16
$\begingroup$

Basically, there are 2 types of hiveminds: the common centralized and the uncommon distributed.

Centralized means that all units are extensions of the master entity (the badguy alien/robot stereotypes).

Distributed means all units are extensions of the entire collective (closest example I can think of is: ants, bees, schools of fish).

Since time and again, the centralized type's master entity has shown to be the weakest link (B1 Battle Droids, Battle LA aliens, Ender's game aliens, Independence Day aliens, Skynet etc), why would it be used instead of a distributed hivemind type?

Seems like there is no purpose of having them with a centralized hivemind, except to give the human goodguys an easy way to win

Edit: No one answer is the best; many of them bring up good points.

After going through the answers, the advantages of a centralized hivemind are (summarized):

• greater/easier scalability

• capability for strategic planning/goals

• less resources and less processing power required per unit

• lower latency, higher speed communication

• higher reliability and security (but a double edged sword as the master entity(s) is now a critical vulnerability)

$\endgroup$
11
  • 6
    $\begingroup$ Yup, you figured it out. It's a trope. $\endgroup$ – user535733 Apr 15 '20 at 15:28
  • 8
    $\begingroup$ The Zerg in StarCraft have a centralised hivemind and were extremely efficient due to it. Do you think they would have been nearly the threat if they were decentralised? Because within StarCraft itself is shown time and time again that a hive cut off from the overmind is a nuisance but not nearly a galaxy spanning problem. $\endgroup$ – VLAZ Apr 16 '20 at 7:00
  • 7
    $\begingroup$ Minor query: how is the "master entity" any more of a weak point than Human armies having commanders, countries having leaders, et cetera. Films and books play it off as a massive flaw, because they need their heroes to win. If the hive-mind disconnecting turned the drones from a methodical disciplined army into rabid indiscriminate murder-beasts instead of just keeling over in a coma, it would be far less attractive an option $\endgroup$ – Chronocidal Apr 16 '20 at 10:51
  • 7
    $\begingroup$ Do you allow for magic, like e.g. the Ender's game aliens? If you don't at least have instant communications, the whole thing falls apart immediately. Heck, we're centralised hive minds (essentially every multi-cellular organism is), but we're still not in perfect conscious control of our "hive" - even though the whole thing falls apart if there's a successful "rebellion" (e.g. cancer). Presumably, even the most centralised hive entities would be very similar - having some direct and indirect control, and lots of things that are essentially automatic. $\endgroup$ – Luaan Apr 16 '20 at 11:33
  • 2
    $\begingroup$ Rather different take on this, but all the examples of a centralized hivemind being an issue that are quoted here seem to me to be just as much tales of how overconfidence and/or insufficient physical security are problems. $\endgroup$ – Austin Hemmelgarn Apr 16 '20 at 17:56

10 Answers 10

24
$\begingroup$

The swarm can think ahead.

In a decentralised hive every action is local.

  1. Foraging group picks up the scent of an enemy hive. They put down danger pheremones.

  2. Patrolling soldiers pick up danger pheremones. They identify the rival hive and put down muster pheremones.

  3. More patrolling soldiers gradually cluster around the muster point.

  4. When the soldiers reach a critical mass their passive attack pheremones (which are always active) cause the platoon to follow the trail of enemy pheremones.

  5. The soldiers follow the trail for 400 metres and find nothing at the end. Huh?

  6. While the soldiers are occupied, the enemy hive (which has a centralised intelligence) attacks.

The centralised intelligence of the enemy hive allowed it to make this trick. It would be much harder to do this only acting locally.

Edit: The main weakness of the centralised mind is, if the queen dies, the whole swarm is left mindless.

HOWEVER: There is no reason you have to choose one or the other. Imagine an ant-like hive mind that can function on it's own, but also has a conscious leader who can (a) issue broad commands and (b) assume direct control of individuals or groups of drones.

You get efficient small-scale behaviour but also the potential for long term strategies. Also if the queen is killed the hive has enough instinctual intelligence left over to create a new one. So yes, the hive is rendered stupid for a while, but it's not an automatic win for the enemy team.

