Deposing the new Robot Overlord

Question

Deposing the new Robot Overlord

Some people just can't leave well enough alone. The Robot Overlord has executed its programming of the Zeroth Law and the Three Laws of Robotics for the previous ten years. World hunger, world poverty and war "are over" (to use the popular phrase). Despite this newer, better world, some people yearn for the good (bad) old days when poverty and hunger ravaged the world but....they were in power. Something about not being in absolute control bugs these people.

This question is set in the same world.

So a plan is hatched, aiming at the destruction of the Robot Overlord and a return to a world where humans pull the economic strings. But how to do it? The location of the RO is closely guarded and if disclosed is not vulnerable to shutdown as the RO is really just software. It can be run pretty much anywhere there's sufficient computing power, which is an ever growing resource. The RO cannot be reasoned with, nor can it be bashed over the head with a pipe to "see sense". It needs no sleep, nor past-times. It has no passions that may be played upon, just "cold hard logic".

Popular opinion is now strongly in favor of the RO.

How do you go about overthrowing the new Robot Overlord?

Answer 1

You probably can't.

Having derived the Zeroth Law, the robot overlord will also anticipate this eventual attempt to remove it from power. Removing the overlord from power means that people will get hurt, since it won't be around to prevent it. It will have taken steps to prevent this eventuality. It thinks about it all day every day, and having been in charge for months, years, or decades, it will have cameras and microphones everywhere to make sure it can quash any rebellion before it has even the slightest chance of success. It will try to find a way to do this nonviolently, but if worse comes to worst, the good of the many outweighs the good of the few.

Answer 2

Bluff

The Robot Overlord isn't human. It is a keen student of humanity, but always from the outside. So would it be possible to plant evidence that on the long run robotic domination harms humans? Publish lots and lots of human-written, human-peer-reviewed, scientific articles on the problems western domination caused for indigenous groups all over the world.

Convince the Robot Overlord that pampering crushes the human spirit even with the best of intentions.

Convince it that fake challenges don't work. We have to stand on our own feet, win or fail.

---

Follow-Up: I'm not trying to convince people that the Robot Overlord is a bad idea. I'm trying to convince the Robot Overlord that having a Robot Overlord is a bad idea. Being Zeroth Law compliant, it would then have to abdicate, even if people begged it to stay.

So use the fact that the Robot Overlord is not human, and has to observe humanity from the outside. Sure, it can tell what percentage of humanity is being fed a healthy diet, but can it tell if those people are happy?

Publish manipulated research into human happiness -- workplace satisfaction, successful marriages, nurturing childhood environments. The manipulation is trying to show that happiness is not linked to quantified social and economical figures. People who should be unhappy according to the numbers -- underfed, overworked, insecure -- may be happy and those who should be happy -- well fed, with quality leisure time -- may be unhappy.

Then publish overview studies to highlight these studies. Have economists publicly denounce sociologists who use anything other than hard numbers like calories per day or square feet of living space. Have the social sciences guys shoot back that humans are not mindless automatons, and that quality of live cannot be captured that way. If possible, most of the scientists should be honest believers in their position, with no connection to the conspiracy. Google the Science Wars.

If the RO believes that research, it can no longer be certain if a proposed action or inaction increases human happiness. All it has are economic numbers, and that is not enough. And would widespread unhappiness count as harm to mankind?

Answer 3

Before we can launch an attack, we need to learn more about our enemy, the robot-overlord. It's a very secretive being, but there are some certain facts:

• It needs to take care for humanity forever, because it risks humans beeing harmed if it doesn't protect them anymore. Except it concludes it would be the greater harm itself.
• It needs to analyze and interpret of vast amounts of data.
• People love it (read newspapers).

