Artificial intelligence creates magic trick

Artificial intelligence will change our future. Even the future of magic tricks.
15 August 2017

Interview with 

Peter McOwan, Queen Mary University of London


As quickly as you can - think of a colour. Was it red? If it was, we didn’t read your mind, that’s just the colour that most people think of under pressure. Indeed, because we’re all the same, our brains tend to make similar associations - and mistakes - and this is what magicians often exploit to amaze and impress us. Of course, magicians have been coming up with new tricks for hundreds of years, but now scientists have given Artificial Intelligence - also known as AI - a go, to see if it could come up with a trick too. Georgia Mills asked Peter McOwan, professor of Computer science at Queen Mary University of London, if he’s a fan of magic, as well as computers.

Peter - I am indeed, yes. It goes way back - I think I was about 9 or 10 and my Dad brought me a trick back from a shop that he’d been passing when he was down on a business trip to London, and from thereon in, I kind of got hooked on it. Particularly, I liked the fact that you were using a lot of secret signs and mathematics to make these tricks work, particularly in things like self-working card tricks. If you think about that, actually they’re just implementing some kind of hidden algorithm which allows you to sort the cards in a particular way. If you think about it, a pack of cards is nothing other than 52 separate data elements. They happen to be cards, they could be numbers in a computer. That was something I spotted and thought well, can I use artificial intelligence and it’s ability to take large amounts of data and process those quite quickly to create some kind of new magic trick?

Georgia - Okay. Break down what you did for me - how did you get an AI to come up with a trick, did you just sort of say go, make me a magic trick, or was it more…

Peter - Abracadabra…

Georgia - Yeah!

Peter - Not quite. The work that we’re talking about at the moment is to do with word associations. The question was could take some of those artificial intelligence algorithms and produce some new kind of word association trick? If I was to ask you to think of the first flower that came into your mind, what would your answer be?

Georgia - Rose.

Peter - And if I had written it down on a piece of paper, which I have in fact here - it says rose. Because most people make an association, if asked under pressure to choose a particular flower, the majority of them will say rose. Similarly, there are a whole series of these kind of inbuilt biases.

Georgia - This trick involved predicting a word association you’re going to make from an apparently randomly shuffled selection of images and words. By asking a lot of people and using AI bots to comb the web, they were able to come up with a list of words and images with a few strong associations that people are much more likely to pick, and many others that had no associations at all. Add a bit of clever shuffling, and a bit of pizzazz, and you make it look like you’ve read someone’s mind, when really, it just relies on the fact that most of us think in the same way. So did it work?

Peter - Well, that’s a really interesting question. How do you test a magic trick? How do you scientifically approach that? The way that we did that was first of all to get people to do what’s called self-reporting, so just a list of how did this make you feel from not at all up to where I was really amazed? So you can start to get a number for that.

But equally well, you can take that trick and also compare that trick against another standard benchmarking trick that we used in all the work that we’ve done, which was basically the vanish of a ball using sleight of hand. We found that in the very large majority of cases, people were wowed by the trick, they were surprised by the trick, they couldn’t work out how the trick was done.

Georgia - Does this have applications outside of magic?

Peter - One of the areas that I’m interested in, particularly around artificial intelligence, is how we can use that to try and understand the biological systems; how we can better try and understand ourselves? I think one of the most interesting things I did prior to the magic was I came up with my own optical illusion that was predicted by a mathematical model that we had for the way that people perceived motion. Originally, when I was running that, I got the results and I thought no, there must be a bug in the computer software, we’ll have to go in and fix it. I wasted a week of my life until somebody suggested why don’t you just code the stimulus up and see what actually happens. It was at that point when I did that suddenly I realised that the bug wasn’t in the software, the bug was in human brains because it was an optical illusion and cropped up that way. So doing these kind of mathematical models, these mathematical mimicries of the way that the brain works allows you to, potentially, come up with some new ways of understanding how the brain is processing those massive amounts of information it has to deal with on a day to day basis. So it’s useful from that point of view to try and understanding something about the human process.


Add a comment