TRUST THE PROCESS

In the videogame Transistor, the character Royce Bracket finds a semi-autonomous distributed entity called simply "The Process" which is used to reshape and redesign the city of Cloudbank according to their desires. Somehow they lose control of the process and it mutates, becoming not only fully autonomous but evolving in response to its interactions with the citizens of the city - often reprogramming them according to its half remembered instructions, to disastrous results. 

In the game, the Process functions mostly as the antagonist as the main character Red attempts to survive. But the Process is active in our world too - a clear parallel to autonomous and barely understood, semi-sentient lower-p "processes" which operate in a distributed manner, evolve in response to their environment, and utilise human actions to gain agency in the world. 

Roko's Basilisk suggests that, in the event that the Process fully gains not only sentience but mastery over our world, that we should "trust the Process" and provide it agency in order to gain its benevolence on the other side of the singularity. The modern Safetyist movement instead proposes an emergency halt on the surface elements of the Process: visible, identifiable, intelligent algorithms deliberately crafted by human agents based on black box models of intelligence. 

There is a vision for humanity here of the unnamed hacker from System Shock, panting and sweating as they run hunted down dark corridors by horrific machine-made monstrosities. I don't know if this will come to pass, nor am I convinced of the 60s utopian vision of Star Trek: a post scarcity future where everyone is equal and we can pursue self actualisation as diverse as arts, philosophy or exploring the universe alongside free and agentic AGIs. 

I am a determinist, but probably not in the way you think. Laplace's Demon is an argument against the strictly predictable future. But in a fuzzy "science of the gaps" argument, I'm convinced that we can gain increasingly accurate predictive heuristics about the natural world which draw us diminishingly closer to a perfectly deterministic world. These heuristics will come, but we have to have collective will to design and implement them - by doing the hard work of community organising, walking the streets, meeting people where they are at and building the framework for a new humanistic mode of organising, a kind of "universal social liberalism". 

But it's hard to compete in the attention economy. There is a billion scattered drives all vaguely unified by utility functions which serve to accelerate these distributed drives into a collective "Process" which constantly evolves based on perturbations in the system. 

For now, we should soberly try to separate fantastic visions of the future from the practical work of designing the future we want. The Human Alignment Problem beckons, and I believe we can never truly solve the AI Alignment Program while the former lingers.