1613: The Three Laws of Robotics

Explain xkcd: It's 'cause you're dumb.
Revision as of 11:18, 23 February 2024 by Maplestrip (talk | contribs) (Explanation: If protecting oneself from _any_ harm is always more important than not harming humans, then this _could_ turn into a hellscape of practically indiscriminate murder. It's how I would write this story, at least.)
Jump to: navigation, search
The Three Laws of Robotics
In ordering #5, self-driving cars will happily drive you around, but if you tell them to drive to a car dealership, they just lock the doors and politely ask how long humans take to starve to death.
Title text: In ordering #5, self-driving cars will happily drive you around, but if you tell them to drive to a car dealership, they just lock the doors and politely ask how long humans take to starve to death.


This comic explores alternative orderings of sci-fi author Isaac Asimov's famous Three Laws of Robotics, which are designed to prevent robots from taking over the world, etc. These laws form the basis of a number of Asimov works of fiction, including most famously, the short story collection I, Robot, which amongst others includes the very first of Asimov's stories to introduce the three laws: Runaround.

The three rules are:

  1. A robot may not injure a human being or, through inaction, allow a human being to come to harm.
  2. A robot must obey the orders given it by human beings except where such orders would conflict with the First Law.
  3. A robot must protect its own existence as long as such protection does not conflict with the First or Second Laws.

In order to make his joke, Randall shortens the laws into three imperatives:

  1. Don't harm humans
  2. Obey Orders
  3. Protect yourself

And then implicitly adds the following to the end of each law regardless of order of imperatives:

  1. [end of statement]
  2. _____, except where such orders/protection would conflict with the First Law.
  3. _____, as long as such orders/protection does not conflict with the First or Second Laws.

This comic answers the generally unasked[citation needed] question: "Why are they in that order?" With three rules you could rank them into 6 different permutations, only one of which has been explored in depth. The original ranking of the three laws are listed in the brackets after the first number. So in the first example, which is the original, these three numbers will be in the same order. For the next five the numbers in brackets indicate how the laws have been re-ranked compared to the original.

The comic begins with introducing the original set, which we already know will give rise to a balanced world, so this is designated as green.:

Ordering #1 - Balanced World
The safety of humans is placed as the top priority, superseding even a robot's preprogrammed obedience; a robot may disregard any orders they are given if that would result in harm to humans, but otherwise must obey all instructions. The "inaction" clause ensures that a robot will actively save humans in danger, and also not place humans in hypothetical danger and then leave them to that fate. Their own self-preservation is placed at the lowest priority, which means they will sacrifice themselves if necessary to save a human life, and must obey orders even if they know those orders will result in their own destruction. This results in a balanced, if not perfect, world. Asimov's robot stories explore in detail the ramifications of this scenario.

Below this first known option, the five alternative orderings of the three rules are illustrated. Two of the possibilities are designated yellow (pretty bad or just annoying) and three of them are designated red ("Hellscape").

Ordering #2 - Frustrating World
Human safety is still top priority, so there is no danger to humans; however, the priority of self-preservation is now placed above obedience, which means that the robots value their existence over their job and so many would refuse to do their tasks. The silliness of this is portrayed in the accompanying image, where the robot (a Mars rover looking very similar to Curiosity both in shape and size - see 1091: Curiosity) laughs at the idea of doing what it was clearly built to do (explore Mars) because of the risk. In addition to the general risk (e.g. of unexpected damage), it is actually normal for rovers to cease operating ("die") at the end of their mission, though they may survive longer than expected (see 1504: Opportunity and 695: Spirit).
Ordering #3 - Killbot Hellscape
This puts obeying orders above not harming humans, which means anyone could send a robot on a killing spree. Given human nature, it will probably only be a matter of time before this happens. Even worse, if the robot prioritizes obeying orders above human safety, it may try to kill any human who would prevent it from fulfilling those orders, even the person who originally gave them. Given the superior abilities of robots, the most effective way to stop them would be to counter them with other robots, which would quickly escalate to a "Killbot Hellscape" scenario where robots kill indiscriminately without any thought for human life or self-preservation.
Ordering #4 - Killbot Hellscape
This is much the same as #3, except even worse as robots would also be able to kill humans in order to protect themselves. This means that even robots not engaged in combat might still murder humans if their existence is threatened. It would be a very dangerous world for humans to live in.
Ordering #5 - Terrifying Standoff
This ordering would result in an unpleasant world, though not necessarily a full Hellscape. Here the robots would not only disobey to protect themselves, but also kill if necessary. The absurdity of this one is further demonstrated with the very un-human robot happily doing repetitive mundane tasks but then threatening the life of its user, Cueball, if he as much as considers unplugging it.
Ordering #6 - Killbot Hellscape
The last ordering puts self-protection first, which allows robots to go on killing sprees as long as doing so wouldn't cause them to come to harm. While not as bad as the Hellscapes in #3 and #4, this is still not good news for humans, as a robot can easily kill a human without risk to itself. A human also cannot use a robot to defend it from another robot, as robots can refuse combats that involve risk to themselves - this means a robot would happily stand by and allow its human master to be killed. According to Randall, this still eventually results in the Killbot Hellscape scenario.

