The computer scientists were stunned. They thought they had given the computer an impossible task: come up with a way for a robot spider to walk without its feet touching the ground. But the algorithm found a way past this constraint. It flipped the robot upside down so it could use its elbow joints as feet.

Computer algorithms are excellent at optimising to a well-defined definition of success, as this example shows. For many of the algorithms and applications of artificial intelligence applied to our everyday lives, however, the outcomes are far less spectacular, and success depends on who is asking.

In cases where entire subsections of society care about the result, such as with the growing field of facial recognition or in prison sentencing, algorithms have been proved to provide inaccurate and biased results. As artificial intelligence finds its way into more and more critical areas of our society, this is increasingly becoming a problem.