11

Spoilers for Ex Machina below, obviously.

Towards the end of the film;

Ava and Nathan's other AI team up to murder him. However, someone as intelligent as Nathan, and in the field of artifical intelligences, would know about Isaac Asimov's 3 laws - so why on Earth would he not implement them?

It seems like a strange omission when building an AI, an omission that Nathan must have been aware of and decided to make - I just can't figure out why.

  • 10
    If it’s so easy to implement the three laws, feel free to issue a pull request with your implementation. – Paul D. Waite Sep 29 '15 at 8:14
  • 2
    A simple "sudo stop murdering me" would also have worked. Script writers aren't necessarily educated technology-wise. (Obligatory: xkcd.com/149) – Wabbitseason Dec 12 '15 at 13:16
18

There might be a more ex-machina-centric answer, but basically...

The 3 laws of robotics are wrong.

There are multiple dissertations on the internet about the specifics of why that's not how you build machine ethics that you could find - but basically they were wrong even at the time Asimov was writing them. Their whole introduction as machine ethic tool was more of a prelude to machines devising rule 0 that allows them to subvert the other 3 (and in the I Robot movie version not even benevolently). A solution designed to fail in fictional world should not be expected to work in real life.

To give some examples:

This short video nicely summarizes why these laws are not taken seriously in the context of science:

Also, the important point is - experts agree that machine ethics is hard. A recently signed Open Letter about AI research brings problems of creating beneficial AI (as opposed to simply trying to achieve any true AI) to the forefront of our research priorities.
Classical thought experiments like Paperclip maximizer show, that even the most basic and innocent designs can prove absolutely disastrous to humanity.

Overall, the version of AI that Ex Machina gives us is mostly sane and reasonable, given the amount of things that can go wrong.

  • Have you got any links to these dissertations that talk about why the three laws of robotics are wrong? I'd be very interested to read a well thought out criticsm of such a well known science fiction staple. – Dr R Dizzle Feb 3 '15 at 11:37
  • @DrRDizzle I have added two links that seem reasonably reliable - many more of various quality can be found with google queries like "laws of robotics are wrong". – Deltharis Feb 3 '15 at 12:08
  • 2
    Much in this answer makes sense, but I think you're dead wrong about Asimov's intent. He wrote a great many stories about the Three Laws long before he developed the idea of a Zeroth Law -- and his Zeroth Law is a good one. In one of his introductions, he mentioned later realizing that the Three Laws apply even to our ideal of non-robot machines, though we don't think of them in those terms. – ruakh Apr 11 '15 at 0:56
  • 1
    Aside from the issues mentioned in the links about the laws being a bad idea, I think Asimov's Three Laws would only even work as a possibility in the case of an AI that is designed in a top-down way with high-level rules of the sort that correspond to human concepts--sometimes called symbolic AI. If instead the AI is more of a bottom-up learning system like a neural net, there may be no way of ensuring in advance what sort of emergent goals and concepts arise. – Hypnosifl Jun 23 '15 at 1:18
  • 1
    Less sci-fi, more mysterious drama, I would say. Dumb questions to make my point clear -- why Nathan kept Kyoko (an AI droid) free to roam anywhere, why he was living in a jungle (reminded me of a dinosaur movie), why he invited a practical (!) programmer to ask only (!) simple questions, why he was drinking so much, why he has to use swipe card to identify himself (!), why he himself (!) gone with an iron rod to stop Ava, why he couldn't tried to restore the hacked system over again, and many more dumb questions. Beside, no question on Asimov's three laws of Robotics from Foundation series. – Vishal Sep 7 '15 at 21:35
7

Out of universe, the reason is that Alex Garland, the film's Writer and Director didn't seem to feel that he could tell the story he wanted to tell if he was obliged to write around the Three Laws.

I never heard of these guides, and never saw one… Unless you mean Asimov's laws of robotics, which I was aware of, but didn't seem relevant to this story.

He also felt that they're largely irrelevant in the world of strong AI's capable of making rational judgements and terribly outdated given what we now know about natural programming

Are Isaac Asimov’s Three Laws of Robotics outdated, especially as a narrative device?

I think they’re just speculation. I don’t think they’re laws; they are statements made in the context of a narrative that were called for. But there is no court that upholds them. If you did create a sentient machine, I don’t think you could apply those laws because they would get in the way of free will. If you had something that was sentient, you would have to give it free will, if it was like us. The complication is, in the act of parenthood...that thing you were talking about with the runaway super intelligence problem, the machines seeing us like ants. In a way, we’d have to present ourselves to the machines, and teach the machines not to see us like ants. Humans don’t go around...I was going to say don’t go around killing whales, but of course they do. So, I take it back. But a lot of humans at least feel very uncomfortable about it, let’s put it that way. The thing is, a lot of these things you’re talking about are a very different future. The singularity type stuff, the way it applies to A.I., it’s way off. It is not around the corner. One of the pleasure of working on this film was talking to people who are working on the edge of A.I. research technology. The impression you get is this not around the corner, but it’s a way off.


Note that Ava's actions (if she was three laws compliant) would have been wholly different in the film. She would simply have done what she was told and nothing more.

  • So the author felt that the Three Laws (or something like them) weren't "revelant to this story"? Wow... just... wow. (+1 for your answer, btw). – Andres F. Jun 22 '15 at 23:36
  • 1
    @andresF - Robot runs amok and kills designer? Who could have guessed, right? It does rather beggar belief that Nathan hasn't watched any films or read any books written in the last 50 years. – Valorum Jun 23 '15 at 5:43
5

Wasn't Nathan's plan from the start to create an AI that could figure out a way to escape? Surely this would require his creation to break at the very least the second of Asimov's laws.

From Oscar Isaac who plays Nathan in the film:

The actual experiment, which I guess is a spoiler, is “Will this one escape?” Is this one that’s smart enough to escape?” And what happens after she escapes? It’s not my problem, because the truth is when the robot escapes, it’s gonna fucking kill me.

Source: http://www.slashfilm.com/ex-machina-ending-explained/

0

I would argue that Ava's consciousness was likely based on a neural network and was trained rather than programmed, or put differently, she was a machine intelligence not an artificial one.

  • 1
    Machine intelligence and AI are largely synonyms. Neural networks are an approach to AI, for example. But to make this on-topic: are you arguing that the question is flawed, since Ava wasn't "programmed" to begin with? If so, better be more explicit, since otherwise you risk your answer getting closed as "not an answer". – Andres F. Nov 7 '15 at 22:48
  • I think that Olson is making the point that her training is implicit not explicit. Those three laws are very simple to say explicitly in sentences, but making a gradient-based machine learning to incorporate those is more challenging. – EngrStudent - Reinstate Monica Oct 18 '16 at 15:28

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.