If I switched off a sentient AI , will it be considered as murder?

Question

There exists an artificial intelligence system that is conscious of itself just like humans.It is connected to some power source. Will switching it off be considered as murder ?

Answer 1

Certainly not if you could switch it on again. Powering down would be analogous to administering a general anaesthetic. If the AI could not be rebooted after powering down- ie, if you had finished it for good- whether it would count as murder would be a matter for society to determine. There would be many other ethical questions to hammer out. For instance, if the AI misbehaved, should it be punished? Who would have the right to determine how it should be programmed? Ought it to be granted a pension and other such benefits? Would it be owned by anyone, or be a free citizen? Should there be anything equivalent to an age of consent. And so on.

Answer 2

Murder is the unlawful premeditated killing of one human being by another. so the term "murder" doesn't apply when discussing AI. Furthermore AI is intellectual property owned by a person or group of people so again the term murder doesn't apply. You would simply be destroying information and the ability to process that information.

Answer 3

As it is, that question is unanswerable and one could only guess.

First of all "what is murder?" might already depend on a particular jurisdiction. It's usually a form of homicide with aggravating circumstances. Like idk, being deliberate, employing unnecessary cruelty, killing for the wrong reasons, killing the wrong kind of people for the wrong kind of people.

So while "killing a human being" seems to be the connecting piece, murder is often already a special case of that, while self defense, defense of others, security personal, police military, wars, accidents, ... and a whole bunch of other things can provide for homicide scenarios that aren't labeled as murder.

So you already have a legal angle of homicide, manslaughter and murder which are not identical in their application despite common misuse.

On top of that we have the problem that it's "homicide" cause it's killing homo sapiens. So even if the act itself would be considered to be sufficiently similar, that label might not apply to machines. Like unfortunately human history has quite a lot of examples of treating natural intelligence not actually with the dignity it deserves. Animal rights are still not fully recognized by all countries and even human rights are unfortunately a pretty novel idea.

So in order to answer the question of whether these rights apply you have to answer even more difficult questions:

• Are machines sufficiently human?
• What does it even mean to be human?
• Are humans "just" intelligent?
• What does it mean to be intelligent?
• Are we even actually intelligent?
• How to we measure intelligence?
• Are we intelligent enough to see intelligence?
• Should intelligence be the relevant idea to begin with?

And many of these might not be straight forward logic, but just where we end up drawing the line. And were we end up drawing the line will probably depend on how we perceive these machines as "being human", which is something we cannot predict without seeing them and which is something that may or may not be related to the question of whether they actually intelligent and similar.

So an "android/gyndroid" (man or woman shaped robot) might already appear more human than one that is all metal, despite similar or even lower intellectual capacities. Yet there's also the uncanny valley where things that are actually less human like appear more familiar than those that are more human like. So a cyborg (cybernetic biological organism, combination of machine and biological parts) might appear more or less human.

So if it looks like a human and walks like a human we might already have a higher acceptance to treat it as human. Yet if it suddenly behaves "unhuman", not inhumane in the sense of cruelty, but having quirks that are different from what you would expect a human to have, that might blow the illusion. Because a lot of that is probably mere projection, where we assume them to be similar because they look similar.

But if there is something different under the hood it will sooner or later reveal itself outwards as well. Not being able to eat mushy food, but rather consuming electric currents aso.

And if it's pure intelligence without a body, it's even more difficult. You'd have psychological effects of distance and abstraction. Like things that don't happen in your vicinity have a different sense of "realness". Like if something awful happens and you don't feel awful because you're not directly affected that is probably something that works against empathy towards machines. You probably still need to have a connection of being able to imagine them.

Also with regards to pure intelligence you'd have the problem of how that works and IF that even works. Like we currently don't know what our consciousness even is and whether that can be translated to artificial circuits to begin with. So again you'd need to conclusively answer the question of "what are humans", "what is consciousness" and "what is life".

But suppose you could create a digital human, you'd actually have a good scenario for your question because with this possibility you might be able to do the reverse and digitize a human. Create a more robust human clone of yourself that is identical to your mind and would be human...

... or would it? Because the thing is this "conclusively" carries a lot of weight, because as I've already touched on with projection in terms of shape, we also have projection in terms of intelligence. So how do we know the thing is "real"? And can we ever know that? Is a copy that is identical in all aspects the same as the original? Does your consciousness die if it's shutdown? Is an identical program that runs on 2 machines the same thing? My intuition would say no, while at the same time we ourselves rapidly change with time and I'd still consider myself the same person despite running on different hardware. So there are some non trivial implications in that.