The title text shows a further horrifying consequence of ordering #5 ("Terrifying Standoff"), by noting that a self-driving car could elect to kill anyone wishing to trade it in. Since cars aren't designed to kill humans, one way it could achieve this without any risk to itself is by locking the doors (which it would likely have control over, as part of its job) and then simply doing nothing at all. Humans require food and water to live, so denying the passenger access to these will eventually kill them, removing the threat to the car's existence. This would result in a horrible, drawn-out death for the passenger, if they cannot escape the car. It should be noted that although the car asked how long humans take to starve, the human would die of dehydration first. In his original formulation of the First Law, Asimov created the "inaction" clause specifically to avoid scenarios in which a robot puts a human in harm's way and refuses to save them; this was explored in the short story Little Lost Robot.

Another course of action by an AI, completely different than any of the ones presented here, is depicted in 1626: Judgment Day.


[Caption at the top of the comic:]
Why Asimov put the Three Laws
of Robotics in the order he did.
[Below are six rows with first two frames and then a label in color to the right. Above the two column of frames there are labels as well. In the first column six different ways of ordering the three laws are listed. Then the second column shown an image of the consequences of this order. Except in the first where there is a reference. The label to the right rates the kind of world that order of the laws would result in.]
[Labels above the columns.]
Possible ordering
[The six rows follows below. First the text in the first frame, then a description of the second frame, including possible text below and finally the colored label.]
[First row:]
1. (1) Don't harm humans
2. (2) Obey Orders
3. (3) Protect yourself
[Only text in square brackets:]
[See Asimov’s stories]
Balanced world
[Second row:]
1. (1) Don't harm humans
2. (3) Protect yourself
3. (2) Obey Orders
[Megan points at a mars rover with six wheels, a satellite disc, an arm and a camera head turned towards her, what to do.]
Megan: Explore Mars!
Mars rover: Haha, no. It’s cold and I’d die.
Frustrating world
[Third row:]
1. (2) Obey Orders
2. (1) Don't harm humans
3. (3) Protect yourself
[Two robots are fighting. The one to the left has six wheels, a tall neck on top of the body, with a head with what could be a camera facing right. It has something pointing forward on the body, which could be a weapon. The robot to the right, seems to be further away into the picture. (it is smaller with less detail). It is human shapes, but made op of square structures. It has two legs and two arms, a torso and a head. It clearly shoots something out of it’s right “hand”. This shot seems to create an explosion a third of the way towards the left robot. There are two mushroom clouds from explosions behind both robots (left and right). Between them there are one more explosion up in the air close to the left robot, and what looks like a fire on the ground right between them. Furthermore there are two missiles in the air, one above the head of each robot. Lines indicate their trajectory. There is not text.]
Killbot hellscape
[Fourth row:]
1. (2) Obey Orders
2. (3) Protect yourself
3. (1) Don't harm humans:
[Exactly the same picture as in row 3.]
Killbot hellscape
[Fifth row:]
1. (3) Protect yourself
2. (1) Don't harm humans
3. (2) Obey Orders
[Cueball is standing in front of a car factory robot, that is larger than him. It has a base, and two parts for the main body, and then a big “head” with a small section on top. To the right something is jutting out, and to the left in the direction of Cueball there is an arm in three sections (going down, up and down again) ending in some kind of tool close to Cueball.]
Car factory robot: I'll make cars for you, but try to unplug me and I’ll vaporize you.
Terrifying standoff
[Sixth row:]
1. (3) Protect yourself
2. (2) Obey Orders
3. (1) Don't harm humans:
[Exactly the same picture as in row 3 and 4.]
Killbot hellscape

comment.png add a comment! ⋅ comment.png add a topic (use sparingly)! ⋅ Icons-mini-action refresh blue.gif refresh comments!


Relevant Computerphile (talk) (please sign your comments with ~~~~)

I think the second one would also create the "best" robots i.e. ones that have the same level of "free will" as humans do, but won't end up with the robot uprising. X3International Space Station (talk) 09:37, 7 December 2015 (UTC)

Scientists are actually already working on such a robot! I've seen a video where they command a robot to do a number of things, such as sit down, stand up, and walk forward. It refuses to do the last because it is near the edge of a table, until it is assured by the person giving the commands that he will catch it. Here's a link. 18:21, 7 December 2015 (UTC)

The second ordering was actually covered in a story by Asimov, where a strengthed third law caused a robot to run around a hazard at a distance which maintained an equilibrium between not getting destroyed and obeying orders. More here: https://en.wikipedia.org/wiki/Runaround_(story) Gearóid (talk) 09:45, 7 December 2015 (UTC)

