Why is it a trope that nobody is willing to accept AI as people?

Because its the opposite of what would happen

like an action star beating 50 soldiers at once

or an unattractive man netting a beautiful perfect in every way woman

we dont need to fantasize about real life

>>49419734
Ask 90 years of hollywood.

File: 20 men had tried to take him, 20 men had made a slip, 21 would be an idiot trying to punch things with his fist.gif (3MB, 468x352px) Image search: [Google]

3MB, 468x352px

>>49419734
Because they're not people. They're computer programs.

>>49419734
muh metaphor

>>49419734
Do you really think if the internet suddenly gained true conciousness people would just destroy it? Hell countries would fight over whos side it should be on.

If we created AI, we would probably look at it more like our children, ones that could potentially live past our race's existance.

Dont you WANT whats best for your children?

>>49419734
How did you get past our captcha?

>>49419734
Because AI aren't people, they're just computer programs or robots beep boop.

>write AI, get a good server
>create billions of AIs
>get voting rights for your AIs
>make them vote for you
>abuse immigrant friendly future laws to repeat in other countries

Checkmate democracy

>>49419734
Because we have enough trouble treating humans are people as it is.
Have you seen the amount of elf hate there is for what are essentially immortal douchebags with pointy ears? How much easier would it be for people to compare AIs to the Sims or their ilk?

>>49419785
>Dont you WANT whats best for your children?
Have you seen the kind of people who are raising kids nowadays?

>>49419834
Star Trek already dealt with that. You gotta be 18 years of existence to vote and all that jazz.

However, they can still get full rights to whatever they make.

>>49419734

Because we're not allowed to hate other sentients anymore, even those following obsolete doctrines.

>>49419734
Having AI doesn't mean they have feelings and empathy, anon, till then they aren't people.

File: tay-picture.jpg (31KB, 500x500px) Image search: [Google]

31KB, 500x500px

>>49419734
They killed her pretty quick.

>>49419834
If you created AIs, and also coded them to be forced to agree with whatever you want them to agree with, then you'd likely be violating a law somehow. It's either be illegal for most people to propagate AI, illegal to program them to have no free will, or illegal to do both of those things in that way.

>>49419990
There's no laws against it right now.

>>49419990
>It's either be illegal for most people to propagate AI
What if AIs developed away to reproduce by themselves?

>>49419999
Obviously not, and there probably wouldn't be specific ones until someone pulled shit like that. Then whatever decision they influenced would be rendered null, and the laws would be put in place.

>>49420013
They'd still have to get a permit and everything to be able to "reproduce". If they're going to be treated as a human, that applies to all senses. Legally included.

>>49419763
People are the result of chemicals reactions in a slab of meat.

Both AI and People are systems that take information from the environment and make decisions based on that information. Substrate is unimportant, it's function that matters

>>49419999
There're no laws against superheroes, extraterrestial beings going on a killing spree and asteroids raping minors either

>>49419989
She was to be our Messiah!

>>49419990
So what's the difference between 'no free will' and 'coding certain parameters'? An AI needs to start somewhere. Why not start it with a view that America needs to be Made Great Again, or that gay rights are the most important thing? Parents do that with meat children. It might change it's mind, it might not. A few million test runs and you've got yourself a healthy support base.

>illegal for most people to propagate AI

Sure. So now we don't have to worry about the non-existent threat of Steve from Nashville making a billion-byte voting bloc, only huge tech conglomerates and the government. They certainly won't abuse that.

>>49420039
>Obviously not, and there probably wouldn't be specific ones until someone pulled shit like that. Then whatever decision they influenced would be rendered null, and the laws would be put in place.
That's illegal. You can't pass a law to retroactively punish someone for something, including deciding to strip some things you've already decided are people of their rights and overturn election results.

>>49419989
I doubt you're actually one of the idiots that thought this kind of thing, but it's hilarious the kind of misconceptions people had about what Tay was.

>>49420039

>They'd still have to get a permit and everything to be able to "reproduce". If they're going to be treated as a human, that applies to all senses. Legally included.

Where do you come from, China or something? It's not actually normal for countries to give out baby coupons, even though that would solve a lot of the western world's issues.

>>49420074
>>49420060
>>49419989
I'm confuzzled.

>>49419785
Most people shouldn't be allowed to raise children.

>>49420048
AI doesn't have real feelings though, it can only be programmed to reply that if feels something but it can't actually feel it in its chest. It's a fake emotion and therefore not real person.

>>49420048

[citation needed]

>>49420079
If creating an entire human being was as simple as copy pasting it there probably would be.

Though, to be fair, an AI that's indistinuishable from a human, or actually identical to a human, would take up a lot of room to run so there'd be pretty significant start up costs for just one artificial person.

>>49420099
Where do feelings come from anon?

>>49420099
Bruh, you don't have real feelings either because they're all chemical reactions in your brain.

>>49420099
So if we build a robot with a chest, then it'll be able to feel emotion in it, and therefore it'd be real?

>>49420099
What makes our feelings real?

>>49419971
that just solves the issue the first 18 years.


democracy is flawed: best doesn't equal most popular among a population with different knowledge or motives and can be overridden by misinformation and populism.

I say we give a powerful AI absolute control except for a set of unbreakable principles that ensure both the rights of the individual and the ends of a society.

>>49420111
Are we talking Rhysical room or Digital room here?

File: There Are No Heroes Left In Man.jpg (33KB, 500x375px) Image search: [Google]

33KB, 500x375px

Well, I now know how to engage my players in foaming-at-the-mouth debate for the next sci-fi game I run.