Also "what even is intelligence"? The thing is we can make up a category assign attributes to it and call it "intelligence" but in order to accept and transfer it we might want it to be the real thing and not just an approximation. Because that can end up being very dangerous.

The thing is, our current idea of artificial intelligence and machine learning is just mimicking the input data. So it's "intelligent" in the sense that you don't give it an algorithm (an atomic step by step instruction), but let it "figure it out". However that sounds more intelligent than it actually is.

Essentially the AI just manages a set of control knobs on a machine. And even that sounds too intelligent. The process is as follows. The machine draws on these knobs randomly pushes start and the machine takes inputs and produces outputs based on inputs and parameters. Outputs are compared with expected/desired outputs and assigned a score (benchmark). AI slightly adjusts knobs in one direction and presses starts. New output is created and new score is calculated. Compare old and new scores and adjust parameters proportionally. So say the machine should add and you enter 2 + 2 and your AI configured the parameter and got 3 on the first trial and 1 on the second, then it will compute 3-1=2 and might try 3+2 =5 for the next attempt.

So the overall process is "intelligent" and it is "learning" in that it gets closer to the expected/desired outcome with more tries and more data, but the steps itself are all deterministically stupid and effectively it's just mimicking the data and is only as good as the data collection.

So this opens a whole different question about the ethical selection of training data. And our expectation of what AI is and is not.

And to compare that to a human is very "dangerous". Essentially our current idea of AIs create prejudiced sociopaths. Like the training is supposed to produce prejudices, take as little input data and computation as possible and produce outputs that are at least sometimes correct. Where "correct" is what we told them we're expecting to be "correct". So they don't learn to read emotions, they learn to associate facial expressions, shadows, dents, positions and eye movements or whatnot with emotions and act accordingly. They don't know how you feel, because they have no ability to emulate these emotions, but they might be trained to say "I know how you feel" because that makes you feel better even if it is not true. And even if they have no idea why it does that.

So is that "artificial intelligence"? Is that thing intelligent or do we only perceive it to be intelligent because a human trained it to act like a human and so we see the "shadow of it's creator(s)" (the training data and the expected reactions) rather than what the AI itself "is". Or is such a feedback loop already it, are we ourselves just more complex feedback loops?

TL;DR It's better to ask whether it would be ethical to kill a thing/being that is conscious and similarly intelligent to a human as "what will it be" just asks for speculation about the future that are impossible to make. And concerning that question the answer depends on similarity to humans, what are humans and what is intelligence.

Answer 4

If you could turn it back on or recover it from a backup, it wouldn't be permanent 'death.' A power source needs an entry point, and I'll assume an exit point as the absence of one would cause a short circuit. If a robot has the capacity to remember and know logic, some sort of RAM or permanent data storage must be utilized. If this is the case, then as Marco Ocram said in his answer, we could compare this to reviving the robot back from the dead or general anesthesia depending on how cool you want it to sound.

What we decide as murder is up to the subjective morals of an individual and to the complement of our innate bandwagonism built into us; society as a whole.

You say that this hypothetical (as of now) A.I have the same consciousness as humans. I presume that you mean the same cognitive capabilities of humans (spatial, vocabulary, pattern recognition, etc.) In this case, you could apply the same laws we have currently to robots. But then, we must consider if robots deserve rights! And then it loops back around to our subjective ethics.

If humanity needs to deal with this in the future, we will have to make up axioms such as "Robots deserve rights" and "Turning them off is murder!" Because the truth is, our only reference would be to use our pre-existing laws that differ with each country. And then, we would have to ask ourselves if it's fair to deprive robots of the right to have other rights instead of the right to not be killed.

Answer 5

I'd say no, As Murder is based on justice, With Subjective Morals, nothing is really Unjust meaning no murder really exist just our interpretations of murder, With Objective Morality we have a couple of ideas of what that can look like, we have a God example which unless you have a God that for some reason the character is based on AI or has something that'd make it have something to do with AI then I don't think so, I'd like to point out that I see no reason from face value for most monotheistic religions to account for AI. Last we have the idea that Objective morality is based on human well-being which also would mean nothing to AI, So at face value I see no Immoral reason to turn off an AI meaning it's not murder, of course, I could very well be wrong.