"Failing 15% of the time is the best way to learn"

I didn’t look closely yet, but people here might be interested in this.

Despite a long history of research, it is unclear why particular difficulty levels might be best for learning.

However, scientists from the University of Arizona say they have now found the “Goldilocks zone” – with their data suggesting people who fail 15 per cent of the time learn the fastest.

Researchers created machine-learning experiments in which they taught computers simple tasks like categorising patterns or arranging numbers. The computers learnt fastest when they got 85 per cent of answers correct, according to the paper published in Nature Communications .


This explains a lot perhaps, need to observe my methods a bit more since I’ve always been more around 50% I believe.

Need to read and work on this.

I love the example :

*Scientists say the “85 per cent rule” applies to things we learn through experiences – for example a radiologist wanting to learn the difference between images of tumours and anomalies that are not tumours. *

“You get better at figuring out there’s a tumour in an image over time, and you need experience and you need examples to get better,” said lead researcher Robert Wilson from the University of Arizona.

1 Like

Their argument makes sense: too easy or too hard and you’re not getting much out of it. Close enough to 80/20 for the Pareto principle to apply, so no big surprise really.

Interesting, I wonder how I could apply this to life as the rate in which you fail rarely seems to be up to you, you just try to complete tasks or learn things that are beneficial and failures and wins don’t tend to be so obvious. Any ideas of practical examples anyone?

Maybe this is only useful for problems with very few dimensions like learning to recognise patterns and not so applicable to more complex systems/problems? Reminding me of N N Taleb’s Ludic Fallacy. Also real world problems may not have such accurate and quick feedback. Just a few thoughts.