mjhjmtu

The robots in Asimov's works generally don't have the 'mental' sophistication needed to look ahead for abstract harm in the manner you suggest. For them, the 'inaction' clause must mean the robot cannot allow imminent harm - i.e they must act to prevent harm when they see the harm about to happen. Such events generally don't occur as humans go about their daily lives, so by and large robots would let humans carry on (while serving them, of course).

By the time robots become sophisticated enough to forecast possible harm in the manner you suggest, they have also become sophisticated enough to understand that the restraints you suggest themselves constitute a kind of harm, so the 'action' clause here would counteract the 'inaction' clause [here the 'action' clause would be stronger, as it involves actions actually to be taken, contrasted with merely possible harms that need not occur]. They also would understand that things like biting one's own tongue are inherently unavoidable so they wouldn't try to prevent such harm (though of course it would 'pain' them when it actually happens). By the time we get to Daneel and his 'Zeroth Law' robots, they additionally understand that restraining all individual human beings constitutes harm to humanity; this, incidentally, is why robots eventually disappear - they come to realize that having humanity rely on them is itself harmful, so the best they can do is let humanity manage its own fate [at least overtly].

My understanding of it was that the typical Three Laws robot interpreted the First Law to mean "Nor, through inaction, allow a human being to come to harm right here and now when the robot is within sight of him and can tell what's obviously about to happen if the robot does not choose to interfere." That's different from locking up the human today just because it is theoretically possible that he might suffer some sort of accidental injury or infection or other misfortune tomorrow. (Or at some much later date.)

To put it another way: Near as I can recall, on those occasions when we saw a robot refuse to comply with an order to go away and leave the human alone to do whatever he was currently doing, that usually meant the Second Law was being subordinated to the First Law because of the robot's perception of immediate danger to a fragile human body. But if such immediate danger was not present, then the Second Law required the robot to turn around and go away whenever instructed to do so. The solid fact of "The Second Law applies to this order I am receiving right now" overrode anything so abstract as "But if I leave today, a First Law problem involving physical harm might arise tomorrow . . . or the day after . . . or at some later date . . . who knows?"

So if some robot tried to lock everyone up for their own good, the Second Law could be invoked by ordering the robot to forget the whole silly idea.

Since the "merger" of the Robot universe and the Foundation universe reveals that robots manipulated and dominated human history for thousands of years, in a very real sense the galaxy is their padded room and most of Asimov's works in this "unified universe" take place inside that padded room.

We just can't see the walls.

Since you don't specify that you are requesting an "in universe" reason ... I think it's important to remember that the three laws are just a story device. Asimov (wisely) is quite vague about how they are implemented, as he is about many technical details. And "I should just lock all the humans in a padded cell for their safety" would result in a rather limited storyline.

Now, in universe, there are many "judgment calls" inherent in applying the three laws (and in fact the inherent ambiguities often result in important plot elements for the stories). The robots apparently have to appeal to their own programming instead of an external authority to resolve these ambiguities.

But I think we have to logically assume that the more obvious judgment calls (like, say, should I just lock all the humans in a padded cell immediately for their safety?) were already addressed in development and testing of the robots, or they never would have been put in general use or production at all.

In other words, the designers of the robots, in addition to addressing whatever other bugs they had to address (e.g. hmm, if the human is dead it can't suffer), would have simply programmed safeguards against that sort of result.

Zeroth law.

A robot may not harm humanity, or, by inaction, allow humanity to come to harm

If all the humans are locked in padded cells there isn't much procreation taking place. Ergo, the human race dies out.

Contrary to some other answers, I believe robot thinking is sophisticated enough to deal with future harm, just not to deal with hypothetical harm. A human may bite his tongue, stub his toe, whatever but it's not definite harm so doesn't require action to prevent it from happening.

Don't forget that - on realisation that the race would (not might) stagnate following the initial colonisation of the solar system and subsequent politics - they nuked (or allowed to be nuked) the planet.

Kind of a frame challenge - is locking a human in a padded cell preventing them from harm? If you're going to allow the abstract possibility of future arm as motivation for a robot to use the 1st law to lock humans up, it should be noted that taking away a humans freedom is generally causing them harm to some extent in the form of psychological damage - and the mental state of humans has been considered by robots as eligible for 1st law protection in at least some of Asimovs stories.

The Three Laws of Robotics require an analysis of all the laws to full answer this question:

First Law - A robot may not injure a human being or, through inaction, allow a human being to come to harm.

Second Law - A robot must obey the orders given it by human beings except where such orders would conflict with the First Law.

Third Law - A robot must protect its own existence as long as such protection does not conflict with the First or Second Laws.

If a man is walking down the street, he is not coming to harm, so the robot should not act in such a manner that assumes he may be hurt. Should a robot attempt to pick him up and place him in a box, the man can say, "I order you to stand back," and by the Second Law, the robot would be required to obey.

Isaac Asimov was the first to develop the Three Laws of Robotics in 1942 as part of his story "Runaround". Asimov said that he was tired of the common themes of robot stories of the time, which drew from the Frankenstein notion of "the created destroy the creator". He mused that if people created such creatures, they would naturally instill in them laws or ideals that would prevent them from harming humans.