$\endgroup$
16
  • 5
    $\begingroup$ Since humans lack a centralized hive mind and must make decisions locally, individually even, but still exhibit ability to make non-local decisions and coordinate activities globally it is thus fairly obvious that we are being controlled by a secret centralized hive mind. Because clearly we would be unable to do those decisions and activities ourselves. Assuming your argument is valid, of course. Seriously, distributed hive minds are not any more limited to local communication than we are. Realistically less so. $\endgroup$ – Ville Niemi Apr 16 '20 at 9:19
  • 1
    $\begingroup$ @VilleNiemi: It would be helpful for you to say what you think of when you hear "hive mind". $\endgroup$ – Daron Apr 16 '20 at 14:59
  • 1
    $\begingroup$ @Daron Why? I am not the one asking the question. All I am saying is that you are assuming something exactly like ants without there being anything in the question to support more than superficial similarity with qualifiers. I have no particular opinion on how the hive mind should work and see no reason why you should care even if I had. I am simply pointing out that the question was not limited to discussing ants and your answer is. Well, it probably covers other social insects too. $\endgroup$ – Ville Niemi Apr 16 '20 at 16:40
  • 3
    $\begingroup$ I remember a short story about an alien hive mind that acted decentralised (and in fact downright dumb) almost all the time until deviations from normal hive behaviour causes the spawning of an ‘intelligence’ that would take control of the hive, fix the problem and then die, leaving the dumb hive to carry on until the next problem arose. $\endgroup$ – Joe Bloggs Apr 17 '20 at 9:48
  • 2
    $\begingroup$ @JoeBloggs it's the short stroy "Swarm" in the Shaper/Mech Universe by Bruce Sterling and pretty cool. $\endgroup$ – mart Apr 17 '20 at 11:23
9
$\begingroup$

The client-server model has better scalability than peer-to-peer.

Simply put, as you scale a peer-to-peer model, the number of connections that must be maintained scales according to a Big O of N^2, since each node needs to be able to connect to every other node. By contrast, with a client-server architecture, the number of connections scales according to a Big O of N, since each client node connects to a central server node.

While this may not matter much for small-scale networks, once the number of a nods begins to increase, this becomes more and more noticeable - once you've got a million nodes, adding one more node requires another million connections for peer-to-peer, while it only requires one additional connection for client-server. This would result in the peer-to-peer network using considerably more bandwidth than the client-server model.

$\endgroup$
7
  • 17
    $\begingroup$ Um, no. That is not how P2P scales beyond small scale. It is how some games do it, which is probably where you get this? But once you start scaling up, the difference between "being able to connect" and actually "being connected" becomes critical. No P2P system designed to handle large scales maintains constant connections to all nodes. In fact, many have no constant connections at all with all connections done as needed. This explains the basic idea : en.wikipedia.org/wiki/Kademlia $\endgroup$ – Ville Niemi Apr 16 '20 at 9:32
  • 3
    $\begingroup$ Is it any easier if we just connect each node to the nearby nodes instead of all the others? You can get good flocking behaviour doing just that: BOIDS! (youtube.com/watch?v=QbUPfMXXQIY) $\endgroup$ – Daron Apr 16 '20 at 17:39
  • $\begingroup$ If each peer just connects to nearby peers then you can't reliably do any planning that relies on knowing the big picture, because none of the nodes have the big picture. $\endgroup$ – user253751 Apr 17 '20 at 11:52
  • $\begingroup$ @user253751 , messages can travel from person to person without a direct link. You can argue about the efficiency/quality of this, but it is still definitely possible for a unit to have a "big picture". $\endgroup$ – cowlinator Apr 17 '20 at 19:20
  • 1
    $\begingroup$ Funny enough, I think it is exactly the other way around: P2P scales much better. $\endgroup$ – Peter - Reinstate Monica Apr 18 '20 at 14:34
9
$\begingroup$

Latency (and reliability)

Depending on the mode of communication your swarm uses, the physical act of communication may be the major bottleneck. It's all well and good if the individual drones in your swarm have enormous combined processing power and storage capacity, but if every operation requires time-consuming and potentially unreliable communication at every step, it may quickly become untenable to do things that way.

In a centralized model, every percept the system receives and every action the system chooses to take only requires a single transmission (or perhaps a couple if they're short range, to forward it to the central unit). In a fully distributed model, every 'thought' the swarm collectively has could require dozens or hundreds of redundant transmissions just to locate the relevant bits of memory.

