These Two Magic Tricks Were Designed by AI Machines

Artificial intelligence is changing the way we interact with computers. It’s improving voice recognition. It’s letting us more easily search for images. It’s driving cool new devices like the XBox Kinect. And now it’s giving us card tricks too.

Witness the video above, from researchers at Queen Mary University in London. It shows off a card trick called Phoney, but this isn’t magic. It’s driven by clever computer algorithms that let the researchers guess which card is being picked based on the colors of the other cards around it, coupled with probability analysis.

The team taps what are called simulated annealing algorithms, which are used to solve really complicated computer problems. This family of algorithms can tackle computationally intensive riddles like the traveling salesman problem. That’s where the computer works out the quickest route for a salesman, even though there is a really, really large number of highways and possible sales stops to be made.

The researchers applied the same principles to create a “magic jigsaw,” dubbed the Twelve Magicians of Osiris. You can see it here:

The rub is that, for the algorithms to successful pull off the card trick, the Queen Mary researchers resorted to a little trickery of their own. The deck in the video isn’t really being shuffled. That’s a human trick, called a “false shuffle.” Without any real shuffling, the unique order of cards in the deck is preserved. The dealer also surreptitiously enters data into the phone via a phony passcode screen. That allows the smart phone to figure out the card that got picked.

And it works most of the time. You see, the researchers also used some of their own data, about which cards are most likely to be selected when people are asked to pick the card they like the most from a set of four. That means that the trick can fail, but the Phoney software is about 80 percent likely to chose the right card in the first two tries.

The researchers learned an important lesson: humans are an important part of any magic trick—even one that’s being solved by a computer. Howard Williams, the Queen Mary PhD student who devised the experiment, says that the human touch can smooth over things whenever the software comes up with the wrong card. But he and his colleagues were surprised at how often test subjects were amazed by the card trick—even though it uses an Android app to ultimately display the “magically” selected card.

He attributes that to the human magician, who walks the person being tricked though the card selection, pulls off the false shuffle, and smooths over any problems, creating the illusion of magic. So don’t worry, magicians, IBM’s Watson won’t be gunning for David Copperfield’s job anytime soon. “We think the presence of the magician is key even though he’s using the phone to help,” he says. “Without the magician, it wouldn’t have the same impact.”