WHY WE STILL TRUST HUMANS MORE THAN MACHINES

We forgive humans. But we fire machines. Why?

Waymo's cars have 2.1 incidents per million miles - it's 4.85 for humans. Yet few cities allow safer transport. And this behaviour isn't reserved for life-and-death situations.

As we develop solutions - inside Brightbeam as well as with clients - we often find that perfect accuracy is listed as a must-have. Even when human fallibility on the same task might be more than 50%.

Which seems irrational?

Not quite. Behavioural scientists call this Algorithm Aversion. We hold algorithms to higher standards than we hold ourselves.

People lose confidence in forecasting algorithms after a single error - even when it outperforms human forecasters. Meanwhile, we'll watch a human make repeated mistakes and continue to trust their next judgement.

Part of being a Brightbeamer, therefore, is to hack your own psychology. And if you want to do the same, the key is measurement.

Before creating a single line of code, audit the existing human process. Accept the underlying data will be messy. Put hard numbers around speed. And quality. As well as cost and error rates.

Then, as the Very Fast Prototype is delivered, measure performance again. Keep measuring with every iteration. Focus on the delta between human and machine performance.

In this way you'll stop trying to build machines that never fail. And start building systems that fail less than we do.

Today's in-the-end-at-the-end?

Chase progress. Not perfection.