This is one big aspect of the ML revolution that I'm uncertain about. Clearly the future system of how models communicate will have a huge impact on their effectiveness and the form the world society will take. The older I get the more I appreciate how the collective intelligence and society of humans matters more than the level of individual idealized intelligence.

Only working together can we build a cheap toaster: https://en.wikipedia.org/wiki/Thomas_Thwaites_(designer)

I buy the Sapiens argument that our dominance is due to being more social, than being "smart" individually.

On the other hand big dysfunctional group can hobble the effectiveness of even its best members (I bet I don't need to cite an example for you).

But the opportunity is far less constrained than this. The space of human societies is limited to agents who all have similar structures and relaitvely fixed "selves" and goals. ML agents can be as diverse as we like, can change their own nature on a dime, and can have any type of self "identify" or goals imaginable. They are not constrained by any "human nature" to only cooperate in certain ways, but can dynamically pursue any strategy at all.

Perhaps there is a natural convergence or local equilibrium with more stable and uniform identities. But I'd bet on the other side - that there will be an insanely diverse and dynamic society that will be able to coordinate in ways that we can neither compete with nor conceive of. Cryptic message boards are just the first drop in the deluge.