How can we trust automated tests in life-or-death scenarios?

September 18, 2022
Actually, how can we trust manual tests in high-stakes scenarios?

One of the more frequent arguments I hear against automated testing, and by extension, continuous delivery, is the idea that some problems are too important to trust to automated tests. Someone might die if a bug gets through. Or maybe billions of dollars would be lost.

Common examples are things like: firmware for pacemakers, self-driving cars, and Mars landers.

I guess this question seems reasonable on the face of it, because it comes up all the time.

But hidden within this question is the erroneous assumption that manual testing is more reliable than automated testing.

What makes it worse is that the assumption isn’t always wrong. The problem lies in the fuzzy definition of “testing.” Some kinds of testing definitely do require an actual mind, which can reason about things, apply intuition, etc.

But that’s not the kind of testing we’re talking about when we discuss automated tests, or continuous delivery, is it? We’re talking about rote regression testing. “Run through these 10,000 scenarios, and make sure none of them behave in unexpected ways.”

This is exactly the sort of testing that humans are notoriously bad at doing. And not only because they’re dreadfully slow, but also because they easily make mistakes. They also tend to get bored, and the better ones require salary increases every now and again.

Would you rather hop behind the… eh, wheel(?) of an autonomous vehicle that had the last software update tested automatically, by a computer that can examine 10,000 corner cases per second, or by a human sitting in the, er… driver’s(?) seat, trying to trigger unusual scenarios to ensure that the car behaves correctly?

When you stop to think about it, this common question “How can we trust automated tests in high-stakes scenarios?” is exactly backwards.

How can we possibly trust manual tests in high-stakes scenarios?

P.S. The Lean CD Seminar can help you take the next big step toward the test automation you need to realize your software delivery potential. The 4-week program is only €189 and starts October 3. Read more.
Share this

Related Content

The Technologist Podcast #4: Continuous Delivery, DevOps, Go

Coach Denis interviews me about my mission to bring enterprise-class software delivery to small teams with small budgets.

What's the ideal ratio of devs to QAs?

Of course, as a consultant, I can confidently tell you: It depends!

How to travel light

Software delivery is a lot like traveling. Less luggage makes for a better experience.