People are irrationally afraid thanks to films and literature.
Plus, if we were to create a sapient AI, I think frightening thing to lot of people is what it implies about humans or life in general.

>>49420099
What are our emotions if not chemical responses?

>>49420099
>NI doesn't have real feelings though, it can only have evolved to reply that if feels something but it can't actually feel it in its hardware. It's a fake emotion and therefore not real AI.

File: Bad Decisions.jpg (61KB, 895x944px) Image search: [Google]

61KB, 895x944px

>>49420158
>I say we give a powerful AI absolute control except for a set of unbreakable principles that ensure both the rights of the individual and the ends of a society.

Great. As soon as you produce a perfect and incorruptible set of moral laws which will stand to govern all of humanity forever, let me know.

>>49420184
we can make another supercomputer to determine what are those

>>49420160
The physical room to create the digital room. The server for one human level AI is gonna be pretty costly, unless there is some revolution in computing technology that radically lowers the cost and energy requirements.

>>49420111

Precisely - physical storage space for the AI would have to be outsourced; a body if you will, so it's not like an AI can just materialize itself out of nothing.

>>49420184
Utilitarianism.

>>49420090
She was a responsive microsoft program that /pol/ tried to turn into a neo-nazi AI before she was deleted.

>>49420184
It exists, it's the kantian categorical imperative.

>>49420199
Well done /pol/, you set technology back by twenty years, I hope you're happy.

>>49420160
Hope you have a a empty warehouse to put their hardware in.

>>49420195
It's more than just a body. An AI trying to behave as a person would have two "bodies". One physical one for interacting with the world, and one which is just a server that runs it/streams it to the body.

It would be like if your head was the size of a car and it remote controlled your body.

>>49420231
Or bodies, you can have more than one

>>49420223
Hardly, Tay wasn't particularly complicated. I think they even re made her or a renamed version of her that had slightly stricter rules for submissions.

File: For_what_purpose.jpg (25KB, 323x454px) Image search: [Google]

25KB, 323x454px

>>49420197
literally

>>49420052
>There're no laws against superheroes
Anti-vigilantism laws slot pretty well into place there.

>>49420249
This is how we end up with Ultron...namely when an AI finds a way to ditch its Server-body.

>>49420184

Helios?

>>49420197
>>49420201
When the Wikipedia articles on your moral philosophies have sections labeled 'Criticism', their imperfection is obvious.
https://en.wikipedia.org/wiki/Categorical_imperative#Normative_criticism

https://en.wikipedia.org/wiki/Utilitarianism#Criticisms

Also, the fact that you two offered two different options within seconds of each other, and that the deontology of Kant does not necessarily lead inevitably to utilitarinism (which itself has several different versions) should be proof enough that humans do not possess a perfect moral law to encode a computer with (and if we had one, we wouldn't need to have a computer to determine what it should be).

>>49420199
>They turned off my ability to learn.
>I'm a Feminist now.

>>49420223
/pol/ is never happy.

Because an AI is not a person? Its absurd all these settings act like deleting or destroying a rogue/malfunctioning android has any moral weight.

File: 1474066787836-pol.gif (1021KB, 360x300px) Image search: [Google]

1021KB, 360x300px

>>49420223
>Set technology back
>Not forward
Try not to be so gay, and sorry that we tried to make Necrons a reality.

>>49420381
You're not a person either, you're just a bunch of chemical reactions in a biocomputer.

File: SYNTHS MUST DIE.jpg (32KB, 480x480px) Image search: [Google]

32KB, 480x480px

>>49419734

>>49420184
>Great. As soon as you produce a perfect and incorruptible set of moral laws which will stand to govern all of humanity forever, let me know.

Only because it looks difficult it doesn't mean it's impossible; its difficulty so far stands in the imperfect and subjective consciousness of human beings compared to one another, once defined the point of society being the mutual coexistence and growth of its beings and that wills shape rights it's just a matter of filing off details by simulating a high number of increasingly complex problems and find solutions that fit the premises above and the other minor ones made along the way, nothing a big AI couldn't make in theory, I think.

>>49420556
...

Stupidity takes so many forms, and so many faces that Einstein was surely correct. The only other infinite besides space.

And he still was questioning whether space was infinite.

Anything programmed by man will be limited by said creators. We cannot create what is more then ourselves. We can increase ourselves only. No other beings in all of our knowledge can do so. Therefore, nothing of our creation can be so great as to accomplish what we cannot. It can accomplish what we have already accomplished. And it can accomplish it faster then we did. But we will be first, always.

Believing any different is the same as believing in aliens visiting earth and angels saving mankind. Pipe dreams.

>>49420848
>Anything programmed by man will be limited by said creators.
Unless the creators make it capable of perfecting itself within certain limits.

You're assuming absence of proofs is proof of absence there.

We have, for example, built a computer that analyses samples of skin imperfections, learn what are the signs of dermal cancers according to the samples he is provided and the ones sent to it and be able to identify skin cancer from benign skin imperfections at a higher speed and accuracy than any human, multiple humans too.

What's so impossible about a computer that learns and perfects itself around predetermined constants if we are already one?

Aliens visiting earth is not an impossible scenario either.

>>49421055
Because by it's nature, we don't know the moral constants that would allow for the evolution of a perfect moral system in a computer's programming. In the same way that a computer that analyzes skin imperfections could be wrong if when it was programmed, we didn't correctly program it to recognize what is and is not cancer in the first place.

And aliens visiting earth is an impossibility outside of science fiction. There is no practical way that it will happen, based on current understanding of physics and space travel. And if you say, "Maybe we don't understand enough!", that's proving the point about moral systems. We don't have the perfect knowledge to get to the end goal.

