Is There a Stability Effect in Machine Learning?

If the system is too chaotic, or if you let the chaotic parts of the system touch the long term memory, that can be a problem for a neural network. So what you might say is that there's a spectrum between explorer mode and exploit mode. And on one side lies catastrophic forgetting. That may not just be a fetrineral networks, it might also be a feature of human culture. On the other end of the spectrum, on the exploit side, where you've found a successful strategy, and maybe it's not working as well as it used to, but for some reason just keep doubling down - that leads to intrenchment effect.