Cosmic Coordination

The universe looks dead. If it is actually teeming with ancient advanced life, why don’t any of them use all those resources we see? Yes, there might be other even more attractive resources we don’t see, but it still seems odd none specialize in using what we do see. Yes everything might be under the control of a unified collective, who agree on a preference to keep the universe looking dead. But pretty much any observation could be explained as due to a vast unified ancient power with an arbitrary preference to make the universe appear a certain way. (more)

I’ve posted before that coordination is harder than most folks realize. Today I’d like to emphasize that coordination on the largest scales of space and time, across the entire visible universe, should be the very hardest. Our far minds tend to assume that stuff at this furthest scale is the simplest, with the fewest relevant details, suggesting easy coordination. After all, if there’s just a few kinds of creatures, each with a few kinds of preferences and ways to act, then a simple mechanism might well be up to the task of coordinating them. But in fact creatures the size of stars or galaxies could have vast complex detail, diverging immensely across the universe, and changing greatly over billions of years. Their ability to monitor each other and enforce any agreements must surely be challenged by the vast distances and times involved. And for agreements to not use ample resources, there would remain great temptations to use such resources secretly, perhaps to support a breakout from the coordination regime.

Yes, billions of years also offer a long time to find and implement the best possible coordination mechanisms. But if those mechanisms must be in place before substantial expansion begins from some central origin, that seems to require maximal development of coordination technology at a cosmically very early development stage. If coordination is hard, that seems an extremely high hurdle.

GD Star Rating
Tagged as: ,
Trackback URL:
  • TGGP
  • Carl Shulman

    Running minds on faster hardware lets social processes run faster relative to construction and travel times. Uploads or AIs running at increased speeds could easily have thousands of subjective years to coordinate before interstellar craft (capable of outracing attack) could be produced.

    • Joshua Zelinsky

      That coordination won’t benefit from increased speeds of AIs and uploads if the bottleneck is the speed of light. No matter how fast two AI think, even if they are at stars which are close to each other, they won’t be able to have exchanges of messages more than once every few years.

      • gwern

        Such AIs can think of complex mechanisms to buffer the blow of lightspeed limits.

        For example, if they are heavily coordinated to avoid burning the cosmic commons, this implies a lot of trust; so they could do things like exchange entire copies of each other (want to know what your years-distant partner thinks about some burning issue? Ask Mini-me over in that computing node.).

      • billswift

        gwern, that is the ultimate in expensive coordination. If the two are roughly equivalent, then simulating one would take all of the resources of the other.

      • roystgnr

        billswift, if the AIs solve internal mathematical problems approximately, then a lower-fidelity approximate simulation might be good enough for coordination, and the resources used for AI minds might then have rapidly decreasing marginal utility. E.g. if the approximation algorithms at the core of my AI converge like O(1/sqrt(N)) and if I would normally solve one problem with 99.99% accuracy, then I can instead solve four similar problems with 99.98% accuracy.

        This is something like how non-AIs handle coordination problems, too. My internal model of my coworkers needs only to be good enough to predict how honest each is, not what their favorite colors are.

  • Doug

    I guess the countervailing argument might be that huge pools of intelligence and processing power unimaginable to us would probably be available to any creatures capable of interstellar expansion.

    As an analogy, imagine that we’re both playing chess and we want to coordinate a draw. For any human chess player this would be difficult. Chess is such a complex game that trying to tell if the other player was actually positioning himself for a win would be extremely difficult.
    Even seemingly innocuous moves could be sitting our partner/opponent up for a win.

    However two computers playing chess that are far more powerful and sophisticated than anything today might have an easy time coordinating on this. At the limit if they’ve solved chess, they’ll always be able to coordinate.

  • Hedonic Treader

    I wondered a while ago whether it could be possible to build a network of self-replicating spacecraft that expands through space, but never mutates, even after billions of years of replication. The idea would be to pre-define a replication and distribution algorithm that manages resource use and coordination fairly while enforcing the tightest possible error-detection and prevention during replication so that no defector mutations ever arise in practice. Both the “phenotypes” of the ships themselves and the algorithm which they use to distribute and use resources would have to self-replicate flawlessly.

    Such a system would very probably be sub-par in its resource use efficiency in comparison to freely mutating evolving systems, leaving much resource potential in the universe untapped. However, they may be significantly preferable from an ethical utilitarian perspective: They could support sentient minds free from suffering in rich hedonistic states without having do deal with the competitive pressures from universal darwinism that would otherwise inevitably drag utility down to unpleasantly low levels (maybe even net-negative in the worst case).