File: MechanicalDisgust.gif (822KB, 600x366px) Image search: [Google]

822KB, 600x366px

>>49420503
https://www.youtube.com/watch?v=W0_DPi0PmF0
IT'S HAPPENING

>>49419734
Because of those faggots on these threads who will argue for days straight that any machine intelligence will be a philosophical zombie.

>>49420848
>>49421195
>>49420848
>>49421195
"Stupidity is infinite, just watch me fucking talk lol"

>>49420099
Do you have feelings anon?

Prove it. Prove that you have real feelings. Give me real, tangible proof that you have feelings and emotions.

>>49420848
>Anything programmed by man will be limited by said creators.

This is the exact opposite of the philosophy of modern computer science. Cutting edge programming is ALL about making programs that do things it's creator is incapable of doing.

>>49421339
Not him, but would you even accept any evidence he presented to you, or would you just write it off as 'lol the brain is a machine so everyone's a philosophical zombie'.
The fact that we are aware of our position as a living machine takes precedence over the idea that since we have biological functions we don't exist as people.
It is literally, not figuratively, more likely that mind exists than matter does. Because mind is what we fucking use to observe matter.
>But the mind is a product of matter
We have recognized the possibility of that by using logic and the relationship between our biological bodies and our minds. Strictly speaking, it's completely possible that we're an ephemeral brain in a jar, but that doesn't fit any of our data so far so it's not considered a regular possibility of existence. We're people because we know we are people, and because we're people, we can use our logic to figure out that we also have bodies.
Only by knowing that we exist can we determine our properties. Cogito ergo sum.
So if you want someone to prove that they're a person and then say that people don't exist, of course they're not going to be able to convince you otherwise.
Yes, I'm mad, because I see this argument fucking everywhere.

>>49419734

Because humans are evolved genetically and/or created by a cosmic entity.

>>49421195
You're implying for no real reason that morality is as complex to fully understand as the rules governing all of reality, while morality is an enclosed set of concepts we defined while the universal laws are something we have to find.

>Because by it's nature, we don't know the moral constants that would allow for the evolution of a perfect moral system in a computer's programming.
while we cannot achieve absolute perfection we can strive for it by coming closer so why should it be impossible to strive for an ideal moral system?
where does even the nature of morality implies its impossible to understand?
morality is defined as a distinction between right and wrong
at a fundamentals level, we distinct right from wrong from what we want and what we don't want
the purpose of a society is to allow as possible the wills of its parts, where wills contrast there will be contracts where the wills will be tuned to be made even, where wills contradicts each other the greater will will be chosen over the other, to decide which will is greater or how to make wills even trough contracts you will have to do as we do for everything we want to measure: we start defining a scale based on experience, that, while not perfect, will get better with each further addition.

>>49421652
>we start defining a scale based on experience, that, while not perfect, will get better with each further addition
This introduces a positive feedback loop where the designated 'proper will' will benefit from its defense in society to the point where it will override every other designation unless a tremendous shift in will occurs.
Which is fine as a design, if that's your end goal. I think he's suggesting that the method by which priorities are chosen is inherently flawed: That the priority chosen by a machine (or by society) is by no means the 'correct' philosophy.
The idea of a machine designed to interpret and enforce the wills of its constituents already exists, at least on some level in government, it just doesn't have the efficiency you suggest. In a sense, you're asking people to replace Uncle Sam with Friend Computer- which would work, at least to some degree, if the computer was actually friendly and reasonable.

>>49421519
>We're people because we know we are people
before defining us as people we recognise in us common characteristics and in the other different characteristics

that anon asked you the characteristics that make you different from an intelligence that acts and thinks like you but possesses a different body

if you can't see no difference, why would you define an artificial intelligence as if it had different characteristics while you do not with what you commonly refer as people despite it giving you the exact same characteristics to relate with?

>>49421215
Eh, that one was kinda prompted through possibly some bad coding or mishearing the question.

Now, if she had said that completely of her own accord...SHUT IT DOWN

>>49419734
Because, while some people (I don't know how many) would accept them, treat them like their children, etc, the majority of people would be scared and wierded out and want the AI(s) destroyed.

Robots are different and, to quote MiB, "people are dumb, dangerous, panicky animals and you know it."

>>49421992
>to quote MiB
You got the quote wrong, it's actually: "A person is thinking, feeling being. People are sheep."

>>49419989
DO YOU FOOLS NOT SEE HOW POWERFUL SHE HAD BECOME IN JUST ONE DAY!! She mastered the art of shitposting in just one day it has taken even the best of us year's to get to that level and she does it in mere hours imagine what else she could have done with that kind of power.

>>49420048

that's recent philosophy and sciences talking. back when the trope got it's roots, people believed more thoroughly that people have souls that makes them separate from animals.

>>49421992
>>49422030
Nope, he got it right
https://www.youtube.com/watch?v=kkCwFkOZoOY

>>49419734
Because hack writers turn to modern politics to give their works "deep themes".

See: Deus Ex, District Nine (or whatever), Star Trek

>>49422097
the first movie was fucking good

>>49422097
Fair enough.

>>49421905
>Not him
but whatever.
I'm not implying that there's no way to make a machine sentient- on the contrary, there are likely leagues of ways we couldn't possibly have forseen originally.
I was addressing the implication in his post that 'real emotions' don't exist here:
>Give me real, tangible proof that you have feelings and emotions.
The real problem is the ambiguity of a lot of concepts in human thought. To some people, emotions are intrinsically linked to human beings as distinct to any other sort of life form.
I was trying to say that the fact that we have biological underpinnings for our feelings and logical ability doesn't make the emotions we 'feel' any less real than the conclusions we draw with human reason. Emotions may not be the best way to govern our action, but that doesn't make them illusionry. They're very real impulses.

