Machine Flips Coin 10,000+ Times

Flipping a coin is a time-honored tradition for making random decisions, but is it truly random? While one could debate that point on both a philosophical and physics standpoint, Andrew Consroe decided to "find out" for himself. He build a machine that he used to flip a coin 10,000 — or more precisely 10,040 — times, analyzing results after the fact with computer vision.

The mechanical setup is quite clever, as a bowl-shaped device with iris-style arms on the bottom. These arms push the flipped coin toward the middle using a stepper and gear system. When in place, a second stepper, along with a cam and spring, flings a poker up to flip the coin. The cycle repeats, and repeats, and repeats — thousands of times as needed.

A camera takes a picture of the coin for analysis, which is augmented with a small dot to aid in recognition. Consroe then went through these images after the fact, allowing him to try different techniques in order to "make heads or tails" out of the results using computer vision. In the end, he was able to get it to automatically recognize this binary data with a 1.1% error rate. He knows this because he actually went through manually and corrected readings that were off!

While certainly no small task, the result was 49.3% heads, 50.7% tails. It's just enough variation that one would wonder if tails does have a slight advantage, or if that was just chance. Here's some further reading on the subject via Smithsonian Magazine, and Consroe put his data up here for your perusal.