Hi ,

​

When I was in Year 9, I studied Australian Geography at school (bizarrely, it was taught using a Hong Kong-centric textbook, but that's beside the point).

For our mid-term exam, our teacher showed us a series of slides depicting Australian landmarks and our task was to name the geographic formations they represented.

I got 19/20 on that exam, and decades later, I can't remember a single answer I got right. However, the one answer I got wrong is burned forever into my brain (in case you're wondering, Uluru is an inselberg).

Getting that question wrong on that exam had a far greater impact on my learning than preparing for that exam ever did.

Here's the thing...

No one enjoys getting things wrong. But getting things wrong creates data - and we can learn from data, so we don't make the same errors the next time around.

In education, this is known as "deliberate practice", and it's how humans gain expertise.

In data science, it's known as "unit testing".

Rather than simply releasing a model (or code) into production and hoping for the best, the idea is to effectively try to "break" the model (code) while in development, so any mistakes can be learned from before they have the chance to cause any real harm.

Data scientists know the value of data. And even though every effort should be made to avoid mistakes, anytime things do go wrong should be seen as an opportunity to gather as data about the mistake, not a cause for despair.

As Thomas Edison once said, in response to a question about his own failings when developing the lightbulb: "I have not failed 10,000 times - I've successfully found 10,000 ways that will not work."

Talk again soon,

Dr Genevieve Hayes.

All past emails are available in the archive.