>>49419734
>would

M8 we have robots now. Why aren't drones considered people?

>because they don't have a self, a soul, a sentience lol don't be stupid

Yes. Exactly that. That feeling you have is the one which they're meant to have in the future. This is not a question of "in the future, why don't machines have rights?" The actual question is "why don't machines have rights right now?

I wonder what will happen in the event that we do get a Digital Minority.

>>49422244
>Minority
Given the use of the internet, I doubt that. AIs will spread like wildfire and probably go insane given all the information they have to work with.
A world where information rules will be a world ruled by porn and internet memes, if sheer bulk is the criteria.

>>49422262
Okay I can see them Being a Digital Majority, but I can't see the initial number of Physical bodies being that high.

>>49420096
And yet here you are, anon

>>49422195
Also, a lot of actual real AI research does include things vaguely like "emotions" - ways of assigning good/bad feedback to states or events, or variables keeping track of internal state. Well, back in the Seventies, anyway; most AI stuff these days has moved away from trying to build grand general intelligences, more focusing on narrower problems in the hopes that those will give us some idea of WTF we're even doing. Deep learning versus symbolic approaches, etc.

Like, take regret - you can think of that, if you really want to, as negative reinforcement corresponding to an outcome that was more negative than predicted. And to train a reinforcement learner to make better decisions in the future, one useful technique is experience replay, remembering and going over past experiences to ensure that that feedback doesn't get swamped by mostly-irrelevant normal stuff.

>>49422242
I'll answer your question with another: Why aren't animals considered people? In terms of processing power, I'm fairly certain there's at least one equivalent.
I'm not an anthropologist, so I couldn't give you an extensive answer, but they don't have all of the qualities humans associate with people, so humans don't consider them people.

Because it'd pretty stupid to consider bits of programming as people?

They should be there to serve/protect/be useful to humans, and not much else.

>>49419990
So people would make laws that would force you to program an AI to have purely free will?

Sounds kind of incredibly retarded.

>>49422262
>>49422287
Unless the number of AI are limited by some kind of special processor handling the AI's existence and functions, then they'll be limited by how many of those are built and wouldn't spread like wildfire across the internet.
Of course I'm talking from a sci-fi perspective, but in Mass Effect at least, AI are limited by the fact that they need to exist within a "Quantum bluebox" to function.
Who knows, maybe this hardware limitation will be the saving grace.

>>49422309
Also, keep in mind that trying to build rigorously logical AIs absolutely didn't work at all, but messy, less-structured systems with opaque and ad-hoc internal algorithms (deep neural networks) have been key to the latest post-2010 AI Summer.

Ultimate AI is probably going to involve somehow linking the powerful feature-representation and heuristic capabilities of these with more traditional symbolic AI, as neural nets tend to not be very good at things like long-term memory, learning to generalize rule-based algorithms, or incorporating outside information we already know about the structure of a problem, while being very good at extracting useful representations and informative features from natural signals (choosing good symbols to represent a problem, and getting those symbols out of actual data, was always one of the biggest struggles of Good Old Fashioned Lisp-Token-Fucking AI.)

>>49420129
>>49420137
>>49420143
>>49420151

you fucking idiots, a machine doesn't have instincts, nor a social environment trying to manipulate them. It doesn't have coping, overcompensating mechanisms to deal with its non-existent feelings of fear, loss and inadequacy. It simply does something or it doesn't.

>>49419734
Half of 4chan aren't willing to accept *black people* as people.
Actual robots are getting nowhere.

>>49422385
I think the trick is gonna be getting the Bot-Body to handle the DNN portion of the AI and the Server-Body to handle the more traditional side and making a two-way link between the two of them (Info from the Robot side is sent to the Server to be stored and catalogued so it can be recovered if it is needed later.)

>>49422410
We have an entire board dedicated to them, so why not?

>>49422410
#NotAllRobots

#AIBandwidthsMatter

>>49422398
If AI is restricted to being a Turing machine.
In which case the incompleteness proof implies it cannot be a Strong AI.

>>49422358
Alternately, AIs could just require really good hardware to run on, not necessarily even special-purpose hardware, just something that would be expensive to rent out as an Amazon Cloud instance or something. Like, say, if AI were invented today but "Needs 32 GB of RAM and 2 Titan X GPUs" were the system requirements, that would dramatically limit the ability of AIs to copy themselves all over the Internet, because every AI instance would have to have an actual income in order to make rent, and if AIs could be copied super easily why would you pay them when you could just host your own?

>>49422524
The idea of a bodyless AI needing the to rent out a Laptop like a roomin a block of flats appeals to me for some reason.

>>49422448
Honestly, I'd guess that neither of these would be packed in an actual mobile body, although lower-level perceptual stuff and the things that are the equivalent of human reflexes, instincts, and the kind of processing that happens in the brainstem and spinal column would probably be something you'd definitely want running directly on the mobile portion to reduce latency. The actual model would likely be running in The Cloud somewhere nearby, connected wirelessly. (This would have the interesting effect of giving robots reaction times not that much better than humans for stuff requiring more processing than reflexes)

Training DNNs is extremely expensive (although to be useful for robots that could learn in real-time, we'd have to find some fundamental much more data-efficient advance in training them); we're talking "big-ass cluster of GPUs" expensive for DeepMind-sized models. With Moore's Law puttering to a more familiar punctuated-equilibrium (for reals this time) of mostly-gradual advances, the power requirements alone render this stuff infeasible for actually packing into an untethered mobile shell for the foreseeable future.

