DOES YOUR TEAM KNOW ENOUGH TO CHECK AI?
In 1962, a woman at a desk in Virginia spent a day and a half doing sums a computer had already done. And she got the same answer. But that’s what it took for John Glenn to trust the computer sending him into orbit.
If you’re putting humans in loops, you want someone like her.
Glenn was about to become the first American to orbit the Earth. NASA’s IBM computers had worked out his trajectory. But he asked the engineers to ‘get the girl’ to check the numbers by hand. ‘If she agrees with the computer,’ he said, ‘then I’ll trust the computer.’
The girl was Katherine Johnson. She was 43. And she knew a thing or two about computers, because she used to be one.
Back then, it was a job. NASA’s forerunner hired ‘computers’ to do its sums by hand, and in 1953 Johnson joined them, in an all-Black, all-female unit. She’d built quite a reputation, and she’d done the trajectory analysis for Alan Shepard, the first American in space.
Her day and a half found nothing wrong. When nothing’s wrong, a good check and a bad one can give the same answer. The difference is what each would have caught. Glenn believed if something was wrong, she’d have caught it.
Back down on Earth, the usual assurance when AI goes into a process is a human in the loop.
But which human? Someone who used to be the computer?
Could the person signing off work through the important parts without the AI? When did they last try? If the answers are ‘not really’ and ‘can’t remember’, what does having them ‘in the loop’ give you?
And what about the new starter who’d once have learned by writing ten reports a day, and now reviews a hundred an AI wrote? Where do they learn enough to spot the wrong one?
Johnson told her grandchildren: ‘First you learn how to work the problem, and then you can go use the computer.’
Glenn had someone who used to be the computer. In a few years’ time, will you?
