You Can't Learn From Outlier Events
You're learning to play darts. You don't care about the actual rules, but just want to consistently hit the bullseye. You do a couple of throws, all somewhere around “okay”. Finally, you hit the bullseye! You try to remember exactly what it was that enabled you to hit it… You had your right foot in front! You try again, right foot out front. You hit the edge of the board. Left foot, same result.
It must've been something else. At least now you know that the legs don't matter. You do more throws. You're back to where you started, all of them okay-ish.
Now you do a throw that doesn't even reach the wall with the board! What happened? You took a drink of water. Probably some residual water on your fingers making them slippery. You dry your hands on your shirt. Another throw. No bullseye, but at least it hit the board again. The next time you break for a drink you take care to dry your fingers again.
Did You Spot The Problem?
The title states that we can't learn from outlier events. Yet, you learned that the position of your legs does not matter and that it's important to dry your hands after taking a drink. But since this example incorporates an element of luck, we expect a Regression toward the mean.
When rolling dice, what do we expect to roll after rolling a 1? There's a 1 in 6 chance for each side, so the chance to roll a 1 again is ⅙, but a better roll is 5 times more likely. So if you were trying to “get good” at rolling dice, you might try blowing on the die before rolling it again. You roll a 3. But it was likely to be better than a 1 regardless of your actions. Unaware of this bias you now get in the habit of blowing on dice.
Regression Toward The Mean tells us that in a probability based sampling, we can expect a less extreme sample after an extreme sample.
When we are learning something new that has a component of luck to it we cannot really learn from isolated extreme cases that involve luck. This also applies to playing the lottery of making online content and our thinking about the effectiveness of punishment and reward in teaching.
Over a longer period of time we can still take these extremes account, like hitting the bullseye 6% of the time with the right foot out front and 2% with the left foot, then keeping the right foot out front is definitely something to look into.