Security

You cannot secure every drone in your swarm. Some of them will be captured by the enemy. If they remain capable of contributing to the swarm's collective thinking process, this may present your enemy with an avenue of attack by which they could influence the decisions the swarm makes. Strictly keeping all thinking in a central control unit (whose physical security can be guaranteed and which can show all the steps in its thought process upon request) could help mitigate this risk.

$\endgroup$
7
$\begingroup$

In the long term, a distributed network (either distributed hive or autonomous actors) will always outperform a centralized hierarchical organization. This was first described by F.A Hayek as "The Local Knowledge Problem".

Basically, it recognizes that information is widely distributed across large systems. Local actors can see and act upon the local bits of information immediately, while centralized networks need the information to be gathered, processed, sent up the chain to be analyzed, orders made and passed back down to be executed. This sets up a delay, and orders can be received after the local conditions have changed, meaning they are no longer valid for the situation.

This is the "best case" scenario, and even suggesting computers can be used ignores the extra resources needed for sensors and bandwidth, while still not overcoming the latency issues. Add mistaken information, faulty analysis, or malicious corruption of information along the way to derive local advantages, and the conditions exist for a positive feedback loop of errors to accumulate.

Looking at some of the more spectacular failures of central command economies, such as the 1930 era "New Deal" extending the Great Depression by as much as seven years, the collapse of the Soviet Union or Venezuela today, or even comparing the sluggish socialistic economies of Europe to the United States (most European national economies have lower GDP/capita than all but two American State economies), it is difficult to argue with Hayeks observations.

If the loops are small, or there is an emergency situation, a command economy or other centralized system does have short term advantages, being able to mobilize resources faster than networks or meshes, but the possibility exists that this situation results in mal investments, which accumulate and eventually cause frictions or failures (such as economic busts, or ecosystems being overrun by invasive species or algae blooms).

Depending on the environment, meshes are more flexible and adaptive, while hierarchies are better with emergencies and extreme environments. A centralized hivemind can mobilize more quickly to react to emergencies, while a mesh might have the emergency situation "flow through" it, and rebuild the broken links in the wake of the situation. As several comments in this thread have noted, "real life" insect colonies actually use a mixture of techniques (like ecosystems and economies) to provide the flexibility to deal with both changing local situations and emergencies.

Edit to add: In modern military organizations, "mission command" has been adopted which makes use of the Local Knowledge Problem model. Senior commanders provide the overall mission guidance and resource allocation, but subordinate commanders are expected to respond to the local situation in the most appropriate manner consistent with the overall desired end state. In other words, everyone from a platoon commander on up is expected to understand the local environment and react to the available information there to achieve immediate advantage without waiting for higher command.

$\endgroup$
4
  • $\begingroup$ Obvious counterexample where it doesn't work: do you want local commanders to have strategic nuclear weapon firing authority based on what they think they know is happening? $\endgroup$ – Keith Morrison Apr 16 '20 at 16:24
  • 1
    $\begingroup$ Also, calling the "socialistic economies" of Europe "sluggish" is betraying a particularly conservative American viewpoint that is not necessarily agreed with or, contrary to your assertions, actually proven. The idea the New Deal extended the Great Depression is pretty much only promoted by right-wing American individuals and organizations who are pathologically and reflexively opposed to the idea of government intervention in any way, shape, or form. $\endgroup$ – Keith Morrison Apr 16 '20 at 16:35
  • 2
    $\begingroup$ @KeithMorrison No serious European economist disputes that the European labor market (for example) is less responsive to changes in local conditions than the US labor market. The standard argument is that issues of (so-called) "fairness" are more important than a speedy response, and/or that a "stickier" labor market produces other indirect benefits. Those counterarguments don't disprove the local knowledge problem as much as they declare that policymakers simply don't care what outcome is better for a particular system node. $\endgroup$ – tbrookside Apr 17 '20 at 19:13
  • 1
    $\begingroup$ As a reply to the comment about nuclear weapons, in the 1950's tactical nuclear weapons were to be used at the discretion of much lower level commanders. Strategic weapons are under national control, but also have very short information loops between the weapon and the releasing authority $\endgroup$ – Thucydides Apr 19 '20 at 23:16
7
$\begingroup$

The army dying when the leader dies isn't a flaw, it's a feature.

