The worst part of change is figuring out what you need to let go of in order to achieve it. Cate Hall (whose writing I admire) has a timely essay on the topic.
Modern life is mess of conflicting and changing realities to which we are more or less poorly adapting ourselves. Learning is hard.
As you might expect from hominids adapted to long extinct physical and chemical environments, the new parameters determining our current physical realities are a challenge for us to update on our own.
Backpropogating a human neural network does not yet have a set of best practices but when it does emerge I’m surely it will be a blogger tying together the layers of training, physical requirements and other weights and cultural measures that improve our learning environment.
Not with me? Google’s attempts to serve AI generated synopses is here to help and their updates might be working as this isn’t bad
Backpropagation is a training algorithm for neural networks, specifically designed to optimize the weights within the network by minimizing the difference between predicted and actual outputs. Backpropagation aims to reduce the error between the network’s output and the desired output
In plain English, you learn by making mistakes and correcting them. You do something again with an adjusted technique and when succeed you update your understanding. You reduce your errors by looking back at what you did and changing your future behavior hoping it will succeed. When it does you adjust. You have learned.
Success might change depending on what you are doing and how your environment changes. Some constants remain. How can we look back on the data in our own lives and in our species and use it to improve our lives going forward?
I don’t know exactly how to approach the current moment but I know I’m adjusting to millions of pieces of input daily and I am still frustrated that I don’t always get the outputs I want. The logical next move is to change.
But change what? And how? What will I be leaving behind in that process? How acceptable is it to let go of what we were so sure we knew? Can we convince others of it? Can we adjust the parameters globally so others adjust too? How do we turn the knobs and dials on the systems that we use to learn at network scale?