Why How You Respond To Being Wrong Matters More In The AI Era

I was listening to a speech given by one of the early pioneers of artificial intelligence when he used a word I had heard before but never really thought much about: backpropagation. I looked it up because I was curious about how it related to the way humans learn and develop curiosity. Backpropagation is part of the process used to train neural networks to adjust after seeing how different their output was from the desired result. In very simple terms, the system makes a prediction, discovers that there is an error, and uses information about that error to make adjustments that can lead to a better prediction the next time. That made me curious about what happens when people discover they are wrong and how they respond to being wrong.

What Being Wrong Can Teach Humans About How AI Learns

The basic concept behind backpropagation is simple. Imagine an AI system that is learning to identify images. It produces an answer, compares that answer with the correct one, and determines how far off it was. Backpropagation helps identify which connections inside the neural network contributed to that error, and the training process adjusts those connections so the system has a better chance of producing the right answer in the future. The error is useful because it provides information about what needs to change.

That is an interesting contrast with the way people sometimes respond to being wrong. People can become embarrassed, defensive, frustrated, or determined to prove that their original answer was reasonable. A neural network does not have an ego to protect during training. The information about the error goes back through the system so adjustments can be made. Humans have a choice about whether new information causes them to make an adjustment or causes them to defend what they already believed.