Spring til indhold
Hjem » Artikler » Creative Measures

Creative Measures

Hello Hello Helloes! 💕

I wanna tell you something exciting.  First and foremost, I finished my master in Cognitive Science with 5 seconds to spare – haha, lol, what a ride. More interestingly, the thesis that I handed in might have some larger than life kinda consequences – woop woop. Because… I made a state of the art algorithm that might have cracked the code to creativity.

As a standard within my field we use the ‘Alternative Uses Task’ to study Creativity. It asks experimental participants to come up with as many creative uses for ordinary objects as possible. A relatively simple use case for a brick could be to use it as a pedestal, while a more creative use would be to grind it up and use it as paint. Since we’re mad scientists, the task is on time, and that means that most people come up with a mere few ideas, and often the same ones.

When all participants have gone through the experiment and come up with a bunch of creative uses for several objects, human raters look through all the answers and score them on how creative they are. This enables us to score people on how generally creative they are, by measuring across their many answers for the many objects. Think of how IQ scores are created

But wow, sweetie. This is some very tedious work because the raters use 4 different dimensions to capture what creativity is and there are often thousands of answers. Just saying – sliiiiiightly problematic. Since this experiment is expensive in both time and money, wouldn’t it be cool if we could put the rating on autopilot? Well, now we can!

So! I set out to make an algorithm that could rate all the answers for us and I accidentally ended up making one that did approximately as well as humans. More specifically, I used some network science methodology called Random Walk Agents and some Artificial Intelligence stuff called Transformer Models (I’m not kidding, those are the real names, haha) to capture the language structure of people’s ideas. 

Essentially what my algorithm tried to do was to calculate the semantic distance between the objects (e.g. brick) and the ideas that people came up with (e.g. pedestal), using a ton of statistics. And interestingly the distances that my tuned and tweaked models produced correlated as much with the average human ratings as humans did with each other! Omg!

But Sebber, what can this be used for?

Well, I’m only just now trying to figure it out. But imagine having a compas leading you toward creative associations within any field of your interest. You’d also be able to contribute to fields outside your domain of expertise without ever risking reinventing the wheel. Just like Jack Sparrow doesn’t know where his compas leads him, he knows it’s always saucy and savy! 

If you wanna know more over wine or rum, you’re more than welcome to stop by The Dolphin Headquarters or find my contact info on our webpage.