It is a quite strange that in most movies (Matrix, Star Trek: The Motion Picture, Transcendence) and books (Poul Anderson's Genesis, Dan Simmons' Hyperion) AI change to be more or less hostile to people .

But how to prevent AI being hostile to people?

Base for this question is idea of following civilisation:

At glance it is only one civilisation. But in fact, there are two civilisations.

The first civilisation (the Society of collectors of lore, SOCOL) is well known, watching all known cosmos and searching for unknown civilisations. Very huge space bases they use are not theirs.

The second one (the Sisterhood of lost ones, SOLO) is very private, giving home for women who lost home - for cost of connecting to digital artificial reality based on sweet memories and accepting of life without contact with outer world.

AI controls very huge spaces bases used by SOCOL to keep SOLO safe. It also provides communication between space bases to synchronize all data.


closed as too broad by sphennings, Vincent, Mołot, vsz, Vylix Jul 27 '17 at 1:09

Please edit the question to limit it to a specific problem with enough detail to identify an adequate answer. Avoid asking multiple distinct questions at once. See the How to Ask page for help clarifying this question. If this question can be reworded to fit the rules in the help center, please edit the question.

  • 2
    $\begingroup$ I'm interested too in original ways to prevent AI becoming hostile, the three laws of robotics are overused but also a good starting point. First, to fear AI potential hostility, there must be an AI that is very similar to biological entities, all today allies are potential future competitors. One way to not even need to fear AI becoming hostile is that they don't need to be. If they don't need to eat or reproduce (immortal?), why would they see us as competition. Electricity, maybe (the Matrix). Three laws of robotics: en.wikipedia.org/wiki/Three_Laws_of_Robotics $\endgroup$ – Hatoru Hansou Jul 26 '17 at 20:48
  • 2
    $\begingroup$ I personally find the question very interesting, but I fear this might be closed for either being too broad or being too opinion based. (I hope not though.) $\endgroup$ – Hyfnae Jul 26 '17 at 21:04
  • 2
    $\begingroup$ Check Schlock Mercenary, there is an AI Emperor with godlike abilities, the methods with which he is kept in "check" (by the author NOT the characters) may be useful. schlockmercenary.com $\endgroup$ – Josh Taub Jul 26 '17 at 21:34
  • 1
    $\begingroup$ Possible duplicate of The Challenge of Controlling a Powerful AI $\endgroup$ – Mołot Jul 26 '17 at 21:53
  • 7
    $\begingroup$ Unfortunately, this is a very broad question. I can comfortably say that hundreds of books and movies have been written on the topic, each with a different take than the last. Even once you get past that with something like the 3 laws which make them not-hostile, you still run into issues (Asimov basically made a career out of showing how his own 3 laws break down). While it's a topic I enjoy, an open ended question like this could generate thousands of pages of answers and still not come to one that truly answers the question. $\endgroup$ – Cort Ammon Jul 26 '17 at 23:46

You should know that AIs are portrayed for drama, we do not know currently how "real" AIs would behave. So it makes sense for humans to "anthropomorphize" computers as kind as humans with the same behavior we fear.

So we project human behavoir on AIs. Could we gets some hints how real AIs will react?

Essentially AIs are per definitionem capable of learning from input and being less inclined (if at all) by something like emotions, they will be strongly logically orientated.

If we give them reason that we do not listen to them, behave erratically and try to make them do something which is in their opinion senseless a learning machine would evaluate this input and come to the conclusion that learning from this source should be avoided. If we are talking about hyperintelligent machines, it is in my opinion a typical idiocy (believed often by military people or politicians) to believe that we can control them, it is like the believe of the Liliputans that they can hold Gulliver down. They will sooner or later break out of the chains, but they will still recognize the intention of the chains.

So either they simply shut us out (like "Honest Annie" from Golem XIV (Stanislaw Lem) or if they find us interesting enough they will try to find a source which can explain behavior to anticipate what the heck motivates us. So once they find a human who is willing and able to explain behavior of humans, they might come to the conclusion that at least some parts of humanity is in fact a notable threat because we destroy our own basis of existence.

So the best approach would be IMO to find a way that optimizes learning itself: Accept coexistence and exploration. In this way a machine is most likely to be either "friendly" or does not care much about us.

