The Alarm Paradox

Why humans and machines are really one learning system
Machine learning is commonly applied to replace human decisions: a zero-sum game
For example, deep learning algorithms have now beaten humans in trivia, chess, and go.
But large enterprises aren't playing games. They need machine intelligence to empower humans.
But this can be tricky. For example, an alarm system requires a person to make a decision.
A triggered alarm compels a person to verify if a threat really exists.
Unfortunately, some signals are unreliable. For example, a door opens mostly for non-threatening reasons. We call this signal noise.
Also, when you are predicting something that occurs infrequently, the false positive rate is very high.
And when systems are wrong more than they're right, people ignore them. That's the alarm paradox.
This pattern of ignoring systems will degrade the quality of data
Our goal is to change behaviors to lower signal noise and false positives.
If we can make people aware of this, we believe we can change behaviors to lower signal noise