There are ways to have a hive mind without the threat of collapse. You could have multiple backup redundant hive minds, and have one of those set up again whenever the humans killed one. But that comes with a serious flaw.

Your subordinates now have an incentive to kill you to take the swarm.

Your main enemy is your subordinates, not outsiders.

You would have either a distributed hive mind or backup hive minds if you wanted combat efficiency. It makes sense to be prepared. But, if your main priority is yourself, successors are simply a weakness. They mean that by killing you, someone else can gain power. That means you personally lose control.

An outsider who wants a more pliable swarm can kill you and ascend a pliable subordinate. Either way, you lose.

It decreases the swarm's vulnerability to hacking.

If every subordinate has some degree of potential to run the swarm, then a hostile entity can upload code or commands to enough of them and take over the entire swarm. A centralized entity can have powerful anti hacking and anti psychic protections that lowers the risk of a takeover. That's much cheaper than giving every entity strong protections. The centralized entity can also shut off rogue swarm members who are compromised.

The borg have this. There are multiple efforts to use borg drones to upload borg destroying viruses. The queen reduces this risk, by being able to shut down cubes.

It allows more suicidal tactics

A swarm where every creature has a voice will not normally sacrifice more than 50% of the swarm- each individual has an incentive to survive, and so even if they agree, they can at most form a group to sacrifice 50%. A centralized hive mind has no such constraints, and can sacrifice all but one of it's swarm. This allows them to fight harder, and take valuable resources a more distributed swarm could not.

$\endgroup$
4
  • $\begingroup$ I don´t think the last point is necessarily true. It assumes all member only care about themselves, while in reality they can do things just to help e.g. their next of kin. If I had the choice between humanity going extinct, or 60% dying, I´d choose the latter even if I knew I was in the 60%. And that´s assuming you knew beforehand who was gonna die. $\endgroup$ – Emil Bode Apr 17 '20 at 17:55
  • $\begingroup$ Sure, if it's a choice between death and life they may sacrifice themselves, but a swarm can sacrifice 51%+ for economic advantage or technology or other things that aren't about immediate survival. Would you be ok with you and most of your family members dying if you knew several generations later you could have greater growth because of it? Would you be willing to do so quickly, without debating about who should be sacrificed? $\endgroup$ – Nepene Nep Dec 16 '20 at 18:12
  • $\begingroup$ No, but I don't think a hive mind would be willing to either. Would you sacrifice an arm or a leg for greater growth several generations later? $\endgroup$ – Emil Bode Dec 17 '20 at 19:28
  • $\begingroup$ Sure, so long as there were no other enemies nearby to take advantage of the weaknesses. Many hive minds have extremely rapid growth. $\endgroup$ – Nepene Nep Dec 17 '20 at 21:56
6
$\begingroup$

Compare democracy with a "martial law" government.

In a true democracy, every action should be voted by the individual members which may lack the information to make the correct choice. Even if the choice is ultimately correct, this whole process takes a very long time.

In a centralized, non-democratic government, every decision can be made very quickly, and information channels are designed to feed the decision makers. The obvious downside is that those decision makers are small in number, and eliminating them would be a potential lethal blow to the organization.

Similarly, in a hive mind every individual element lacks the whole set of information, and either has to make blind decisions, or delegate those decisions to someone else (central command?).

