Human Learning with Machine in the Loop - Wisdom of Experts

What would you guess?

During a behavioral finance class over Zoom, my lecturer held up a bag of coins to the camera. He asked everyone to type their best guess of the total value into the chat, with one condition: whoever guessed the highest had to actually buy it. The lecturer had been running this exercise for years. Every single time, the highest bidder overpaid. The person with the most optimistic guess ended up holding a bag worth less than they paid.

But here is what's interesting. The median of all the guesses was almost always very close to the real amount. Not the winner. Not the smartest person in the room. The median. The crowd, taken together, knew something no individual did. The wisdom of crowds, playing out live in a classroom.

This is the same intuition behind an approach in machine learning called bagging. If a lot of weak learners vote on an answer, the result is often better than any single one of them. Each weak learner makes its own mistakes, but because they're all wrong in different ways, the errors cancel out. What's left is close to the truth.

The same logic applies to human learning. One source is one data point. Many sources start to resemble something more reliable. Even more so when those sources are domain experts. When people who have spent years in a field independently land on the same idea, that convergence carries real weight.