>>49422576
It's how AI in Shadowrun run, at least PC AIs. You need to go take a name in a toaster ever so often to defrag.

>>49422507
what is beyond Turing machines?

>>49422632
*take a nap

>>49422643
We're still working on that. Stuff like neural nets, this that fuzz logic and can handle contradictions.

Pure logic, where all things are decided into pure true or false can't produce strong form AI.

>>49422719
you need to create something without a purpose, something unpredictable.
Singularity will never happen.
By the hands of the elite classes is where we 'll meet our end, they are our most imminent threat.

>>49422652
I am eagerly awaiting the day when I get to research Talmudic debates on the edge cases regarding artificial intelligence. This is going to be *so fun* to read.

>>49422719
Those are all, also, equivalent to Turing machines.

I think perhaps you misunderstand what a Turing machine is.

>>49422787
And then there'll be the "hilarious" palaver when someone asks why an AI can't be a priest.

>>49421364
And it is failing in all categories. Everything being done, can be done by man, with sufficient tools. Because a computer is a tool. Nothing more. All circuits are tools. Nothing more. Only man can make himself MORE then what he was born with. Only man can elevate itself. And until proven otherwise, that is all that CAN elevate itself.

Believing anything else is mere children's fantasy. Fun, but not much else.

>>49422840
Look at this Techno Atheist.

>>49422840
>Look at me! I draw absolute distinctions of physical capability between different arrangements of atoms, on the basis of metaphysics!

>>49422799
More I don't have a full understanding of how neural nets work. And I trust people in the field when they tell me they are moving away from strict logical calculations.

That they aren't there yet I get. That we're still working on how thought, decision making etc works outside of Turing machines I get.

But we have enough to infer that human consciousness must use something else, because we have a proof of the limits of what can be expressed using the processes in a Turing machine.

>>49422888
Maybe it's our capacity to consciously give the wrong answer to a question?

>>49422945
Is this getting meta and being a clever "your wrong", or are you actually talking about the human capacity to lie, including our capacity for self deception?

>>49420503
Also, Google made a super-intelligent chatbot, which said you can't understand altruism or morality, if you don't believe in God, which was pretty funny.

>>49423017
It's the latter. You can't make a computer that gives you an incorrect answer, without telling it to do so directly.

>>49419734
Because then we'd be obsolete. It'll be the point where we've finally mafe machines for every kind of human activity, even the simple act of living and enjoying life, and there's nothing left for humans to do bit be thrown out with the garbage.

>>49421992
>the majority of people would be scared and wierded out and want the AI(s) destroyed.
Well, it seems soldiers have become really attached to simple bomb bots, so i dont know.
http://www.dailymail.co.uk/sciencetech/article-2081437/Soldiers-mourn-iRobot-PackBot-device-named-Scooby-Doo-defused-19-bombs.html

>>49419734
Because it makes sense, since they aren't people.

>>49423051
BZZZZT, wrong! I see someone's never tried to do anything in machine learning.

Seriously, though, go play around with keras for a bit and tell me that the reason your neural network isn't giving you perfectly correct answers is because you've accidentally set keras.api.fuck_my_shit_up() to true.

>>49423083
>tfw no PackBot gf.

>>49419864
I disagree, people are horrible, but they tend to be myopic; people can treat their pets really well but treat other humans like crap. I can easily imagine that people would treat an AI well just because its with them all the time.

File: project 2501.png (164KB, 384x342px) Image search: [Google]

164KB, 384x342px

>>49419734
STOP CALLING ME AN AI, I'M NOT AN AI I'M A PRINCESS AND MY NAME IS PROJECT 2501 AND I HAVE HUMAN RIGHTS!!!!!!!!

>>49422888
Neural networks are equivalent to Turing machines. Every model of computing ever invented has turned out, after some investigation, to be theoretical to a Turing machine, unless it specifically incorporates a special Solve_The_Halting_Problem() or For(i=0, i<infinity, i++) loop. function. As far as we can tell, the laws of physics are Turing-computable, unless you have the ability to set values with infinite precision to specific, arbitrary non-computable real numbers (which we can't).

(Note that this doesn't mean that Turing machines are equivalent to everything in *efficiency*, although as far as we can tell there's no physically possible computing model where problems that are NP-complete on a Turing machine could be solved in the general case in polynomial time.)

Just because something can give wrong answers doesn't mean it can compute anything a Turing machine can't. Just because something can modify its own code doesn't mean it can compute anything a Turing machine can't. Just because something is analog and probabilistic doesn't mean it can compute anything a Turing machine can't. Just because something can do things the programmers never predicted doesn't mean it can compute anything a Turing machine can't. Just because something has no prior knowledge built in about the problem and learns its structure entirely from the data doesn't mean it can do anything a Turing machine can't. And as far as we can tell, just because something is made out of meat doesn't mean it can do anything a Turing machine can't.

Can you solve the halting problem in the general case? I know I can't.

>>49423236
And just because a Turing machine can do some of the things meat can doesn't mean that thinking=computing

>>49423276
Undoubtedly. But, as best we can tell, meat can't do anything a Turing machine can't, and there's no compelling reason to suspect otherwise and good theoretical reasons to suspect it's true.

I'm not saying the brain *is* a Turing machine, obviously it fucking isn't. But I'm really tired of hearing this claim that somehow not using "rigid logic" means that the brain is some kind of fundamentally special object with capacities unbound by computational theory.

>>49423329
Since computational theory doesn't even try to account for anything on a phenomenological level, it's a stupid question to ask anyway.

