Systems remember what happens in the end few systems remember the choices that were made this difference is really important when many agents start working on a large scale at first it seems like no deal if the system forgets things or a task gets done then the next task starts and the system keeps moving
But after a while the same unusual problems come back the same arguments happen again the same corrections need to be made by hand.
Actually nothing went wrong the system just did not remember things this is how I think about ROBO when it is part of the Fabric Foundation.
I do not think about whether agents can get things done whether the system gets better after every tough situation or systems that work on their own are always getting experience.
Every time a dispute is solved every time something is overridden and every time a strange decision is made about a route the system learns something
If that information is lost once the problem is fixed the system has to pay the cost all over again later that is the hidden cost of forgetting things.
A healthy system that works on its own uses problems to create memory.
Patterns become rules that the system follows.
Exceptions become part of the systems logic.
Things that used to need an operator now happen automatically.
Unhealthy systems work differently.
Problems get fixed nothing really changes.
Operators remember how they fixed the problem not how the system was fixed.
Over time the difference, between the two systems gets bigger.
One system gets knowledge and the other system just gets used to how things are done.
The difference becomes clear when the same problems happen again and again.
How often do the same problems come back?
How often do operators have to solve the kind of problem twice?
If the same problems happen often the system is learning.
If the same problems keep happening the system is not getting better.
People often think that speed and getting things done quickly are important.
Systems that last a long time are actually getting better at something else: remembering things.
Because systems that work on their own do not get better by doing things more often.
They get better by remembering why things worked or did not work in the past.
The systems that remember things well are the ones that eventually do not need to be supervised.