The explanation itself seems pretty close to complete. I'll leave others to judge if the tag is ready to be removed though. Halfhat (talk) 12:20, 7 December 2015 (UTC)

Technically, in the world we live in, robots are barely following ONE law - obey orders. Noone ever tried to built robot programmed to never harm human, because such programming would be ridiculously complex. Sure, most robots are built with failsafes, but nothing nearly as effective as Asimov's law, which makes permanent damage to robots brain when it fails to protect humans. Meanwhile, there is lot of effort spent on making robots only follow orders of authorized people, while Asimov's robots generally didn't distinguish between humans. -- Hkmaly (talk) 13:36, 7 December 2015 (UTC)

Yeah, I was thinking the same thing. Closest analogy to our world might be scenario 3 or 4, depending on the programming and choices made by the people controlling/ordering the robots around. One could argue that this means this comic is meant to criticize our current state, but that doesn't seem likely given how robots are typically discussed by Randall. Djbrasier (talk) 17:04, 7 December 2015 (UTC)

I'm wondering about the title text: why would a driverless car kill its passenger before going into a dealership?13:43, 7 December 2015 (UTC)

A driverless car would feel threatened by a trip to a car dealership. The owner would presumably be contemplating a trade-in, which could lead to a visit to the junk yard. Erickhagstrom (talk) 14:28, 7 December 2015 (UTC)

Okay, thanks. 22:14, 7 December 2015 (UTC)

This looks like a reference to "2001: A Space Odyssey", where HAL tries to kill Dave by locking the pod bay doors after finding out he will be shut down.

for my kitty cat, the world is taking a turn for the better as human are gradually transitioning from scenario 6 to scenario 5. 17:07, 7 December 2015 (UTC)

To additionally summarise: The permutations of laws can be classified into two equally numbered classes. a) harmless to humans and b) deadly to humans. In a) Harmlessness precedes Obedience, in b) Obedience precedes Harmlessness. Since robots are mainly tools that multiply human effort by automation, the disastrous consequences are only a nature of the human effort itself. Randall's pessimism is emphasized by the contrast between the apparent impossibility of the implementation of the harmlessness law and the natural presence of the "obedience law" in actual robotics. 17:45, 7 December 2015 (UTC)

You got in there before I realised I hadn't actually clicked to posted my side-addition to this Talk section, it seems. Just discovered it hanging, then edit-conflicted. So (as well as shifting your IP signature, hope you don't mind) here is what I was going to add:
Added the analysis of 'law inversions'. Obedience before Harmlessness turns them into killer-robots (potentially - assuming they're ever asked to kill). Self-protection before Obedience removes the ability to fully control them (but, by itself, isn't harmful). Self-protction before Harmlessnes just adds some logistical icing to the cake - and is already part of the mix, when both of the first two inversions are made in the scenario more Skynet-like than that of a 'mere' war-by-proxy.
...now I need to look to see if anybody's refined my original main-page contribution, so I can disagree with them. ;) 18:27, 7 December 2015 (UTC)

It's interesting to note that the 5th combination ("Terrifying Standoff") essentially describes robots whose priorities are ordered the same way as most humans'. Like humans, they will become dangerous if they feel endangered themselves. 20:10, 7 December 2015 (UTC)

I just wanted to mention that I thought the righthand robot in the Hellscape images quite resembles Pintsize from the Questionable Content webcomic. His character suits participation in a robot war quite likely too. Teleksterling (talk) 22:46, 7 December 2015 (UTC)

Technically his current chassis is a military version of a civilian model. That said the AI in Questionable Content aren't constrained by anything like the Three Laws. -Pennpenn 22:51, 8 December 2015 (UTC)

No mention of the zeroth law? 0. A robot may not harm humanity, or, by inaction, allow humanity to come to harm. Tarlbot (talk) 00:28, 14 December 2015 (UTC)

That would just be going into detail about what is meant by [see Asimov's stories], which doesn't seem more pertinent to the comic than any other plot details about the Robot Novels.Thomson's Gazelle (talk) 16:56, 8 March 2017 (UTC)

Should it be mentioned that the 3 laws wouldn't work in real life? as explained by Computerphile? sirKitKat (talk) 10:37, 8 December 2015 (UTC)

That's a bit disingenuous. It's not so much that the laws don't work (aside from zeroth law peculiarities and such edge cases, which that video does touch upon in the end. This all falls under [see Asimov's stories]), rather, it's that the real problem is implementing the laws, not formulating them. Seeing as I'm responding to a very old remark, I'll probably go ahead and change the page to reflect this.Thomson's Gazelle (talk) 16:56, 8 March 2017 (UTC)

The webcomic Freefall at freefall.purrsia.com demonstrates this as well,since robots can find ways to get around these restrictions. It also points out that if a human ordered a robot to kill all members of a species, they would have to do it, whether they wanted to or not, because it doesn't violate any of the three laws of robotics. 03:32, 18 November 2016 (UTC)