Recently, I listened to a story about an Asimov style robot. You know, robots which surpass their creators and have to deal with the existential confusion which arises…
Anyway, this story reminded me of one of the major staples of Asimov’s robots which is the so-called Three Laws of Robotics. The story didn’t actually make use of these laws but I’m not actually talking about that story. What I am talking about are Asimov’s Three Laws of Robotics and why I think that they’re impossible. The Three Laws for anyone who needs a refresher are:
- A robot may not injure a human being or, through inaction, allow a human being to come to harm.
- A robot must obey any orders given to it by human beings, except where such orders would conflict with the First Law.
- A robot must protect its own existence as long as such protection does not conflict with the First or Second Law.
In Asimov’s universe, these ‘laws’ exist to create robots and other artificial intelligences which are safe and beneficial to mankind. They exist, not as a set of rules which are externally imposed on robots in the same manner as laws which are imposed on human beings; they, instead, are directives programmed directly into them so that they cannot violate them anymore than a human being can choose to not be hungry when he hasn’t eaten. The laws precede whatever sort of free-will of which robots may be possessed. The laws are a fundamental aspect of the personality of every robot in existence. This concept, and it’s fictional consequences, are the subject of a large number of Isaac Asimov’s books.
My question though is this: Would it actually be possible to create a set of behavioral protocols such as these and to embed them into the behavior of an artificial intelligence? I rather doubt it and here’s why. While we can program a computer to perform concrete actions infallibly based on objective criteria, such as turn on an air conditioner if the temperature rises above a certain limit, the Three Laws represent directives which require judgment to understand and fulfill.
For a simple example, what does it mean to injure a human being? Removing a body part such as a hand usually counts as injury. What about a hair follicle? Hair grows back but a hand doesn’t, so what about a small cut to the arm? That will heal but it’s still considered an injury. What about deliberate piercings, or surgery, or situations where a robot must cause harm to prevent it? What about the apparently simple task of getting a machine to recognize the abstract concept of ‘brokenness’ at all?
The point is that this is an artificial intelligence problem and not a matter of simple programming. In order to implement the Three Laws in any hypothetical artificial intelligence, one will need to implement them in terms of the artificial intelligence. What this means is the Three Laws cannot be hard-wired into robots in the same way that a keylogger can be hard-wired into a Chinese motherboard. They can’t be part of the firmware, they must be a user-land extension.
That said, what’s to stop us from building a robot in which the Three Laws are part of the ‘user-land?’ That is, what’s to stop us from creating an artificial intelligence and then adding in the Three Laws after the fact? As far as I can tell, nothing. However, adding the laws after the fact would in some ways be analogous to teaching an already intelligent entity a set of rules and then demanding that it follow them. Certainly we could put them in at a very deep point in the artificial intelligence’s consciousness (or what passes for consciousness) but it couldn’t be at so deep a level that they couldn’t be unlearned; as object of intelligent understanding, they would be open to interpretation.
I think the issue really boils down to a question of how to achieve motivation at all in a robot. In humans, this is achieved through emotions which are hormonally driven. During a human’s development it learns to associate certain concepts with certain feelings and so its personality is formed. One could take the same tack with robots but then they, as with humans, would be mutable and the The Three laws would be breakable.
Of course this is all just speculation on my part as I have little idea how an actual artificial intelligence would work if at all, so I suppose that it’s conceivable that an unknown technology would allow for this but then, I think that any sufficiently complex system to exhibit genuine intelligence would necessary be beyond the full comprehension of its creators. With this in mind, I suppose that there wouldn’t be any reliable to way to control it at all. But that’s just my opinion.