The general idea is that you have lots of ‘normal’ nodes and some which are able to adapt (using a generic algorithm or one of the more recent open ended evolutionary algorithms).
This is in some ways an old topic, but the recent design changes, which include instant join of new nodes and a generally quicker time to earning rewards for providing resources, make it more attractive.
If the evolving nodes know who each other are they can work together to adapt their behaviour (eg tweak control parameters) as a group (population) generation by generation. The rest of the network acts as the environment, and nodes select for better combinations of the adapting parameters according to say, earnings from the network.
It may not be trivial to create an evolving type of node but it doesn’t seem too hard at first sight. The first step would be to identify some tweakable node behaviour parameters and design a suitable mechanism for verifying performance, which would need to include adherence to the evolutionary protocol (to avoid false reporting of performance for example).
In the past I’d be diving into this because the principle has interested me for decades, but for now I’m just putting it out there.
You know that it’s also possible for 3rd parties to build a level-2 data organization on top of level-1 what the bare Safe Network is. Not sure how this would look like and what benefits are, might be a trade-off between privacy (what for most cases might not be desired) and enhanced speed with caching.
So evolving to maximise income, that is they adapt to join in where their is more being stored, they could evolve to the point were uniformity is out the window and chunks end up not having enough nodes replicating them because they fall outside the more income making grouping.
While the distribution of chunks across XOR space is fairly uniform it will not be perfect and evolving to make sure they end up in the more used space could potentially cause issues.
Not saying its a definite but is something to consider.
How could tweaking node parameters help a node or node group do this? If it’s a weakness in this scenario I expect it would be even more vulnerable to human tweaking, or external AI etc.
It’s an interesting question, and I’m sure there are other scenarios we might want to consider or test.
I actually feel that is almost certain that some form of intelligence will be in nodes soon. Dynamic adjustments of parameters etc. should all be done that way. I lean towards genetic approaches too, but certainly, I see the removal of magic numbers almost forcing a degree of intelligence.
The trick will be allowing these to evolve, so initially they will all die as they are kicked out, but then they should improve over iterations. That is the hard part, how to get the rewards coded in and let the dynamic algorithms (or neural net) test assumptions over and over again.
I should have said, the reward is gonna be close to our Network Fundamentals which we may need to carefully consider as there may be ways to achieve those that harm us humans (in terms of data access)