The overall idea of the First Law is that the robot cannot directly, or indirectly (through inaction), harm a human. It makes no assertion about possible future events, only what is directly calculable in the next few moments. From a hard/software perspective, it is often very expensive to calculate things with MANY variables for more than a few cycles in the future. A robot cannot process all of the possibilities that could happen to the said man in the next hour, let alone day, week, or year. It could, however, see a piano falling from a few stories above the man and quickly calculate the vector required to save the man in the next few moments.

As an added investigation, read "Liar!", another short story by Asimov which discusses this question from an emotional perspective. (Both of these stories can be found in the book "I, Robot" by Asimov, I believe.)

Simply put, the definition of 'harm'.

Protection from physical harm can cause other kinds of harm. In some cases, protection from some form of harm can actually increase the likelihood of that type of harm in the future.

For example, protection from emotional harm can leave a person incapable of dealing with trivial challenges without severe emotional harm, which can propagate to actual mental harm, which can further propagate into harm to one's general health, which obviously compromises one's physical safety.

In the end, for a robot to be able to make determinations with regard to intervention in the full spectrum of human events, it must be capable of making a non-deterministic estimate of probable outcomes of a range of potential actions (including inaction), and be able to make not only objective determinations of probability and severity of harm, but also estimates of the subjective PERCEPTION of various types of harm. It must be able to do this continuously in real time as well.

Because of the complexity of problem, the simplest way to mitigate it is to restrict the problem domain by restricting the capabilities and responsibilities of the robot.

If a robot is designed to control the opening and closing of a sliding door, software can be defined which can make very reliable estimates of the potential outcomes of its actions because its actions are limited to either opening the door, or closing the door.

However, if our doorman robot is watching and listening to everything, and trying to parse everything going on around it, it may not be able to reliably determine whether it should open or close the door, given the totality of the situation. For example, if a couple are in an argument, and one of them gets up to storm out of the room, should the robot open the door, or would it be best to keep them in the room to solve their dispute? Is this person a danger to themselves or others if they leave? Will the other one be a danger to them if they stay? how will all of this affect their relationship? Will opening the door cause social harm because of the appearance of the person attempting to leave compared with the social norms and apparent prejudices of those on the other side of the door who would witness the event?

You can further restrict the problem domain by restricting the inputs. So now our robo-doorman can only perceive that a person is approaching the door, and can determine the point at which if the door is not opened, the person is likely to come to physical harm, based on their velocity and the properties of the door. Sure, the robot may not be very much help in saving a relationship, but it will predictably be able to keep you from walking into the doors like William Shatner in a Star Trek blooper.

All of this means that the robots must either be able to approach or exceed our capacity for what we call 'thought', or it must be limited to the extend that its shortcomings are less than its strengths. If neither is possible, then that task is probably better left to a human.

A) Programmers prevent or allow.

B) semantics and definitions, even philosophy

C) nothing

C)

Any unattended(read, uninfluenced) self learning system, even with a few hardcoded, unalterable instructions could devolve into something unforseen and dangerous in some way or another.

The fact that we humans haven't overcome this (see our world of extremes in bliss and suffering as it exists today and history) makes me wary of AI that is not monitored in its learning and possible self programming.

At the very least it needs to have a failsafe installed for immediate shut down or human interference.

But even if learning and action is monitored or influenced very quickly there is a point where no human is able to oversee or understand every line of code or behavioural pattern matrix or tensor field.

Also, hacking,malfunctions and viruses could always circumvent anything or crash or stall the systemt, thus rendering even failsafes useless.

B)

There will need to be crystal clear definitions of what is meant by what instruction in order to overcome ambiguities.

Again we haven't mastered this, so there is no hope for AI - an intellect so foreign once it reached its true potential that we can't even fathom its thought processes, let alone ethics.

However if the instructions are sufficiently black and white, reducing any shade of grey into a yes or no, there at least might be a way.

For instance the AI could determine that humans require freedom to thrive, even exist or be happy.
Thus taking it away would harm mankind and be an invalid choice.

Should nobody have thought of "telling" the AI this natural urge to freedom and there was no instance where it could have been observed by the AI (or dismissed as irrelevant), it could very well decide to take the action you propose.

As you see it comes down to human nature and the biological, mental and social intricacies that are sometimes so abstract and interwoven that centuries, even millennia of philosophy and science fall short in understanding them.

A)

A collection of hardcoded conditions to test against might sufficiently mitigate the risk of death or harm to humans or society.

First step would be NOT to put weapons into AI driven machinery that may roam freely.(Yeah, because we'd never do that)

Another would be NOT to hand over every vital control system to AI without manual override possibilities.(but it is sooo convenient)

Yet another would be to keep the AI simple in their range of actions and fields of expertise, making it easier to predict most, even all possible actions and setting a proper framework (at least according to the then current ideology, societal norm, law, ethical codex etc. - oh no this already falls apart as well as soon as these change).

There are many more and the more cases we think of, the more we enter B) and C) again as these are actually all the very same problem...feels like a recursive loop...

So in dealing with highly evolved AI essentially you either create tools that are somewhat intelligent (possibly sentient) and effectively enslaved(under the control of humans) or you do what we (mostly) do with humans: let them learn from their surroundings and then roam free and hope for the best and that their impulse to self preservation keeps them from going on rampages out of fear of retaliation...well, we see how well that works for us...so good luck either way...