember a few years back there was lots of interest in elephants, such as Suda from Taiwan, who seemed capable of painting. There were YouTube clips of these beautiful creatures with paint brushes in their trunks, delicately working on canvases to seemingly produce self-portraits. On further investigation, even though elephants are highly intelligent animals who mourn lost family members and have predigious memories, this demonstration of art was merely the result of reinforcement learning. The animals were trained through repetition, trial and error, and reward and punishment to be able to replicate the same painting again and again. There was nothing inherently creative going on and sadly, in some cases, the elephants were brutalised if they didn’t comply. 

The point is, many of us were duped because the idea that elephants can paint seems plausible. Our senses and our brains deceive us all the time. We think a cat brushing against our leg is a sign of affection but then discover it’s leaving a scent to mark its territory. If a couple on a first date do something that arouses their survival instincts such as a rollercoaster ride or a skydiving experience, they are far more likely to find each other sexually attractive than if they did something more sedate like shared a meal or went for a walk in the countryside. Psychology textbooks are full of such examples - so perhaps we are just inclined to attribute the potential for creativity to a machine because we see intent where there is only binary data. We humanise the inhuman. We create creators where there are none just as we see dragons and faces in the clouds.

But then, one could counter even this argument by asking if is it true that there are uniquely human pre-requisites for the creative process such as frustration, curiosity, or inspiration? Is intent and consciousness a necessary condition for creativity or are we just shifting the goalposts here to deny AI systems the keys to our most jealously guarded playground?

Or if we really want to fiddle with the lid of Pandora’s box, we might ask if it is possible that we are more like finely tuned Turing machines than we might want to admit? Just because we're not aware of the coding or the programming that's going on underneath the surface, doesn't mean that we aren’t following similar algorithmic processes in order to arrive at our moments of insight. There are plenty of fields of study that find some uncanny similarities between our minds and the neural networks which form the basis of our AI systems. Brain imaging studies, game theory, Norbert Weiner’s early research into cybernetics and Natural Language Processing all draw these parallels. Perhaps we are creating AI systems in our own likeness - meat machines making silicon offspring.

And it's at this point, when we have to start questioning the relationship between creativity, free will and consciousness that we start to run into the limitations of this kind of inquiry. For the time being we can only conclude that the human brain and the neural networks at the heart of complex AI systems are black boxes. One is carbon based, the other silicon. Both display emergent properties that can be unexpected. Both are capable of breath-taking outputs. One is sentient - and it’s unclear to what degree our AI counterparts are or are not self-aware. We just don’t know.

And this leads us quite neatly to ask whether Ai systems will displace human creativity, or will they instead come to complement ourncreative process? Perhaps it’s just because I’m an optimist but  I am inclined towards the latter point of view. I think we will learn to use AI systems as tools or collaborators rather than find ourselves mixing paint and sharpening pencils  for our robot overlords.

There have always been moral panics around the ways in which technological advances will impact creative spaces. When the camera was created in the early 19th century many artists believed that the photograph would be the death of canvas, and this was not of course the case. In fact many artists appropriated the camera to augment their own creative process by taking snapshots that they could then print and use at home in their studio.

Likewise, when the first electric guitars emerged in California in the 1930s it didn’t spell the death of acoustic music so much as a broadening and enriching of the genres of music that emerged in the following decades. And I think that the same thing will be true of the way in which we work with artificial intelligence. Technology will always amplify intention - if our intention is to create, we will be able to do more quickly, in more interesting ways and in different ways when we have access to AI. 

Used congruently, generative AI platforms can democratise the creative process. With limited training and investment, my students have been empowered to  create illustrated storybooks, orchestral arrangements and animated film clips without having to deal with publishers, recording studios or actors and filming sets. Today if you have a creative vision and are willing to embrace the learning curve that comes with these tools, you can compete with professional artists around the world from your basement just as Steve Jobs built Apple from his garage back in the 70s.

On the other hand, of course, there is the danger that we will soon be trying to navigate marketplaces flooded with homogenised low quality creative outputs that are so prolific that it will become hard to find the really worthwhile content from all the dross. Browsing the Amazon Kindle store or YouTube which has become hijacked by faceless YouTube channels can be depressing. Again - technology amplifies intent. If the intention is to maximise attention, manipulate the algorithm and make money, then AI will serve us up the quagmire we conjure into being. 

If we use AI to create deep fakes, photorealistic footage of celebrity affairs or politicians caught in scandals, then the same impulse that gave us the Sistine chapel or Van Goch’s Sunflowers will be used to cast us all into a hall of mirrors when it is almost impossible to discern truth from fiction. I don’t think it’s a coincidence that cultures throughout history have depicted the creative impulse as a flame. The same element that can warm us and protect us from the dark can be used to burn down the world around us. An AI system isn’t a camera, or an electric guitar - because each of those tools is domain specific while AI is protean in its potential. 

