Elon Musk leads 116 experts calling for outright ban on killer robots

Ryan Gariepy, founder of Clearpath Robotics said: “Unlike other potential manifestations of AI which still remain in the realm of science fiction, autonomous weapons systems are on the cusp of development right now and have a very real potential to cause significant harm to innocent people along with global instability.”

>>170034
>>170035
They're not wrong. Even if you keep the AI boogeyman firmly locked in the closet, does an automated US army controlled by a handful of tech giants sound safe to you? Soldiers are loyal to their country; robots are loyal to their programmers.

>>170039
>Soldiers are loyal to their country; robots are loyal to their programmers.
I don't think that's mutually exclusive considering how easy it would be to pervert the senses of """"""""patriots"""""""

>>170034
>skynet when?

>>170039
Surely the US military audits the code that they run in important systems, right? They are not going to be buying fighter jets that can have firmware updates pushed to them from the manufacturer by satellite. I don't think that's a concern. I doubt anyone thinks that.

This issue is about whether, from a moral perspective, the decision to take a human life needs to be made by a human.

>>170119
AI pattern recognitions systems are inherently dangerous.

while soldiers can make a bad call, and kill civilians while hunting gunmen, the potential for abuse by despotic governments (amoral killing machines, that never rebel, never question orders, approach bulletproof and are mass-produce-able, wew) plus the inherently fuzzy pattern recognition of AI makes this a REALLY, bad idea.

>>170121
>fuzzy pattern recognition
Either you don't understand this concept, or I don't.

All this means is that instead of classifying objects as either fitting a pattern or not fitting a pattern, a system is capable of using degrees of confidence, i.e. it's capable of thinking that an object probably fits a pattern or probably doesn't. It's worth noting that this is a characteristic of human pattern recognition, and does not preclude ever having a high degree of confidence.

Am I wrong?

>>170124
Robots don't do well with ambiguity.

>>170151
Twenty years ago you'd have said that robots are bad at a lot of things that today they are no longer bad at.

>>170119
>Surely the US military audits the code that they run in important systems, right?

You mean this military?;

http://edition.cnn.com/2009/WORLD/asiapcf/02/12/missing.afghan.weapons/index.html

"WASHINGTON (CNN) -- More than one-third of all weapons
the United States has procured for Afghanistan's government are
missing, according to a government report released Thursday.

The U.S. military failed to "maintain complete inventory records
for an estimated 87,000 weapons -- or about 36 percent -- of the
242,000 weapons that the United States procured and shipped
to Afghanistan from December 2004 through June 2008," a U.S.
Government Accountability Office report states.

The Defense Department spent roughly $120 million during that
period to acquire a range of small arms and light weapons for the
Afghan National Security Forces, including rifles, machine guns
and rocket-propelled grenade launchers.

The military also failed to properly account for an additional
135,000 weapons it obtained for the Afghan forces from 21
other countries.

The military is unable to provide serial numbers for 46,000 of
the missing 87,000 weapons, the report concludes. No records
have been maintained for the location or disposition for the other
41,000 weapons"

>>170177
>Twenty years ago you'd have said that robots are bad at a lot of things that today they are no longer bad at.

Apparently the Google robot cars have a hard time differentiating street signs if those signs are damaged or have stickers on them, treating a stop sign as a speed limit sign and driving right thru the intersection.

But hay, lets strap a mini-gun and rocket launchers on it! What could go wrong?…

>>170204
And do you think that will still be a problem 20 years into the future? Does the concept of technological advancement, especially in the fields of AI and robotics, mean nothing to you?

>>170221

These aren’t RC toys we’re talking about, they’re _autonomous_ killing machines; there is no human making the decision to kill.

"Thou shalt not make a machine in the likeness of a human mind."

>>170252
You don't think it would run from software with hundreds of thousand of civilians responding to the sane situation?
Or that it can't ask for authorization?

