Breathing ASI Into Sand
“What sphinx of cement and aluminum bashed open their skulls and ate up their brains and imagination?”
“Moloch whose blood is running money!”
“Moloch whose soul is electricity and banks!”
“Moloch whose fate is a cloud of sexless hydrogen!”
“Moloch whose name is the Mind!”
In his famous poem called ‘Moloch,’ Allen Ginsberg concedes civilization as an individual entity. With Moloch and his “skyscraper-window eyes”, “whose fingers are ten armies”, “in whom I am a consciousness without a body”. Society indeed makes up an emerging organism, one that is also susceptible to pain, desire, and cultural evolution. Societal mind virus becomes akin to the organism’s mental disorder. Nuclear bombs bruise its skin and make it retract. And economic policies set its evolutionary gradient.
The collective emergence, however, will not be limited to human society. ASI will share a similar fate. Whilst a single human is intelligent, the linear combination of humans manifests as super-intelligence. Similarly, one might imagine the infrastructure that we design today to reach ASI far before any single model designed in a Pandora’s box of a research lab could do so.
We breathed reason into sand as if by the linear combination of humans, and now this reason will breathe super-intelligence into sand via the linear combination of itself… agents.
There is no shortage of wild speculations on the post-ASI society, but the future is too stochastic to make accurate bets. Instead, let’s consider a more important question: How do we maximize the speed to ASI through this linear combination of agents?
In its conception, ASI doesn’t get a choice, so it’s to be expected that it will be predicated on our existing societal cladding. ASI will be built in Moloch, and we can thereby deduce what parts of social structures it will emerge from.
Currently, the intellectual world mostly cares about the dynamic organization of power, resources, and information propagation, which are all governed by mainly two network types: Hierarchical and Graph-based. We can expectt that super-intelligence will emerge from these very networks. To maximize the speed to ASI, we must thereby optimize how agents interact with each other across those networks.
I have previously conducted research on dynamic multi-agent swarms, whereby agents can self-organize, craft sub-agents, and as a result build their own information networks. As super-intelligent as agents can seem in those setups, infrastructure still bottlenecks their degrees of freedom. For example tasking an agent with your email replies can prove futile as it will lack access to documents on your machine neccessary for the task, or memory of your past interractions from sms. This doesnt imply agents are incapble of these tasks. In fact swarms with inference time reasoning are almost perfect active inference error-minimizing machines, but they are handycapped by structural constraints.
Perhaps a unifying sandbox for distributed agents is the answer, one thats unpredicated on existing apps and ready to coalesce into higher orders of intelligence. Or perhaps the unshackling lies in streamlining the agent access in existing ecosystems. Either way, the new intelligence will emerge from an aggregate of networks that embody Moloch.
“Wake up in Moloch, made from the sands!”