    If there are no alien civilizations in our future light cone, and if an entity such as a singleton could design and launch such an initial plan and then ensure the absense of further darwinian competition from earth, this might actually constitute a tipping point in future history that would be most favorable on utilitarian grounds.

    The key would be the deliberate decision to give up some – maybe even much – efficiency in cosmic resource use for the sake of hedonistic quality control. The coordination would then rely not on game-theory between diverging, evolving agents but on the non-mutating replication of a pre-defined distributed algorithm that enforces the hedonistic quality control and prevents defectors from ever arising.

    Your thoughts?

    • anonymousfromUK

      > give up some efficiency in cosmic resource use

      I doubt that there would be a significant tradeoff. Digital copying of source code has an intrinsically low error rate. You only have to protect the goal, not the phenotype of the ships.

    • Tim Tyler

      Self-directed evolution doesn’t need such drastic measures to avoid a race to the bottom. A powerful totalitarian government would be able to control the direction of evolution pretty well – by deciding what gets to live.

      • Hedonic Treader

        Over the course of billions of years, and distributed throughout space? Doubtful. When it comes to cosmic scales, totalitarian central government doesn’t work. The only kind of coordinating singleton I find remotely realistic in such a scenario is a non-mutating pre-defined algorithm that determines all decisions and doesn’t allow for defectors or deviations.

      • Tim Tyler

        A galactic federation probably wouldn’t be very “central”. It would be distributed. Successful galaxies will fairly clearly need galaxy-wide governments to coordinate the demanding task of spreading to other galaxies. It is not clear to me why you apparently think such things are not possible. Communication problems and delays across a mere 100,000 light years? That seems like peanuts.

      • Hedonic Treader

        Tim, I’m assuming a good probability that FTL signal transmission – let alone travel – will be physically impossible. If that turns out to be wrong, and you can jump your police drones to the other end of the galaxy within an hour, I would give credit to the possibility of a stable totalitarian galactic government. We could call that “central” despite the big numeric distances.

        However, if FTL is impossible, and space colonization happens in such a way that openly evolving systems (societies, species, technological replicators…) build a “galactic federation” throughout such large distances, their evolution will greatly diverge. To the degree that they cooperate at all, they will get all the usual game theoretic problems of cooperation vs. defection, secrecy vs. transparency, shifting loyalities, military surprise attacks etc., and no one can directly enforce any central utilitarian plan by relying on everybody’s respect for agreements.

        It seems all but impossible to me to police defectors if messages need 100.000 years to reach them, and police drones need 300.000 years to enforce actual punishment. Of course, there would be game-theoretic equilibriums of sorts, but they will probably include enormous costs of hedonistic utility in comparison to a plan that is designed to flawlessly well-align the utility function of all players in the very long run to optimize subjective well-being (rather than fitness of competing agents).

        As for intergalactic colonization: If the initial plan starts out with a sufficiently good understanding of the universe, the distributed, but non-mutating cooperative colonization algorithm could take into consideration leaps between galaxies. If that doesn’t work due to the large distances involved, surely maximizing utility within one galaxy is still better than scaling war and torture throughout a whole supercluster.

      • Tim Tyler

        100,000 years is not such a long time – when you consider that the universe is billions of years old. I figure coordinating a galaxy-sized creature would be perfectly possible. I do not envisage FTL signalling, though.

      • Hedonic Treader

        You’re right about the age of the universe, but it isn’t the relevant time frame here because the speed of evolution will dictate the difference between convergence and diversity. We already have evidence that rapid evolution can happen within much less than 100.000 years, and this will be especially true after the imminent memetic takeover that you describe in one of your video essays on youtube. From then on, living phenotypes and sentient minds will be shaped by memetic evolution even more than they are already. The pace for this can become extremely fast.

        Of course, there won’t just be one colony 100.000 years away from the next; it’s more likely that a regional continuum of many communicating star systems will come into being. So there will be faster memetic feedback systems, but they will be distributed and dynamic in a way that contradicts your premise of guided evolution by any coherent decision-making entity or governmental body.

        If posthumanity wants guided evolution for ethical reasons, e.g. to create sentience within a well-defined state-space of high hedonistic value, it will need to find ways to overcome such limitations. One potential solution could be the launch of one non-mutation replication algorithm optimized for this ethical goal. Of course, if we already have an evolutionary equilibrium of some sort due to undetected alien life, then this is a moot point.

  • anonymousfromUK

    Also, in general, coordination may get easier in the future. Posthumans may be very well-optimized for a “big” world. Beware of overgeneralizing from humans.

  • kevin

    Teeming, not “teaming”.


  • Tim Tyler

    “But if those mechanisms must be in place before substantial expansion begins from some central origin, that seems to require maximal development of coordination technology at a cosmically very early development stage.”