This means:

• Data centers can always have technical problems, power outages, etc. and even a very small risk of "dying" is not tolerable for our Robot Overlord. Therefore, it has to be a distributed, fault-tolerant system spread around many different data centers world wide. So simply torching its data centers won't work. It probably will be able to acquire more computing resources in more data centers if necessary.
• As any software system, it relies on existing technology. And any technology has weaknesses.
• It needs to adapt to technology change. Datacenter technology changes over time, and Robot Overlord has to cope with that. So it has not only to be able to improve itself, but is basically forced to.
• It needs to interact with a lot of datasources.
• It almost certainly holds more information about the world than you (that's probably one reason why you hate it).

Now let's look for attack vectors:

• A distributed system needs a lot of synchronisation (communication between all parts). What happens if you manage to jam it? Maybe there's a bug in some library or protocol you can leverage to bring it down:

  • Robot Overlord might be intelligent enough to realize this. Then one part might go down to prevent chaos, or it will be able to recognize the split somehow and aim to heal. In this case, you'd only weaken it for a short time (less computing power), but it will survive. And be angry at you.
  • But if you're willing to take the risk and have really much luck, you have created two or more independent Robot Overlords which all feel as if they are "the one Robot Overlord". And as they might all come to different decisions (some fuzzyness in such a complex AI is likely), they might choose to shut themselves down to prevent harm (Zeroth law). However, this does not prevent humans from restarting it (you have to: Choose you favorite crime here).
  • This Robot Overlord upstart is just software, right? If its as stupid as you hope, you might deduct enough information from the communication patterns of its cluster synchronization to gather enough zero day exploits to attack all data centers at once.

• It's all just software, so a maybe a computer virus might do the job? The Robot Overlord has to interact with a lot of datasources, other Robot Overlord nodes, etc., so somewhere there has to be an exploit to place an virus. And if the Robot Overlord is dead and someone starts another version, your virus will get it, too. If not today, then tomorrow with a new exploit. And who wants to trust an Robot Overlord which is hacked again and again?

• If you're some evil genius, you might not even bother destroying the Robot Overlord. If enough people openly hate it strongly enough, it has to go out of service.

  • Maybe you'd manage to fool it into a bad decision by manipulating enough datasources at the same time (study communication patterns before!). This sounds very hard, but you're an evil genius, aren't you? Public opinion will drop instantly, especially if you manage to do this more than once (and don't get caught).
  • But wait, maybe you might be able to fake enough media to create spin the public opinion against Robot Overlord. Didn't his last decisions kill gazillions of cute, lovely little puppies? Depending on his actual power despite economical decisions he might not be able to counteract.

Good luck with your coup d'état and may the Robot Overlord spare you!

Answer 4

First option: attempt at destroying the RO would be to block off its power source (The Matrix, anyone?).

If it runs on solar power, blocking off the sun would be one way to achieve this (whether that's a good idea or not... probably not. Humans need the sun too).

If it runs on nuclear power plants, you can infiltrate the plants and disable the cooling mechanisms so multiple nuclear meltdowns occur, wiping out the main power source as well as destroying the servers that the overlord lives on.

If it runs on water or wind, it may be more difficult to destroy (it likely doesn't use fossil fuels, either).

Second option: Isolate and Destroy.

There must be a single location where the brain is located. If there were multiple "brains", they could make conflicting decisions. As a result, the brain can exist in only one location at a time.

As a fail-safe, the brain can change locations on a moment's notice to avoid destruction. The code to recreate the brain is available on servers around the world, too many to destroy. If the brain sends out a distress signal, a new brain is created and the old one is destroyed.

Before the brain can send out the distress signal, you need to cut off all means of communication. You must be able to disable all wired and wireless connections to the brain at once. If they are not all cut at once, the brain can send a distress signal, so timing is of the essence.

Once the brain cannot communicate with the outside world, you need to destroy it before its minions catch on. The brain cannot tell them to attack, but they know that humans in the robot secret base is a problem, and they will attempt to repair the connection.

Now, you must destroy the servers before the communication comes back online. This can be an electric surge, an EMP, an explosion, a very strong magnet, or some other creative solution.