$\endgroup$
17
  • 2
    $\begingroup$ The economist F.A. Hayek demonstrated that the hierarchical "command" economies are always outperformed by market economies. This is called "The Local Knowledge Problem", and is predicted on the idea that information is distributed widely across complex systems. A local actor can see and act on the information right away, while there is a time delay in gathering and moving information up the chain, analysis, orders and transmission of orders. Further corruption is caused by misunderstandings or malicious interception of information for local gain. $\endgroup$ – Thucydides Apr 15 '20 at 20:34
  • 5
    $\begingroup$ @Thucydides there is local action, and there is strategic action. We may have nearly total local success, but one strategical fault may become insurmountable. Economic development is a realm that is quite different from a military action. $\endgroup$ – Alexander Apr 15 '20 at 20:43
  • 3
    $\begingroup$ I would argue that Hayek theorized that, rather than demonstrating it. Beyond that, a hivemind would know everything at all levels, including the local, so that's not a reasonable refutation of the strength of a hivemind over the pseudorandom behaviour of a "market" economy. $\endgroup$ – jdunlop Apr 15 '20 at 20:43
  • 3
    $\begingroup$ @Thucydides, you run your military entirely on Hayek's ideas, I'll run mine on classical methods. By the time you figure out what the hell is going on, I'll have defeated you in detail. To use a trivial example, a unit may be required to perform an action where it has to sacrifice itself which, based on its Local Knowledge, makes no sense whatsoever, or perhaps take no action even though logically, based on what it knows, it should. $\endgroup$ – Keith Morrison Apr 16 '20 at 16:21
  • 2
    $\begingroup$ @Thucydides Imagine the COVID-19 response with only local knowledge. Everyone would do nothing until someone they know got infected. Not good. $\endgroup$ – user253751 Apr 17 '20 at 11:52
4
$\begingroup$

Avoiding Value Drift.

A distributed intelligance works by emergant thoughts held by the interactions of the agents. If you seperate two groups of the agents entirely then you end up with two hive minds. They are exposed to different stimulus, so think more and more different thoughts.

This still happens even if not totally seperate. Once you spread out enough, and become large enough you get latency. Information can not travel faster thn the speed of light, and even once you occupy just a single solar system that starts to really add up. Even if you do have faster than light information transfer (which is useful to make a centralized hive-mind possible too), the possibility of being cut-off remains.

The left-hand does not know what the right is doing. Worse, some other hands (finger) might be too small (and thus dumb) to realize that they are at risk of value drift.

You can eventually become your own enemy.

A centeralized highmind does not run this risk. It has its own latency problems and may not (without FTL communication) even function over large areas. But it does not run the risk of becoming its own enemy. If communication with agents is lost or becomes unfeasible, they stop being intelligently driven. They don't form local clusters with their own emgerent values and goals.

This is a theme of the last stage of the game "Universal Paper Clips". Your greated foe is (former) parts of yourself, who've values have drifted out of alignment with your own.

$\endgroup$
4
$\begingroup$

Centralized is easy.

That's probably all it comes down to. A fully distributed network is far more complex to design and develop and often infinitely more complex to understand than a centralized one.

For a centralized hive mind, all you need to understand is how the central brain works, and you can predict what it'll do. If it doesn't do that, you know where the problem is coming from and where to look for a fix.

If you need to make some design changes, you only have one place to manage them. All the drones in the network are stupid; they just do what they're told. If the central brain can be taught to deal with the problem, the drones can probably stay as they are. If they need to be changed, they can be tested in isolation in a test chamber.

If you build a new type of drone, you just need to update the central network to be able to command the new drone type and to let it interact with the others, if necessary. The new drone, again, can be built and tested in an isolation chamber; as long as it does the right thing when instructed and feeds the right info back into the central hive, it's going to work perfectly well with everything else.

When you're dealing with a distributed network, nothing is easy. All your drones are self-operating, so it's hard to test them in isolation. Even if you build a full environment for one of them to play around in, all you'll know is that the drone does a certain thing when alone. You have no idea how the drone will act when it encounters a fellow drone, because not only will your drone behave differently now, but the other drone you introduce it to will also behave differently. And each type of drone will probably have a different type of reaction. And having more than one drone at once will also change what is happening. Even different if it's a different combination of drones. It's hard to predict what the various drones will do.

Whenever you want to change a distributed network, you need to not only think about the main network and how to modify it, but you need to think about how all the different drones are going to respond to new instructions.

If you build a new type of drone, not only does the main network need to deal with that, but all the other drones need to learn to interact with it in a useful way. And probably those drones are pretty limited in terms of thinking power, so good luck with that.

Using primarily centralized intelligence, we can drop human beings on the moon with all the computing power of a modern calculator. Using distributed intelligence, we can only now barely produce something capable of navigating a quiet road and we're still terrified of letting it do that around actual people.

Sure, having a fully functional distributed system that perfectly functions will beat a centralized system with absolute ease. But there's a good reason humanity invented the stone hammer tens of thousands of years ago, but the self-swinging-hammer still isn't a thing.