With this in mind, and in an effort to navigate away from these darker waters, let’s briefly return to Garry Kasparov for a moment. When he was defeated by Deep Blue back in 1997, he didn't spend very long licking his wounds before he pivoted to see the potential of the machine that took his crown. A year later, he took part in a new kind of chess tournament in Leon, Spain in which he and Veselin Topalov would compete against one another while augmented by their own respective chess engines. Dubbed ‘advanced chess’ but now more commonly referred to as ‘centaur chess’, it opened a whole new field of creative play. 

In these centaur matches, a human player uses a computer to model the various moves available given the board position and then he or she retains strategic oversight and decision making based on the feedback produced by the AI system. Kasparov and others have shown that an average human player augmented with an average chess engine will consistently outperform a sophisticated supercomputer working without a human in the loop. In my imagination I’m humming Jack Johnson’s iconic beach tune ‘Better When We’re Together’ or perhaps Bette Midler’s ‘Wings Beneath my Wings’ or maybe even ‘We Can Work it Out’ by the Beatles - I’m on a roll!

And reading about Kasparov’s journey led me to think about how actually, in many ways, we are all of us centaurs. We are all hybrids using AI systems to augment our human interactions and activities. We use GPS to navigate from A to Z. We use mobile phones to project our consciousness to different places and times. And even in the art world, this idea of combining an individual artist's creativity and an industrialized or mechanized process is nothing new. 

Think about the work of Andy Warhol in the 1960s for example with his famous ‘factory’ which was populated by a plethora of studio assistants. Warhol would have a concept, an idea, but many of his iconic silkscreens with figures like Marilyn Monroe, Jackie Onassis, or Chairman Mao were produced by those in his extended creative team. He literally industrialized his creative process and he wasn’t shy about admitting it. He coined it ‘art by telephone’. He confessed that he was excited by the ideas he had but didn’t have the time or inclination to then carry out the production of his vision. So he outsourced those parts of the creative process that he was not inspired by while maintaining an oversight and signing the canvases to authenticate them. Canvases which still sell for tens of millions of dollars when they come to market.  Would anyone refute that Warhol was an artist and that the works were his? If not, how is this different from you or I prompting Midjourney to create an image or Suno to record a song to our specifications?

We may not be able to state definitively that AI systems are capable of creating original works on there own at this point in history but the Warhol benchmark as I have called this sniff test, would certainly seem to give us precedent to argue that AI tools can be used to augment human-inspired artefacts and texts.

Do I have to get my hands dirty to be created, to be an artist? Or is art and creativity more a cerebral and intuitive process? 

With all this in mind, I think probably the direction of travel is that we're going to end up with some sort of hybrid creative process. A human being may have an idea and will prompt an AI model, which will then execute on that idea and iterate possible outcomes. Then the human at the other end of the process will curate the outputs and refine them towards the initial vision that they had. And indeed, this is the kind of process that I now find myself using more and more. 

When I'm producing a thumbnail for a YouTube upload, for example, I'll have a concept. I have an idea in my mind's eye. I will use a Lora art generator that's been trained on my image.and prompt for that idea. The image for this episode, for example, is an image of me as a robot artist in light-filled studio surrounded by canvases and paints. I will enhance that prompt using a large language model. Then I’ll run that prompt, producing 4 images at a time and see what is produced. After each batch of images, I’ll modify the prompt to get closer to the image I want to produce. This might happen quickly or it may well be that I need to produce maybe 10 or 15 runs with until I land at an image that approximates what I had in my mind's eye. 

In effect, I will throw away 95% of the outputs. But once I find something I like, I will then take that version into an editor and I will do in-painting and out-painting. There will be parts of the image that I want to change, parts of the image I want to extend or erase. Finally, I'll download the edited image into Photoshop and I'll play around manually with color filters, the hues, the warmth of the image and so on and add and arrange the text I want included. This whole process will take anywhere from 20 minutes to an hour or more depending on how effectively I collaborate with the AI models and how skilfully I prompt and iterate on the first outputs. It requires a mixture of technical knowledge, good prompting and a clear idea of what I’m trying to achieve.

I would argue that this is a creative process. An idea that was only in my mind's eye is realized in the real world and can be shared with others. I couldn't have produced that outcome, certainly not in the time frame I had available, without the use of these AI models. The AI systems couldn’t have produced the image without my guidance and initial inspiration.