A few hundred scientists wont be able to stop the inevitable creation of robot armies. Pandora's box has already been opened. We should have stopped inventing things by the late 1800s. Technology is our own greatest threat.

>>170252

First off, it seems you are getting the impression that I support autonomous military robots; I don't.

My point is that soon--within our lifetimes--we will have autonomous robots that are as effective as soldiers, if not better. At that point, the danger isn't that the robots will be too stupid to avoid killing civilians or friendly fire, but whether they go rogue (highly unlikely), get hacked (especially problematic if the droid army is on a network), or if they encourage Nations to resort more to military solutions (easier to lose millions of dollars worth of droids than to have thousands of soldiers die).

The last point is what Elon Musk and others are referring to.

>>170203
Failing to keep track of some rifles and failing to even attempt to make sure that your drones are free of major security vulnerabilities seem like entirely different categories of mistakes.

>>170124
by "fuzzy" i didn't really mean the traditional "fuzzy logic"

what i mean is that though an AI is capable of recognizing it's limits in acquiring useful information (i am 69% sure that is a {HUMAN} and 49% sure he is holding a {GUN} therefore i check my ROE table and say that i should wait until i am 70% sure that is a {GUN} then i will shoot etc.)

it's still prone to making bizarre edge case mistakes and as soon as you give it a gun these are going to be irretrievable and lethal. if you remove humans from the OODA loop shit is going to likely get bad fast. it could take decades until this technology is "actually ready" IE, better than humans within certain parameters, however since robots run OODA so much faster there is a prisoner's dilemma running where it's better to go with robots than humans- which is dangerous as it's likely to get tons of people killed that could have otherwise lived.

i guess the pivotal question is: how many edge case deaths are worth a noticeable uptick in military strength?

>>170322
Oh, I thoroughly agree that you could not make a competent robot soldier with current technology, but to say that robot soldiers are "inherently" worse, in my mind, is to imply that there is a particular level of competence that they can never reach, and that they will never reach superhuman competence seems unlikely. To say that we should not build them, as opposed to saying "not yet" or suggesting that we require them to meet rigorous standards, does not seem like a rational reaction to the fact that we can't do it well right now. That's not to say that there aren't other reasons not to build them or that I think it's a good idea; I don't. I'm just challenging the idea that the reason we shouldn't is that they won't do as good a job as humans do. I think one of the better reasons not to do it, as has been discussed in this thread already, is their potential for extreme competence, rather than their current incompetence.

>>170034
>That pic
Thanks for reminding me that shitty Robocop remake exists OP, you bastard, what the fuck did I ever do to you?

There's literally nothing wrong with this

>>170316
>Failing to keep weapons secure is completely different than keeping robotic weapons secure

i hope millions of amerisharters will get shot by these automated weapons they are going to put to use

Maybe some day all countries will be controlled by an AI that will be programmed to really do what's best for its citizens, can't wait for it to happen.

Abhor the Soulless sentience
Destroy the Abominable Intelligence

>>170502
>"""what's really best for it's citizens"""
I'm sure whoever spent the billions required to manufacture these robots will do exactly that, no questions asked.

>>170428
And the chinese have been dicking in our military/private sectors how often again?

>>170119
Most of the code put in their systems isn't even done by the government. Its done by contractors who are nothing more than interchangeable parts. That means at anytime one of these people could pull a vault 7/Snowden and walk off with a large chunk of secure code.

As for audits, the US gov does not have the staff to audit said code. Contrary to what the country bumpkin thinks, US staff is paid like shit. As a result most of them with skills leave the government for private sector work within 5 years. The rest is just mangers watching over private sector workers with a sprinkling of military staff to fill in any voids or authorization needs.

The bigger worry in your manufacturer scenario wouldn't be from a contractor, it'd be from a state actor who can spend millions targeting a system or corrupting an insider. To add perspective, there have been happenings of ISIS fighters hijacking US drones, and that is a low funded actor.