    It seems to be a dubious premise. Signals travel at the speed of light. They will find their audience. What is the problem? Are you thinking useful coordination know-how will be mistaken for alien propaganda by super-smart agents? That sort of scenario doesn’t seem likely to me.

    • Joshua Zelinsky

      But in order to coordinate they would need likely not only to have such messages but to have an exchange of messages. Note incidentally that we don’t see any evidence of even single sided communication broadcasts of this sort.

      • Tim Tyler

        Note, that I was discussing the general possibility of bioversal coordination – *not* the hypothesis that such coordination exists today. I think that latter hypothesis is without merit – but let’s not use arguments against that as evidence against the possibility of future large-scale coordination.

        Economists would like to use their toolkit to understand the future. However, if there is large-scale coordination, then many of their tools become useless. Not the greatest outcome for them, perhaps.

  • Tim Tyler

    Re: “for agreements to not use ample resources, there would remain great temptations to use such resources secretly, perhaps to support a breakout from the coordination regime.”

    You are talking about rebels against the galactic federation? Surely those have no chance – the federation will just crush them utterly.

    • Dan Weber


      Although, in seriousness, if the consortium’s rule is “use resources invisibly” it’s impossible to sneak around that, by definition. If you are visible, you aren’t sneaking.

      Super weapons might keep people in line. If the response to visibly consuming a tenth of a star’s output is that a planet buster gets sent to your world in a century, the permanent residents will enforce that rule.

      • Finch

        Re: Super weapons. It sort of depends on whether defense or offense is dominant technically, in the limit. Since we have little idea of the technologies involved, it’s hard to guess.

        But if it turns out that defending star systems is easier than destroying them, then hegemony will be hard to enforce.

        Conversely, if destroying star systems is easier than protecting them, the no-mutations got-everywhere-first entity would be hard to displace.

      • billswift

        >“use resources invisibly” it’s impossible to sneak around that, by definition. If you are visible, you aren’t sneaking.

        It is still unstable, if using resources more visibly gives you an advantage that would make you stronger than another group, especially defensively, where an attacker attacking across interstellar distances would need an incredible (likely impossible) disparity of force for success.

    • GNZ

      I suggest at its limits war is more a MAD (mutually assued destruction) solution than a conquest one. At present we have nuclear weapons and many consider this MAD.

      I imagine small robots that can engineer the conditions for somthing like a black hole to form if they get past the enemy defenses. Now send so many that it is impossible for the enemy to protect everywhere.

      Now your rebel planets have an ace up their sleeves if the galactic federation decides to come for a fight.

  • robotindemise

    If light speed is the only stumbling block, and we’re asserting a universe peppered with intelligent beings needing to coordinate, but haven’t detected them, how much should we adjust our belief that electromagnetic radiation is the only option for coordination?

  • Dan Weber

    On super weapons, I was thinking of the “existential threats” from the great filter. If some weird physics experiment can destroy a solar system, it can also be a weapon. A consortium looking to enforce conformity can quietly sail a little satellite with its LHC someplace, and activate it.

    Thinking more traditionally, it seems really easy to drop rocks on planets in an inner solar system, and really hard to detect it being done maliciously. It makes it hard to conquer an area so that you can populate it, but makes it easy to lord over.

    • Finch

      I sort of think offense is favored as well. For example, what are you supposed to do when the overlord tosses a black hole at your sun at 0.99c?

      But this might just be because we don’t know enough, and to beings a million years ahead of us the answer would be obvious.

      Maybe emergent civs looking to be stealthy could intercept their emissions and send only the emissions of a non-emergent civ to any monitoring points. Picture a big projector screen on the Earth-emergent civ axis playing a movie of a dead solar system. 🙂

    • gwern

      > Thinking more traditionally, it seems really easy to drop rocks on planets in an inner solar system, and really hard to detect it being done maliciously.

      It’s hard without access to high-tech, but I was surprised to look at the various arguments in (and especially the weapon/detector discussions in the spacewar section beginning with

      Apparently sensor technology is good enough that under plausible assumptions, planet-based defenders have huge advantages over spaced-based attackers – they can see the latter coming from almost indefinitely far off, can use far more powerful weapons (because spacecraft have serious heat-sink issues while planets are gigantic heat-sinks), and so on.

    • GNZ

      If the federation has the weapon at some point hte potential rebels have it too.
      What if the rebels attack first? (afterall they should know they are a threat before the federation does)

      You could potentially blast a buffer zone of dead systems, maybe millions of them.

  • Pingback: Overcoming Bias : Shulman On Superorgs