Similarly, when I'm producing a script for an episode of this podcast, I jot down my initial thoughts. I will often then have conversations with AI models to help me flesh out those thoughts and to suggest directions for research. Once I've got a whole series of notes in front of me, I will record my voice using an AI transcription tool called Audiopen. 

I'll take the transcript from that recording and I'll start working with it and shaping it as a text file until I'm happy with what I have before recording the final episode using the script. Any bad takes or stumbles are then edited out using an AI augmented editing platform called Capcut. I also use it to generate a suggested set of show notes with time stamps and social media shout outs to promote the episode. I will modify and humanise them but they provide a useful starting point rather than having to compose these from scratch.

It took a while to fine tune these kinds of workflows but I’m pretty happy with them now and I move seamlessly from tool to tool, some AI augmented, others entirely human driven, to produce each week’s content. Working as a podcast centaur (not a label I would have expected to use by the way!) allows me to significantly reduce the recording and production time so that I'm more likely to be able to maintain a weekly output whilst holding down a full time job. 

Elements of this same kind of augmented work have found their way into my lesson planning, resource creation and in my email correspondence and report writing. I’m sure with a little thought you can devise your own creative processes that serve your needs too.

So to finish today, I'd like to offer five simple prompts you could use today with any LLM such as Claude, Chat GPT or Grok to augment your own creativity in the classroom.

The first of the five is using AI as a devil's advocate. It's really helpful to be able to find ways of getting outside of our own thinking process to move beyond our own assumptions, blind spots and ideas. You have no doubt already done that by having conversations with friends who might adopt a benevolent but adversarial position to your own. This is the reason why debate is considered such a useful pedagogical tool - it helps us to tease out the nuances of the topic we are learning about by holding an open mind to different positions or readings. You can just invite the Chatbot to challenge you on the position you input or you could request the script for a constructive round table discussion on your topic with speakers holding well argued but divergent points of view.

Next, we can use AI as a brainstorming partner. We can ask it to give us multiple potential ideas for a topic that we're thinking about to provoke our own creative thinking. If preparing for a lesson on magnetism, we might ask for 10 different starter activities to grab our learner’s attention or ask for 5 different acronyms for memorising rhetorical devices that could be used when planning a persuasive speech in exam conditions. Then we should be prepared to adapt the output until we arrive at something we are confident is fit for purpose.

The third way we can use AI to augment our creative process is as a mashup tool - these models are excellent at producing unusual blends or remixes of inputs. We might ask it to suggest ways in which the French Revolution can be presented using the format for ‘Strictly Come Dancing’ or we could ask to devise a production of ‘Hamlet’ inspired by the setting and musical theatre of ‘West Side Story’. Many of the outputs from these kinds of prompts will be truly awful but there will be some golden nuggets along the way too.

Fourth, we can use AI as a constraint generator. Often creativity emerges as a consequence of restraints and limitations. Poetry is an obvious example that comes to mind; writing a poem of precisely 14 lines, composed in iambic pentameter and with a strict rhyme scheme would seem likely to crush creativity but the inverse is true. Counterintuitively, when we are forced to wrestle with obstructions and restrictions, we think outside of the box and come up with novel and interesting new ideas. So we can invite AI to suggest and recommend some constraints for a lesson activity. I’ve done a competitive card sorting activity with one member of the team blindfolded which was a lot of fun. I’ve also asked pupils to write a short summary of a text we’ve studied and randomly assigned constraint cards - one pupil couldn’t use the letter ‘t’ at any point in their writing. Another had  each new sentence with sequential letters in the alphabet. These examples don’t constitute breakthroughs in pedagogy but they were fun to devise and the pupils seemed to enjoy the challenge too which is half the battle at times. 

And then finally, number five, we can use AI as a rapid prototype creator. Once we've got our idea, it can be difficult as an creative to overcome what I've heard referred to as the ‘tyranny of the blank page’ or the ‘tyranny of the blank canvas’. How do we get started? It can seem overwhelming. How often have we heard pupils mouth the phrase ‘I don’t know how to start?’ Procrastination is often just a behaviour which masks the fear of getting started.

Well, AI does that superbly well. You might ask it to generate a suggested essay outline. Five starting sentences for the opening paragraph. A quick and dirty lesson plan on a particular topic. And you might end up throwing 80% of that content away and replacing it with more refined ideas that you're happier with. And that’s ok because this is just about overcoming the inertia that can hamper the creative process and allow us to make progress.

So there we have it - certainly a wide ranging conversation today and no doubt a topic I’ll revisit at some point in the future.

Thanks for listening—I hope you’ve found some useful takeaways from this episode. Please spread the word if you think others would enjoy the show, and don’t forget to check out the AIcademia YouTube channel for practical tutorials on using AI tools in education. Have a great week, and I look forward to catching up again soon!