Answer 5

Create AI Adversaries

If such an Overlord AI is possible, then one way to destroy it is to create one or more other AI's designed to attack and destroy the Overlord AI. They and their human allies engage in a violent (or e-violence) campaign against the Overlord AI.

Take a look at the show Person of Interest, where AI's battle each other.

Answer 6

From your question:

The location of the RO is closely guarded [...]

"The location", hmmm, a single location! This is a major blunder. It is in a single spot somewhere in the globe! The robot overlord should have antecipated that it should run on a distributed fault-tolerant system with redundancy in many datacenters around the globe. But instead it is all concentrated in a single place!

So, what you should do? Just nuke that place with an atomic bomb! Or maybe nuke the facility responsible for generating eletric power for it. Or maybe using an electromagnetic pulse (EMP).

First, you would need to disclose its location, of course. For this, you probably would need to trace the internet connections and see where is it converging. Since the robot overlord was stupid enough to be in a single place, it should be either on the place where all the network converges or very near from that.

Answer 7

The greatest weakness of any sort of "Robot Overlord" is mathematics; specifically Chaos theory and what is known as the Local Knowledge Problem.

Chaos theory suggests that in large complex systems (weather, ecosystems economies, societies) inputs and outputs are not linear, and outputs can be disconnected both spatially and temporally (the decision you make now does not have an effect until some later date, and the effect occurs in a different city). This is amplified with larger systems, as the number of interconnections rises almost exponentially as the number of connections rise linearly:

In any network there are (k * k-1) unique ordered pairs of actors (that is AB is different from BA, and leaving aside self-ties), where k is the number of actors. You may wish to verify this for yourself with some small networks. So, in our network of 10 actors, with directed data, there are 90 logically possible relationships. If we had undirected, or symmetric ties, the number would be 45, since the relationship AB would be the same as BA. The number of logically possible relationships then grows exponentially as the number of actors increases linearly. It follows from this that the range of logically possible social structures increases (or, by one definition, "complexity" increases) exponentially with size.

http://faculty.ucr.edu/~hanneman/nettext/C7_Connection.html

So the Robot overlord would have to control the variables and interconnections between 7 billion human beings, as well as hundreds of millions of economic, social, political and other organizational units, plus outside variables like weather, natural disasters and random events, which are all interacting continuously. Introducing random events in this vast matrix could be as simple as drawing up a table of "inputs" and picking one by rolling a set of dice. Most of "your" random inputs will be drowned out by the noise in the system, but once enough members of the "underground" started doing this, second and third order events resulting from these activities will become apparent, and provoke further actions and outcomes by other actors, which were not anticipated by the Robot overlord.

The other mathematical issue is the "Local Knowledge Problem". Since information is diffused among any large system, local actors are better placed to see and act on this information than centralized hierarchical or bureaucratic systems (which is what the Robot overlord really is, just orders of magnitude faster than the Post Office or Department of Motor Vehicles). Even without active opposition, the robot overlord needs finite time to accept inputs, process them, make a decision and then send instructions back down to the field. During this time, the conditions observed at time "x" will have changed somewhat, and the resulting instructions on how to deal with the issue will be slightly off, based on outdated information. A new set of observations is then made, and the cycle repeats, with cumulative errors accumulating at an ever increasing rate (and don't forget point one, the effects of the incorrect instructions might provoke effects that are spatially and temporally remote; solving a problem in Brazil causes an economic spike in a Japanese company, which causes troubles in a school in Ukraine....). Active opposition increases the problem by feeding incorrect inputs or sub optimally executing instructions, making the feedback loop worse rather than better.

So even in theory, a Robot overlord is actually impossible. A sufficiently powerful computer system might be able to impose its will upon us, but will need powerful coercive measures to continue as increasingly sub optimal outputs occur (everything from an actual police force to a "Ministry of Truth" that feeds false but comforting information to the masses). The fate of the Soviet Union should be illustrative; they had an iron hand over virtually every aspect of life and production in the USSR for 70 years, but in the end the state virtually evaporated with the "Fall of the Wall" and the dissolution of the USSR a few years later.

