« Delicious is now part of yahoo.... | Main | Peopleware & teams »
December 19, 2005
Non-linear managers
There is a theory in neural networks: Any number of linear layers has the same computational power as a single layer of linear processing nodes. Basically, introducing a linear function in between two linear processing layers doesn't increase the computational power of the network. If you are interested see the classic work Pereceptrons by Minsky and Pappert. Now what later researchers showed is that a non-linear processing layer increases the computational power of the system. So we have Input -> Linear Layer -> Output == Input -> Output != Input -> Non-linear layer -> Output. For example a linear network cannot solve the XOR problem but a non-linear one can.
Now organizations with managers can be thought of as a layered processing system with Inputs, middle management layers and outputs. If you are a manager in this type of system your goal should be to be a non-linear manager. This means don't simply summarize the results of the layer before you and pass it on. If you do you haven't increased the computational power of the system, and your manager could simply sum the results. You should be a non-linear function, increasing the computational power of the system and creating an organization that is more powerful with you than without you.
Posted by Chris at December 19, 2005 10:33 AM