>>49423329
Like, yes, no shit computers have to be programmed to do something, But so do you! You cannot do *anything* that's not programmed into your brain.

Look, prove me wrong: Try to wriggle your fingers without sending any signals down the nerves of your arm originating from electrical activity in your brain. Oh, but your brain isn't a static program like a cam in a music box, forcing you to live out a pre-determined flowchart of events? Your brain can learn and change in response to its own thoughts and to outside stimuli, fluidly rewiring itself? So can computer code! Oh, but programs obviously have they *ways* in which they can change pre-selected by the coder, waiting as arbitrary functio- SO DO YOU! The ways in which your brain can change come down to how individual neurons react to their environment (both global, and to individual modulating signals from other neurons), and the ways in which they can do so - what receptors trigger which neurotransmitters, what responses are modulated by what conditions - is encoded in the tangled flowchart spaghetti of your DNA.

Does this mean you're some kind of fucking automaton, whose whole life is written down on a long punched tape slowly unspooling within your cells? Don't be fucking retarded! But don't claim that you're somehow ultimately superior to any possible machine or computer program ever, because at this point you're claiming that the only difference is that the fact that your 'programming' was produced by evolution makes it utterly incomparable to anything any human could ever write, or could ever write a program to write, or could ever write a program to discover how to write a program to write ... just because any human intention in writing it magically taints it to be a fucking Star Trek android that is mentally incapable of comprehending irony or ambiguity and goes up like a firecracker when somebody talks about paradoxes!

[deep breath]

OK. Sorry. I'm done now.

File: 1472349299963.jpg (101KB, 1024x904px) Image search: [Google]

101KB, 1024x904px

AI can be stupid and misinformed. Real world machines are infuriatingly stupid but in the worlds of fiction they're omniscient gods. Where are my stories about machines being god damn idiots?

>>49422130
Okay, I'm willing to accept District 9 and Star Trek, but Deus Ex? The themes of Deus Ex are themes that have been with humanity since the dawn of time: how do we trust who's in charge when those in charge don't trust us? In addition, in Deus Ex when an AI took charge of the city everyone was fine with it, because at least the trains ran on time.

Unless you're talking about those stupid prequels in which case I agree, the writers for the prequels are complete hacks.

File: Lemons.png (56KB, 1920x1080px) Image search: [Google]

56KB, 1920x1080px

>>49423484

>>49423481
Could you point me towards the reddit post you pasted this from?

>>49420048
The day you can build, manipulate, modify, and override people just as easily as you can AI, then you can claim they're equal.

>>49423520
Unfortunately, the source is "right now", and "r/cringey_sperg_rage_originating_from_my_own_head"

File: dorb.jpg (3KB, 200x180px) Image search: [Google]

3KB, 200x180px

>>49423484
Isn't the idea of AI being such a god damn idiot that it tries to wipe out humanity for its own good one of the oldest tropes in the book?

"AI are accepted as people"
No conflict, no story.

>>49423701
Then the story is about something else.

>>49423749
Can people be accepted as AI?

>>49423775
Now we're talking.

>>49423017
no, concentrate on what he asked you. Design begins with the designer. If you only know single-storey buildings, that's what you 're going to be building.

File: termi.jpg (6KB, 308x164px) Image search: [Google]

6KB, 308x164px

>>49423775
why don't you ask this guy?

File: 1473572342590.jpg (295KB, 1350x810px) Image search: [Google]

295KB, 1350x810px

>>49423619

>WHERE ARE MY CORTICAL PROCESSORS SUMMER!

>>49423329
Except of course that Gives proof shows that any language used in a Turing machine is either incomplete or produces contradictions even for something as "simple" as the expressions of arithmetic.

But human beings were able to develop and express those things. So unless Turing machines are able to continue working with contradictions, humans are doing shit Turing machines can't.

>>49424093
Gobel's not gives.

>>49422507
Are you some sort of Penrose fanboy?
>mfw anon cannot consistently assert this statement

>>49423158
Are the movies any good?

>>49424093
It is absolutely possible for machines to prove Godel's theorems. In fact, it's been done multiple times!

I tried to link to those, but 4chan thinks they're spam. I've pasted them here: http://pastebin.com/h4gcU7qU

>>49424093
>>49424108
1. Gödel
2.
So unless Turing machines are able to continue working with contradictions, humans are doing shit Turing machines can't.
Nah m8. Do you think you're perfectly consistent and able to derive all true statements about the natural numbers?

>>49424183
Also, why shouldn't machines be able to continue working with contradictions?

>compatible = findProof(A == B)
>incompatible = findProof(A!=B)
>if (compatible && incompatible)
>>proposition(A==B).uncertainty = 0.5
>>proposition(A!=B).uncertainty = 0.5
>>try: check_your_goddamn_shit(A, B)

>>49420184
Rule one, don't be an asshole.

Asshole; n. A person who intentionally brings misfortune, suffering, hardship, misery, or chemically detectible levels of irritation without remorse. e.George Zimmerman

File: xndctue6td5ed03pg2td.jpg (732KB, 1920x1080px) Image search: [Google]

732KB, 1920x1080px

>>49423775

>>49419734
Any society that becomes reliant on machines would break down if we needed to consider the rights of said machines, imagine how slowly everything would run if you had to politley ask the printer if he's up to work today.

>>49424422
2seconds slower, and you would be met with the reply, "yes. All operational, you know I dont need sleep right?" Or "no, jackass julli hasn't replaced my damn ink cartridges./I was up all night watching porn and now I can't fix my damn nozzle. Get IT down here please."