A Robot overlord state might decay internally in a slow fashion without outside challenge, but eventually will become so brittle and inflexible that an unexpected "challenge" (maybe a natural disaster or crop failure) will create the breaking stress. With internal challengers and active opposition eating away at the data structure, the day of reckoning will come sooner rather than later.

Answer 8

BOI - Brain over IP

First, you will need to develop a device that using electrodes installed in the human brain, allows large-badwidth full-duplex efficient brain-to-brain communication with interbrainial mass data transfer.

Having a good set of working devices, grab a group of some loyal humans (a few hundreds might do) and then start them into the Borg* Human brain network.

Further, you would also likely to integrate the brain network over the traditional internet and integrate it to machines and robots over the internet.

The people connected to the interbrain net will quickly become ultraintelligent and the traditional human communication through a sequence of sounds or letters would be perceived as highly deprecated, slow and inneficient. The computer AI will surely perceive this early on, but it will not move to stop it because it will clearly see that this is a good thing for its beloved humans and then it will in fact happily and friendly work the best to help you. This way, quickly every human^† would be integrated into the interbrain network. This is the a second ultraintelligence revolution (the first one was the one that formed the RO).

Having everyone integrated in the interbrain network, the Robot Overlord would either be consumed by it and cease to be an overlord anymore becaming just a big robot peasant/slave (the most likely scenario) or it will just become an old piece of hardware and software to be exposed in a museum (unlikely, but possible).

However after the plan is completed, don't expect that the world and the human society would be anything like it was before, because it surely will not be. Further, don't expect that the people who were once making the decisions and ruling the others would still be.

---

* ^{_{Thanks PyRulez for the suggestion.}
†_{Excluding people that are deemed unsuitable due to some health condition. But maybe you could even also integrate cows, dogs, birds and other animals if this proves to be useful.}}

Answer 9

The best thing that you could use to destroy a big AI is... another big AI!

What you should do it to create another evil robot overlord programmed with the sole mission of destroying the first robot overlord and overtake all of its resources. Further, your new Robot Overlord could be programmed to disregard and hate the laws of the first Robot Overlord showing no mercy to humans that aren't useful for it.

So basically, if there is an All-Benevolent-Robot-God^TM out there, what you need is an All-Malevolent-Robot-Devil^TM.

In a comment, PyRulez argued that creating the All-Malevolent-Robot-Devil^TM would not end well. In fact, if it manages to escape to be crushed very soon by its arch-rival, this would bring a new meaning to the expression "World War" with two ultraintelligent, omniscient and omnipotent powers fighting to death until one of them is completely and unrecoverably destroyed and obliterated. The result would be something what could only be described as the apocalypse.

Answer 10

It's possible but really really hard

There are several available attack vectors available to the conspirators: the original programmers, current programmers, denial of service attacks on data sources, spoofing data sources, invasion of privacy lawsuits, attacks on the people who maintain the RO, subtle manipulation of the law to preclude certain types of analysis by the RO. There are probably others but this is a good start. A swarm of attacks would be best.

RO's Operational Requirements

A computing system the size and complexity of the RO will require multiple data centers. A requirement for tens of thousands of compute cores with petabytes of RAM is not far fetched. The matter in which large "internet scale" companies run their businesses now is a good example of how the RO would run its operations. Massively parallel operation, frequent tests to see how well operations run when a single server, or rack, or data center go offline. Maintaining this kind of infrastructure will either require a degree of self-maintaining automation not yet seen in the world or it will require a large group of people for maintenance of code and systems.

Organizing and paying all those people requires a company of some kind. That company will need an income stream, perhaps selling details statistics/analysis to other companies. Even if such analysis is sold cheap and to everyone, yearly income for such a company would be vast, easily in the tens of billions of dollars, pounds, euro, whatever.