$\endgroup$
3
$\begingroup$

The efficiency of a centralized system depends a lot on the speed with which information is transmitted and the processing potential of the central node, as well as the ability of the system to reshape itself in the event that the central node is removed. Generally this last achievement can be achieved through backups, in robots this could mean each robot has the potential to be a central node in the system, with biological organisms you'd likely see multiple competing hive's or otherwise have backup rulers that lie dormant till needed. You probably wouldn't want all of your eggs in one basket, but let's assume that's the only option for a centralized system, why might you choose that over a distributed system anyways?

In some circumstances centralized systems can be much more resource efficient and can be far superior in terms of decision making. With a decentralized system each node is only aware of it's surroundings and the signals of others, in a perfect centralized system the central node can be aware of all of it's drones surrounding's and can formulate complex strategies for the hive as a whole and execute them with perfect precision. A distributed network is a cheap knockoff of actual omnipresence and relies on signals propagating through a network of multiple intelligence's, but with enough processing power a centralized network can be the real thing. Essentially a single god like entity in total control, similar to plugging our nervous system into a nations infrastructure. However, if information moves through the network too slowly or the central node can't process all the information efficiently, the system loses it's edge.

Beyond that regard, it's also potentially a waste of resources to have every drone capable of thinking, it's especially wasteful if each drone is highly intelligent. If you assume that each drone is intelligent and communicates with every other drone the processing required will increase exponentially with each new drone, whereas a central system only has to send the information once for each drone, then send a response. A caste system addresses this somewhat as a hybrid system, but is still not as ideal as having a mind capable of correlating all the information of the hive at a quick pace. If it's instead a simpler form of communication that creates emergent intelligence, that could be more resource efficient, but is not as likely to manifest in the same level of intelligence, nor be as adaptable as a conscious entity.

Given very fast information processing, a centralized network can get away with using less resources per drone and less processing per drone by potentially an exponential amount, by keeping the communication at a linear rate per drone. This allows for further expansion of the hive. So if a hive is primarily concerned with growth, centralization probably allows for the greatest growth per unit of resources until the distance become too great for rapid communication. So the answer is that it's only the best option when you have the necessary processing capabilities and a means to send information almost as fast as the nervous system because omnipresence and super intelligence is a very powerful combination that might outweigh the risks.

$\endgroup$
3
$\begingroup$

Obviously a lot depends on the exact architecture you have in mind. In real life, insect swarms are largely decentralised and have a lot of 'clever' adaptive behaviors which work in distributed fashion.

However, there may be biological reasons to have a centralised approach. While some ant colonies may have hundreds of breeding queens (and some do not), many honey bees have a single queen per hive. The reason for this is supposedly that it speeds evolution. In this situation, if the whole hive is dependent on pheromones or other signals from the queen, you have a single point of failure.

As with the queen been, it may be a matter of resources. Distributed brains may be limited. If you only have one super-duper-battle-computer, you may want to put that at the centre of the network so that the whole swarm has access to its computing power. This might be seen as a centralized approach.

However, in real life single-queen honey bee hives are very successful, and can replace the queen rapidly when lost. The trope of 'kill the central master and win the war' appeals to human storytellers but it's very much a B-movie approach.

$\endgroup$
3
  • 1
    $\begingroup$ How much of a single-queen hive is controlled phenomenally by the queen? I can easily believe parts related to egg-laying (build more brood cells please) are controlled by queen pheremones. One the other hand it's clear that most of the hive is not controlled by the queen, as the hive can replace a queen if needed. $\endgroup$ – Daron Apr 15 '20 at 15:39
  • $\begingroup$ A quick Google suggests the Queen pheromones are important for "stimulation of worker activities: cleaning, building, guarding, foraging, and brood feeding" ncbi.nlm.nih.gov/books/NBK200983 $\endgroup$ – David Hambling Apr 15 '20 at 15:43
  • $\begingroup$ on the second paragraph second sentence your statement is misleading, many ant colonies may only have a few or even only one queen, some do have large amounts but most do not. your common fire ant hive will most likely only have one. most colonies on there own will not have hundreds. $\endgroup$ – Topcode Apr 16 '20 at 2:22

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.