Related poll : https://manifold.markets/Guillaume/im-starting-to-fear-the-super-ai-ne?r=R3VpbGxhdW1l
Will we see a 'super AI network' as a result of AI starting to collaborate together before 2035 ?
To qualify as a 'super AI network' as I define it, the network should :
Feature exclusive emerging ability due to the fact it's working in a network
Feature some sort of parallel processing ability
Be adaptative (you can add or remove nodes at any time)
1,000The first two points are technically fulfilled by any neural network, and MoE models (which are already quite common) would probably count for the third: https://en.m.wikipedia.org/wiki/Mixture_of_experts
Although I feel like that argument isn't quite in the "spirit" of the market, I still think in order to get to the vibe you're pointing at in your diagram, it would be enough for neural networks to get bigger and more modular in exactly the same way they already have been, rather than some novel turn of events suddenly occurring. So I believe it's very natural that this question will resolve YES
@CDBiddulph Hum, I thought MoE models where more like conditionnal trees and that only a few 'experts' where involved at each run. If the point is valid regarding all the current criterias I will resolve to yes of course.
I suspect MoE is not exactly fullfilling the 'Be adaptative' criteria, because they need an extensive full model training to be operationnal, so you can't really use new expert in a plug & play fashion in the network (at least that is what I understood about MoE models).
If it appear MoE technically evolve to get the 'plug & play' ability without a new model training (or the model training becoming sort of transparent and automated process in the network), I think it would then satisfy all the current criterias of this market.