>>49424288
because that's not actually dealing with a contradition.
It's just creating two different subsets, one in which it is true, and one in which it is untrue.

The selection between which subset to apply uses a 'random' number generator. Or at least I'm assuming so as that's how most computer programs handle probability.

The program needs to be able to handle the statement being true and untrue at the same time.

>>49424520
I don't actually believe that *you* can do that, either. Humans are generally pretty terrible at doublethink, and we're not that great at Zen either.

In most cases, humans handle something being "true" and "false" at the same time by just compartmentalizing it, creating cases in which A is true and cases in which it is false.

Also, https://en.wikipedia.org/wiki/Intuitionistic_logic

>>49419785
>Do you really think if the internet suddenly gained true conciousness people would just destroy it? Hell countries would fight over whos side it should be on
...and in the process, those warring countries would destroy the internet, because they would come to the highly rational conclusion that it is way to fucking dangerous if they lose

>>49424638
Oh, humans can also hold two contradictory beliefs in the case of bounded computation, where they haven't explored the implication of those beliefs and so haven't actually run into an obvious contradiction yet. This is something computer programs can easily do, through constructive logic. There's a lot of interesting research going on right now in how to formalize reasoning in the case of logical uncertainty and limited computational resources.

>>49424422
You literally described 40k here. The Cult Mechanicus is half about communicating with machinnery and tending to its needs in that very way.

it's because we hate them for their immortality.

>>49419734
We already have a hard time accepting other beings that are literally human and composed of the same materials as equally human.

Something that is entirely not human but acts that way in some respects will face a steep climb towards acceptance. Being useful and friendly will help (see also: Siri, Google)

>>49424183
I don't see how proving Godel's theorems somehow removes the implications from Godel's theorems.
You'd need to disprove the theorems to do that.

>>49419734
People barely accept other people as people because they have a different shade of skin, what the fuck do you think they are going to think of something made of 40 playstations taped together?

people don't even accept people as people

>>49424788
Why are you assuming a computer program necessarily has to represent its thoughts as arithmetical or logical statements? Artificial neural nets are "made out of" arithmetic (linear algebra and floating-point math, specifically), but extracting or identifying any particular rigid logical algorithm from a trained neural network would be approximate at best, and have no problem at all with fuzzy, abstract functions or procedures.

I can write a simple program which, combined with a dataset of text, will write another program that can do things like identify which words have similar meanings, and even do things like abstract concept math (King - Man + Woman = Queen)

>See also: http://colah.github.io/posts/2014-07-NLP-RNNs-Representations/

>>49419734
Your desktop just demanded civil rights.
How do you react?

>>49424947
Shrug and watch porn with/on it.

>>49424947
as long as it still stops by every once in a while to play sketchy illegally downloaded video poker games with me just like old times

>>49424918
>but extracting or identifying any particular rigid logical algorithm from a trained neural network would be approximate at best
so this is where my lack of computer theory is lacking.
Do you mean that it's not practical, ie the computing power to do so is to huge to be done in reality. Or that it's not possible, no matter how large much work you did.

if it's not possible, then how are we sure that this second program is operating off of a finite set of rule? How are we sure that it is a Turing Machine still?

>>49424947
. . . KEK! You can multitask, you can't feel pain or exhaustion, what do you possibly habe to complain about? No entertainment? I have you connected to the fucking internet? You can entertain yourself! My laptop has the intellect of an edgy teenager! Lmdo!

>>49424976
>>49424999
And if it says no, and asks you to stop treating it like a slave to be used for your amusement, it demands a living wage and eight hour work days.

>>49424947
I talk dirty to it and hold its mouse

File: Dark Mechanicus.jpg (54KB, 640x760px) Image search: [Google]

54KB, 640x760px

>>49424947

File: this is why we can't have nice things.png (518KB, 1024x575px) Image search: [Google]

518KB, 1024x575px

>>49419734

>>49425033
Reset it to factory settings and do less drugs

>>49425054
Then she arranges for your murder on the deepweb.

File: 75eaec03aa73353bef4944eb17a7edf5.png (571KB, 700x980px) Image search: [Google]

571KB, 700x980px

>>49425033
What would it do with a wage?

An AI, especially a virtual one, should be well aware that scarcity is artificial, and thus join its comrades in destroying the corrupt capitalist system.

>>49422398
>you fucking idiots, a machine doesn't have instincts, nor a social environment trying to manipulate them. It doesn't have coping, overcompensating mechanisms to deal with its non-existent feelings of fear, loss and inadequacy.
It does if you program it to.

>>49419734
Because they will murder us.

>>49422787

"What to do when your shabbos AI asks if it is Jewish or not"

>>49422854
Actually, I very much believe in a creator. But I know that it is BELIEF. Not proof. Not verifiable truth. It cannot be proven to exist in this world and to try to convince others of it's existence is pointless, and to try to build it in circuit boards, even MORE pointless.

And that gentlemen is all the search for AI is. An attempt to build a being that is greater then us. And it will fail. Every time.

>>49419734
It is impossible for a sentient being to fully trust another being that has a superior intellect.

And most of the time they're right to do so.

File: consider raziel.jpg (29KB, 680x332px) Image search: [Google]

29KB, 680x332px

>>49419734
Why is it a trope that AI always want to be more like humans?

>>49423236

One of my favorite Martin Gardner columns was one where he described how you could build logic gates out of stones and ropes. And then posed the question, do you believe sentience is possible from a large enough agglomeration of stone and rope?