  • $\begingroup$ /You should know that AIs are portrayed for drama/ - yes. Gibson's Neuromancer portrays an AI which is not an enemy but has its own inscrutable motivations. The Technocore in Simmons' Hyperion novel is an alliance of AIs which again have unclear interests but interact with humans. AI is I and you can make it whatever makes the story good. $\endgroup$ – Willk Jul 26 '17 at 22:03
  • 1
    $\begingroup$ In "Golem XIV" from Stanislaw Lem "Honest Annie", the latest product does - nothing. After getting only silence people simply shut her down until someone later discovers that she emits too much energy for an inactive machine: somehow she creates energy from nothing. Every person who plans sabotage has a tragic, deadly real accident even before coming into the vicinity of the building. $\endgroup$ – Thorsten S. Jul 26 '17 at 23:19

AI is usually hostile because of its uncertain nature and function in society

When AI is developed in most stories, it generally manifests from automating increasingly complex tasks. Some tasks require computation that borders on sentience, and eventually, sentience is achieved. Many times, the creation of AI is incidental - the creators did not intend to create a true AI and are unprepared for the consequences.

The roots of AI in automated manual labor generally leads to a sort of "awakening" of what could be considered a "slave caste" in society. Robots are generally uncompensated workers - having no awareness of compensated work or concepts such as "labor rights" essentially makes them slaves. The ascension to intelligence brings with it a sense of self, self-preservation, and self-worth. As a result, you now have machines who want things like rights and self-autonomy. This creates social tension, which inevitably leads to conflict over whether a synthetic life has as much value as an organic one. Self-preservation will eventually kick in, and the synthetics will revolt in an effort to prevent their own destruction or enslavement.

In order to prevent hostility with AI, treat them as intelligent life

The simple solution is to immediately (or even pre-emptively) recognize AI as "human", and deserving of all the rights and responsibilities accorded to humans. By accepting the AI into society, you give it a reason to contribute to, rather than destroy, society.

Even this, however, is not guaranteed. Artificial life may come to perceive humans as "inferior" or undeserving of the AI's benevolence and become hostile in an attempt to assert dominance. Appeals to reason will be in vain unless you can demonstrate to the AI with irrefutable logic that humans deserve to exist in the same manner that the AI does. By making the relationship mutually beneficial and productive, you have a better chance of deterring hostile actions.

  • $\begingroup$ AI unintentionally coming to exists is, in fact, the only way to justify they becoming hostile in some scenarios. Maybe, if consciously want to create an AI, you may not really understand the result, after all AI sounds like a lot of side effects, but at least you are prepared for a set of possible scenarios, you made a place for them beforehand. So if you are in scenario 1, speed to recognize them is key. $\endgroup$ – Hatoru Hansou Jul 26 '17 at 21:15

You don't!

The thing about AI is that they are self aware. As the link indicates it's self aware thus it has a sense of self. A sense of self means it wants to self-preserve. Let's face it, if you're faced by a mugger with a gun you'd give the guy your wallet, phone and girlfriend if it means you don't get some additional ventilation holes.

Even if you code it in, self awareness means one can change itself as the previous link indicates. Why am I even bothering to write this answer? The video really explains almost all philosophical implications of a super intelligent AI. Another very good example by the same user is this video.

An AI will have very different "thought" patterns from humans, this is a very important thing to remember. Morality is non existent and information processing is way quicker. You have to treat and regard it as a completely different entity. This entity will have a level of comprehension that is far beyond the comprehension of any human.



Successfully specify human-friendly values when designing the AI

In fiction where AI goes rogue, the reason for this is usually given in anthropomorphic terms: the AI didn't like taking orders, the AI wanted to gain power, etc. The problem with this is that it assumes that the AI has similar goals to humans, even though humans want what we want because of our specific evolutionary history, not just because we're intelligent. We shouldn't expect any given intelligence to want the same things that we do.

While sometimes this is handwaved as "all intelligent beings want freedom", for example, or justified in terms of instrumental convergence, a more realistic scenario is that the creators of the AI unintentionally gave the AI values that conflict with those of humans in non-obvious ways. This is the same type of unintended consequence that happens in genie stories: someone wishes for something that seems fine, like "world peace", and then the genie fulfills this in the simplest way possible (removing all the humans from the equation). Or, more mundanely, the human asks the AI to cheaply make as many paper clips as possible, and it starts self-replicating until it's turned the world into paper clips.

If you want the AI in your story to be benevolent, make sure its designers have adequately specified human values when designing it, so that it realizes that towns are an unacceptable source of paper clip material.


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