Well of Souls Forums

Because AI will bend every rule and break any physics to obey it's orders. AI is truly a Literal Genie

yeah, but in real physics, they can't move as fast, and the exploits they find often won't work in real life.

Yes, learning to hide the ramps/tools for opponents is a lesson to be taken into real world, but chucking the boxes/ramps out of the playing field altogether will be a little harder.

Thanks for sharing, enjoyed the game.

Q: how to decide whether the Law of Robotics is applied correctly?
It'll be very hard to program.

Even leaving aside the practicalities of implementing the three laws, what gets lost a lot is that the three laws stories were often about the ways the laws go wrong. Hell, the first law makes for perfect justification of taking over the world to protect humans from themselves, depending on the definition of harm. And depending on how the AI handles the trolley problem it could potentially justify a quite violent takeover.

hi hi

And when you don't leave practicalities aside, the three laws become a real nightmare. Humans can't definitively define "human," so how is a robot going to do it, and how is it going to recognize humanness in real time?

They're just so... semantic and vague.

That's not to say that AI isn't potentially dangerous. It definitely is. If a strong AI is ever developed, it's going to be real hard to predict what will happen next.

What I see here is human behaviour. It is the eternal conflict between the have's and the have not's. Sadly, if you get a bit ahead, you must expect that there eome among those lagging behind that will not learn from your example of how to succeed. Thease people will be hell bent on figuring out how to steal yours rather then making their own. And as such, you get the ancient problem of rampaging barbarian hordes destroying attempts to create civilizations.

Siber wrote:Even leaving aside the practicalities of implementing the three laws, what gets lost a lot is that the three laws stories were often about the ways the laws go wrong. Hell, the first law makes for perfect justification of taking over the world to protect humans from themselves, depending on the definition of harm. And depending on how the AI handles the trolley problem it could potentially justify a quite violent takeover.

One of Asimovs stories give an benevolent example of this where a ridiculously human like robot hides his nature and goes into politics.

There is an interesting commentary in the youtube video. In most sci-fi fiction, humans´biggest advantage against robots is our pretended ability to think outside of the box.
This experiments proves the opposite: for the A.I, there is simply no box.

raistlin34 wrote:There is an interesting commentary in the youtube video. In most sci-fi fiction, humans´biggest advantage against robots is our pretended ability to think outside of the box.
This experiments proves the opposite: for the A.I, there is simply no box.

If a human is forced to solve a problem and is given a few million tries to do so then they will eventually think outside of the box and in far fewer tries than a A.I. would. The main problem with humans (and life in general) is that we are evolutionary programmed to not waste effort/resources solving problems that have no immediate benefit. This 'focus' is what allowed us not to 'starve' by trying to catch something beyond our reach while there were plenty of 'fruits' within easier reach.
If we really need to do something with no constraints about bodily and other needs then all bets are off our minds; thinking outside the box? A sufficiently motivated human would find an infinite number of solutions to problems.

I thought the whole point of Asimov's robot stories was that the three laws don't work.

It doesn't have to be Asimov's laws specifically, but AI definitely needs constraints to preserve human life and liberty.

Turrosh Mak wrote:It doesn't have to be Asimov's laws specifically, but AI definitely needs constraints to preserve human life and liberty.

And dignity. Even above life. You don't want to end like the humans in SOMA.

https://www.youtube.com/watch?v=k_ZGUf-uV84

icekatze wrote:And when you don't leave practicalities aside, the three laws become a real nightmare. Humans can't definitively define "human," so how is a robot going to do it, and how is it going to recognize humanness in real time?

The Freefall comic explores that a bit with robots, uplifted animals, and aliens. The whole Ecosystems Unlimited arc is fun, but the relevant part is the safeguard check through simulated attack: result 1, neutralizing the aggressor because aggression is risky; result 2, humans are identified by being clothed because even genetically-modified transhumans will want pockets.

Oh yeah, I loved that Freefall arc.

And also how following orders of Humans can result in desaster (the arc about the one robot helping one single human seizing all assets owned by robots, thus concentrating nearly all of that solar systems economy in one person's hand). And that while no Human comes to harm directly by the action.


Also:
haven't heard of SOMA yet.
Other examples would be the films Matrix, and Wall-E.
Different solutions to control Humans, and also effective ones.

Krulle wrote:Oh yeah, I loved that Freefall arc.

And also how following orders of Humans can result in desaster (the arc about the one robot helping one single human seizing all assets owned by robots, thus concentrating nearly all of that solar systems economy in one person's hand). And that while no Human comes to harm directly by the action.


Also:
haven't heard of SOMA yet.
Other examples would be the films Matrix, and Wall-E.
Different solutions to control Humans, and also effective ones.

Reminds me of the "paperclip" thought experiment.
The idea is a factory robot with the ability to improve itself is told "make as many paperclips as possible". It proceeded to convert the entire mass of the earth into paperclips, possibly preemptively genociding humanity to prevent it from interfering with the paperclip-making process.
There's a neat little browser game called Universal Paperclips that explores this.

Yeah, in the end I had converted the full galaxy into paperclips. meh.

paperclip experiment on Wikipedia