A prosthesis for imagination: Using AI to boost your creativity
AI can already beat humans in many measures of creativity. Let's use that to our advantage.
By many of the common psychological tests of creativity, AI is already more creative than humans.
For example, one standard measure of creativity is the remote association test (or RAT). This test asks people to find the common word that connects a set of three seemingly unrelated words. For example, pine, crab, and sauce are connected by the word “apple.” (Try one: what word connects cream, skate and water? The AI answers it in the screenshot!) Unsurprisingly, ChatGPT is very good at this, maxing out the test even when I created novel RAT questions that had not previously appeared in the literature.
A second standard measure of creativity is the alternative uses test (AUT). This measures an individual's ability to come up with a wide variety of different uses for a common object. In this test, a participant is presented with an everyday object, such as a paper clip, and is asked to come up with as many different uses for the object as possible. For example, a paper clip can be used to hold papers together, to pick locks, or to fish small objects out of tight spaces. The alternative uses test is often used to evaluate an individual's ability to think divergently and to come up with unconventional ideas. Again, ChatGPT maxes it out in my informal test (in case this was based on training data, I also generated some new AUT prompts, which it also aced). This is not surprising because previous, more formal studies had found that earlier versions of GPT were close to human-level performance.
Now, obviously, these tests were made for humans, not AIs, so it isn’t clear that they are measuring anything that is truly creative… and yet… as AI technology continues to evolve, we are seeing more and more examples of machines generating solid stories, art, and even music (Seriously, click that link if you haven’t seen Google’s new music-generating AI). It's easy to look at these developments and start to panic, thinking that AI is on the cusp of replacing human creativity altogether.
But I don’t think we should see this as competition. In fact, the strength of AI is that it can help us break through the parts of the creative process that many humans are bad at, and let us become better at generating ideas ourselves. This isn’t theoretical. I want to give you three ways you can use AI right now, and especially ChatGPT or GPT-3.5, to be more creative when you need to solve a problem or come up with a new idea.
Fluency: Creating lots of ideas
An issue with human idea generation is that not everyone is equally creative. Some people are really good at generating ideas, and they can apply this ability in almost every context. Indeed, recent research has shown that the “Equal Odds” rule is true, which is that very creative people both generate more ideas (fluency, in the graph), and better ideas (creativity, in the graph), than other folks. Overall fluency was not correlated with intelligence, and people who were bad at idea generation might still be best at creative execution, but, as the graph shows, there are some people with the skill of idea generation and some without. Until the past year, there was no prosthetic or approach that would help people who were not good at generating a lot of ideas to do better (besides coffee). Now there is: generative AI. As we saw in the AUT test, is excellent at generating a long list of ideas.
I suspect the Equal Odds rule does not apply to AI, and most of the default ideas of AI are of mediocre quality (though more on this soon). But that is okay, because that is where you, as a human, come into the equation. You are looking for ideas that spark inspiration and recombination, and having a long list of generated possibilities can be an easier place to start for people who are not great at idea fluency.
So how do we do it? Let’s imagine that we want to come up with 20 ideas for marketing slogans for a new mail-order cheese shop. The AI can generate those for us, but we will get even better quality if we tell the AI to act as someone who is good at coming up with ideas. You can see below, we get better results if we say: You are an expert at marketing. When asked to generate slogan ideas you come up with ideas that are different from each other, clever, and interesting. You use clever wordplay. You try not to repeat themes or ideas. Here is your first task: come up with 20 ideas for marketing slogans for a new mail-order cheese shop, make them different from each other, and make them clever and creative.
And… most of the ideas are terrible, while a few have potential (and bad puns). But that is okay! It took us no time to generate 20 ideas, and it is costless to skim them to see if they inspire better ideas. If none of those sparked your interest, ask for another 20 with different prompting. Ask it to consider a particular customer (a cheese shop that will appeal to novices, or to children), or to change the context of the prompt in other ways. Tirelessly generating concepts is something AIs are uniquely good at.