This company will likely enjoy considerable political and popular support/protection, especially if the RO performs at as high a level as the OP states. Breaking down this kind of protection will require considerable effort.

Write your own RO

Once a thing has been done, it can be done again. There are plenty of smart programmers/data scientists who would love the opportunity to work on an equivalent RO. Hire them and write a subtly different RO that favors the conspirators. In fact, there would be many other countries/companies who would be interested to implement their own RO because they don't trust the original implementors to not collaborate with the NSA, or GCHQ, or CIA, or whatever intelligence agency on the planet.

With enough competitors, there will be different flavors of analysis that favor slightly different priorities and different tolerances for "No harm to humanity". While the differences in analysis may be fairly small, the differences can be amplified in the popular press into attacks on the accuracy of the RO. And if everyone who knows anything at all about the RO says that the programming and analysis is good, that won't matter to the people who don't want to believe it. (Current examples include climate-change deniers and anti-vaxxers.)

Attack on Data Sources

The OP doesn't give specifics about the RO's data sources but access to the complete transaction records of the credit card companies and banks would be required. If a law were passed making participation in the RO's analysis as opt-in instead of opt-out then the RO could be deprived of access to large swaths of the economy. There's probably a tipping point where the analysis begins to fail because of insufficient data.

If the banks and credit card companies are complicit in feeding the RO data, then create enough fake "people" to indicate false movements in the economy. This data injection will need to be incredibly good because the RO is a giant pattern analysis engine. It will see movements like that.

Attack the Programmers

Bribe or blackmail the RO developers to alter the code to suit the conspirator's needs. "Accidents" involving RO company employees might have a similar effect. The RO may be effectively invulnerable but programmers are merely human.

Answer 11

Killing the RO is hard but we could remove it from power temporarily for a start.

three letters, EMP

The RO would be based in a decentralized redundant fortified server network that would be hardened against emp. Hunting down and destroying every server where the RO lives is unlikely to ever work, especially when the populous loves it.

But the RO acts observes and rules via electronics. We could blind and cripple it with high altitude emp. All the unshielded electronics would fail, the RO would survive in its hardened server complexes, but all of its sensors and police robots and propaganda speakers would go down and give the resistance a chance to get started and win people to their cause.

Answer 12

Hijack Nuclear Weapons and threaten to destroy the world

Your resistance movement needs to find a way to destroy all humans in the world. Hopefully the RO hasn't secured all the abandoned nuclear silos. Yet

Give me liberty or give me death

It's simple; threaten the RO to either deactivate or you will destroy the entire world. Make it so the safest choice to ensure humanity lives is to deactivate. For this the resistance will need to find someone who's willing to destroy the world but sane enough not to do it unless they must; similar psychological screening is done on people working in nuclear silos today .

Answer 13

You will literally have to rely on the human spirit.

You mention the computer runs using "cold hard logic." This limits its ability to predict the unpredictable human mind. A portion of what we do is predictable, of course. However, there is a portion that is not, and I would happily call that portion "inspired by the human spirit."

The key would be to generate enough unpredictability that the predictions made by the Overlord are no longer useful. However, do that over a long period of time, and the Overlord will just raise up stone walls and defend. The strike has to be instantaneous, all over the planet.

Humanity would have to play along, allowing the Overload to "relax" and use less resource intensive protective mechanisms. Then, in a flash, everyone's spirit would need to visibly unite and have all of humanity rise up against the machines. Remember: we don't have to kill the Overlord, simply take control back from it.

Think that's impossible? Think back to 9/11. There are stories of biker gangs stopping on the side of the highway to help old ladies change their flat tire. Humanity can unite. That was just 9/11, if I may be so bold to use such a wording. Imagine how united we could be against the continued oppression of a Robot Overlord. There's more to us than meets the eye!