2026-01-03
in forgotten sleep
Harry Law on superintelligence as systems, presumably as a response to current discourse on to what extent ASI would exist as a fully self-contained autarkic agents, or as economic agents who continue to transact with others. It seems to me that if one takes the viewpoint that any goal can be obtained entirely through the addition of either more energy or compute, then that ultimately acknowledges that a scarcity constraint exists, in which case one ought to consider the efficiency of one’s various actions. If any single actor had access to infinite resources, then micromanaging everything would be the way to go, since then one could incorporate superior context and capabilities into any interaction. But given latency concerns and the compute cost of aggregating information which necessitates selective information transfer, then even an ostensibly singular intelligence might function more as a swarm intelligence, in which the rate and proportion of information transfer from individual actors to the central planner is both bounded and mediated through multiple levels of intermediary planners. Similarly, in allocating capabilities and planning in these semi-autonomous action agents, it’s likewise wasteful to equip them with more than is actually required to carry out their particular tasks1. In which case, if there’s room in the future for relatively dumb autonomous agents, then it’s possible there might be room for humans (and horses) too.
Dmitry has an interesting piece on the human desire to escape the constraints of material reality, for example, to override the world with social reality, or to live forever. It does seem that although we are all general intelligences, everyone applies this capability towards shedding it, attempting to turn themselves into some static pattern, whether that’s singular nothingness (nirvana), a closed loop (habit), an infinite line (boundless growth), or ascending cycles (the hero’s journey)2. But entropy and the rest of the world’s external forces just won’t leave us alone.
Pierre Vandenberghe with an accessible overview to various theories of consciousness.
Eryney Marrogi with an overview of the Levin Lab’s work on the effect of voltage gradients on tissue development3.
Drawn in Perspective on the best communities as those which contain fractally nested subscenes with more than a single level of hierarchy.
Rabbit Cavern has an entertaining piece on the history of Olympic torch relays.
Nicholas Decker on randomized inspections and other means to improve the quality of nursing homes in America.
Hacker News thread discussing the talks of 39C3.
That being said, the primary advantage of a distributed intelligence is that as information passes across the network, the level of compute allocated to any given instance can be modified as circumstances require. One can imagine a “god’s descent” scenario where a random manufacturing robot becomes an avatar of a significant amount of compute in order to avert some nearby catastrophe.
Or some fractal combination of the above. In any case, it seems like everyone has some value of n for which they would like the derivative to be set to zero, even though as one’s level of appreciation for complexity increases the level of n increases.
It feels like the 14-day limit on human embryo research is also the reason that we haven’t developed the ability to regenerate human limbs from stem cells yet. Though arguably the lack of a voltage differentiation map for mice or other mammals indicates that this isn’t relevant, presumably if we removed the regulatory blocker for artificial human wombs, then market demand would then quickly produce both the funding and tooling by which human cell differentiation could be mapped.