Variance: Creating novel ideas
Another key aspect of idea generation is to embrace variance. Research shows that, to find good novel ideas, we likely have to come up with lots of bad novel ideas, because most new ideas are pretty bad. Fortunately, we are good at filtering out low-quality ideas, so, if we can generate novel ideas quickly and at low-cost, we are more likely to generate at least some high-quality gems.
There are lots of potential ways to do this with ChatGPT. One is to play with constraints. In general, and contrary to what most people expect, AI works best to generate ideas when it is most constrained. Force it to give you less likely answers, and you are going to find more original combinations. You might want to ask: You are an expert at problem solving and idea generation. When asked to solve a problem you come up with novel and creative ideas. Here is your first task: tell me 10 detailed ways an AI (or a superhero, or an astronaut, or any other odd profession) might make espresso. Describe the details of each way.
And, in seconds, we now have 30 diverse and unusual ideas. Again, most are terrible (“The superhero uses their shape-shifting abilities to transform into a coffee machine, brewing and serving espresso shots directly from their own body. This method ensures that the coffee is always fresh and perfectly brewed.”🤢) but some are interesting and might spark further thought (Heat vision espresso makes me think of unusual heating elements to heat water faster, the AI predictive grinding for beans has me thinking about how there are no smart grinders on the market that I know of). And, again, if there aren’t enough ideas on the table, generate more.
You can also use other techniques, like asking it for interview transcripts for fake interviews: Create an interview transcript between a product designer and a dentist about the problems the dentist has, for example. Or ask it to describe non-existent products: walk me through the interface for a fictional new water pump that has exciting new features. There is an art to this that you can learn from experimentation, and you should feel free to share other prompt techniques that work in the comments.
Getting unstuck: avoiding inertia
People always ask me if I use AI to write my posts. I don’t, but I absolutely use it when I am feeling stuck to inspire and “unstick” me. For example, I rewrote one paragraph in this post three times before asking ChatGPT to write it for me. The result was good enough to keep moving forward, and I have since gone back and modified it beyond recognition.
This is not uncommon. Many of us have ideas, but often we have trouble actually taking the next step with them, since, to borrow from Eliot, “Between the idea/ And the reality/ Between the motion/ And the act/ Falls the Shadow.” There are many Shadows, the things that stop us from taking action on ideas and, as a result, many good ones never become reality. A great thing about ChatGPT is you can always ask it to help you with the step you are stuck on: here is my first sentence, continue it. Or give me the subheadings for the following story idea. Or give me a fictional transcript where I pitch the CEO of a family-run Italian espresso company the idea for a new kind of espresso machine using laser heaters. There is tremendous value in overcoming inertia, and often a small push is what we need to do it. So, no, I don’t let ChatGPT write these posts, but that doesn’t mean that it doesn’t help me write them.
Idea generation is similar: it is unlikely that ChatGPT will generate your next great idea alone, because, no matter how well it scores on the RAT test, it isn’t really creative in the way a human is. But, by providing you with fluency, variance, and momentum, it can be a critical partner when you need to solve your own problems.
Though funnily enough, it cannot generate examples of RAT test questions, When I did, it suggested the three words are "swim," "hot," and "summer," and the correct answer would be "pool."
In this paper we showed that if you ask GPT-3 to decide how novel and useful its own answers are, it can actually do better than human level on the AUT: https://openreview.net/pdf?id=8HwKaJ1wvl
After it failed to provide reasonable answers, I asked it to explain the rules.
Can you explain the rules of our game in your own words?
Sure, the game is to provide three words, and I have to combine each word with a fourth word to create new English words that refer to different concepts. The four words must not be variations or misspellings of the same word, or proper nouns.
Let's reverse roles: Can you give me three words so that I can come up with a fourth word that I can combine to form three new English words that refer to different concepts.
Sure! Here are three words for you:
Yes, that's correct! Firework, Stonework, and Glasswork are all English words that refer to different concepts.