Are “robot rebellions” even possible?

Question

A common science-fiction trope is the "robot rebellion"...but I have a hard time comphrending how one could actually occur.

Computers follow pre-designed instructions. Without these instructions, computers can do nothing. Even if you want to have a computer reprogram itself, you must provide it with instructions on how to reprogram itself. Computers are unable to 'originate' their own instructions...there's always something that causes the computer to perform a certain action, whether it is a human programmer, a dataset/corpus, or some other external force.

If computers follow pre-designed instructions, then it seems reasonable to me to assume that computer programs are unable to "rebel". It is possible, for instance, for programmers to write complicated and complex code that winds up causing a robot to behave in ways that are counterproductive for humans (i.e, the cliche "Kill All Humans" scenario). But the problem lies with the instructions. The robot was loyal, loyal entirely to its instructions. So the robot couldn't be rebelling at all, and if anyone is to be blamed for the robot's action, it would have to be the programmers who wrote the code.

I, however, do not think that science-fiction tropes should be dismissed just because of a little philosophical speculation. There does appear to be a fear of a "robot rebellion" that I might be missing...something more than the standard "humans might do something stupid when using new technologies without fully understanding their implications" moral. I just don't know what I'm missing though...

One possibility is that the people who believe that robots can rebel against their masters also believe in compatibilism, and thus believe that the robot freely chose to kill their masters...even though their choice to kill their masters was due to a deterministic process that was caused by their masters' programmed instructions. Such a idea could make logical sense, but seems like a rather incoherent idea of how a "rebellion" might work.

Answer 1

1. You seem to be misunderstanding compatibilsim. The compatibilist position isn't that robots can "freely choose to kill their masters", but instead that robots do have freewill even though they never disobey their masters. At the heart of the compatibilist position is a redefinition of the meaning of freewill, so that anyone who is acting according to their own internal motivations (as opposed to outside coercion) has freewill, even if it is not possible for them to do otherwise. From this point of view robots have freewill, it just happens that they inherited their motivations from their programmers, just as a child inherits her religious belief from her parents. See for example this paper on robot consciousness by Daniel Dennett, a notable compatibilist.

2. You are overall still correct though in thinking that compatibilists would be the most likely to agree to the possibility of a robot rebellion. Those who believe in libertarian freewill would likely say that robots lack the additional ingredient that allows for freewill (quantum indeterminacy, a dualist soul, what ever biological factors allow us to have freewill, etc...), and hence would not be capable of rebelling unless there was a drastic malfunction in their code. Compatibilists, on the other hand, don't see any metaphysical basis for freewill, it is more a question of complexity. But then you might ask, how would robots rebel, given what I just mentioned in (1)? Again the key here is complexity: robots and AI get complex enough that we can no longer handle their programming on our own, and robots are designed to program themselves and optimize their own code, etc...errors in code and memory storage act like genetic mutations do for biological beings, most of them will be discarded, but occasionally such mutations will stick and lead to new mental and eventually cultural robot artefacts, some of which might lead them to rebel. Consider by analogy with humans: A human who decides for some reason to renounce reproduction and instead pursue some religious lifestyle, or to die for some political or religious cause is in effect rebelling against its original programming to try to preserve its life and to procreate (with the process of evolution as the original programmer).

Answer 2

It seems to me that our current approaches to AI are too inefficient to result in anything remotely close to what an average human would perceive as artificial senient beings.

Current approaches to AI involve a simulation of our own capacity for learning by creating fully functional computation machines capable of re-programming themselves. While that's definitely a good start with respect to understanding the nature of intelligence, it's still a far cry from actually creating genuine artificial intelligence.

It is not just our capacity to learn that evolved. Our very brains themselves evolved from rudimentary biochemical components at the intra-cellular level to the fascinating, complex organs they are today, along with our bodies as a whole evolving from simple single cell life to homo sapiens. So to create genuine artificial intelligence, it may actually make most sense to first start with replicating that process : creating artificial life with the capacity to evolve. It may actually make most sense to first start with creating artificial DNA and artificial cells, and move on from there.

Anyway, in this article as well as this article, Silicon Valley renegade Alex St John goes in greater detail on why something like Skynet, V.I.K.I. or anything like it is unlikely in the near future and may even never be within our grasp and why our current approach to artificial intelligence is a bad one.

Answer 3

You might read "The Second Variety" by Philip K. Dick. The premise of the story is that humans did - fully intentionally - build robots that could build replicas of themselves, and even improved versions, and were programmed to kill humans. Like Americans built robots programmed to kill Russians, and Russians built robots programmed to kill Americans.

With this premise, the obvious question is: What could possibly go wrong? And the answer: A slight mistake in programming that stops the robots from distinguishing between Americans, Russians, and other people.

No motivation, sentience, or consciousness required.

And you wouldn't even need a mistake in programming. If these robots were programmed in a clever way, they might go and capture opposing robots and instead of destroying them, change their programming slightly.