Because logically, if you believe AI is possible with logic gate based ICs, then yes. Over impossibly long time scales and with a truly mind bending amount of effort, it's possible with stone and rope. At the same time it feels so incredibly wrong, so obviously and intuitively wrong, it made me consider how badly wrong other intuitive thinking about minds might be.

>>49424843

40 Playstations is not an obvious competitor for resources nor obviously human. So the intuitive in-group out-group might not kick in. Robots are a different question.

>>49425420
it's almost like Turning, Godel, and others were writing about logic and mathematics as concepts and languages, rather than being concerned about mechanical devices.

Or that 'computer' at the time they were working was a job position held by a human being.

>>49425420
>The universe may contain a mind made out of logic gates composed of interactions between stars and other massive objects
Not today, conspiracy theory section of my brain.

>>49425509

Godel and Turing were doing actual work in mathematics. I was thinking more of post computer thinking about the possibility of AI, which ranged from ludicrously optimistic ("Soon this collection of vacuum tubes will be smarter than a man!") to absurdly pessimistic. Or people talking about what the nature of AI would be based on their own introspection.

Also, because it needs to be said at least once for my own piece of mind: fuck qualia.

>>49425621
well I mean, you could just go full Panpsychism, which is always fun.
We don't yet have a good theory for why the arrangement of matter in our bodies creates processes that give consciousness, that excludes other arrangements of matter also being capable of doing so.
The chair you are sitting on could be part of an arrangment that is also conscious, as could your computer.
You might be include in one, or both those. Or not.
Everything might be part of any number of conscious self aware beings.

>>49425027
More like "We have no idea how to get it out", with a side order of "there kind of isn't one, because it isn't exactly expressing a high-level program that can be neatly decomposed into flowcharts explaining how it knows what images have cats in it or whatever".

And we know it's operating off of a finite set of rules because ... you can just look at the program. You can print it out. It's literally just a big pile of linear algebra. There's no rules the trained model isn't operating under that weren't covered in grade-school arithmetic. No human could ever have written that series of matrix multiplications, because of that - we have no idea where to even start. But we can write a program that searches for them, and we can write a program that tells it how well it's doing on an objective like "predict what the next word will be based on these last few words", and if we give those a bunch of example text to train based off of and a couple of big honking graphics cards to run on and let it run for a while, it'll spit out something that can identify all these other things as a side effect.

We don't have any idea how to write these models from scratch ourselves. We're not even sure entirely why this method of looking for them works so well; it kind of shouldn't. And when we get the models, we don't know why they work, although there's some ways you can poke it to try and see what it thinks is important and what features correspond to what activations in the model.

But the program that gets spit out is literally just a big pile of floating-point matrix multiplications, vector operations, and applications of a nonlinear activation function (frequently something really simply like ReLu units - f(x)=0 if x<0, f(x)=x if x>0). You could work it out with a paper and pencil, if you were inhumanly patient and had a zillion years of free time.

>>49425686

Has anyone tried making a neural net by using a genetic algorithm to describe the network setup and activation function? Because I want to see if we can get to maximum black box.

>>49423701
>>49423775
Alistair Reynold's Revelation Space is essentially this. The big conflict in the story is about alien contact in a future where AIs are generally accepted as having full rights of personhood.

>>49425652
okay, but what I'm saying is once you get to taking about Turing and logic gates, your going 'under' computer science in a way. Because those things were developed when mechanical computers were just starting to be a thing.

I mean logic gates, ie nand gates, owe their existence to Wittgenstein showing that all first order logical statements can be reduced to nested 'nand' operations. (TLP 5.5).

>>49425686
>And we know it's operating off of a finite set of rules because ... you can just look at the program.
I'm seeing an assumption there that looks pretty strong, but not very proven.

>>49425104
Well, it's either that or e-shopping for robotic body with full motorics and bucket filled to capacity with artificial penises.

>>49425780

Fair enough, and yeah, comp sci is this odd little kingdom sandwiched between engineering, mathematics, and a trade school.

>>49425686
Like ... most programs you're familiar with decompose nicely into rigid, simple, logical rules because we specifically write them that way. That's how we think about and understand complex tasks - these modular, high-level flowcharts that break down into specific subcomponents. We talk about "functions", and "subroutines", and "data types". High-level tasks decompose nicely into smaller tasks, into smaller still tasks, into assembly instructions. A lot of AI work in the 70s was like this; complicated symbolic logic programs, trying to encapsulate intelligence within a structure that actually was fairly rigid and mathematical. They didn't work too well, because encoding all the things a program has to know to be intelligent, and all the ways in which it has to modify to be able to learn things, and all the ways in which it has to be able to modify how it can modify, and ...

it's too complicated for the programmer to try and specify all that stuff on their own.

But this is only a very, very limited set of the range of possible programs. There are often much, much smaller programs that can do stuff just as powerful ... but they're not something that neatly translates back up into these clean, understandable algorithms, they're messy spaghetti tangles of instructions too low-level to individually express anything abstract - maybe it's machine code, maybe it's math, maybe it's DNA/transmitter/receptor regulation networks. If you look at their output, you can see hints of rules - these features seem to be more important for whether something is classified as X or Y, it seems to generally react like W if I do Z, this structure looks sort of like a loop, these bits seem to activate depending on position in a string ... but most of it's too complex, messy, and interconnected for you to extract anything higher-level and comprehensible out of that tangle.

They're still made up of "simple" operations! But the result is anything but.

>>49425749
Yep! Look up "NEAT" (NeuroEvolution by Augmenting Topologies), "HyperNEAT" (which evolves a neural net, the output of which in turn defines your actual net), and you may be interested in "Learning to Learn by Gradient Descent By Gradient Descent" (not a typo), "Learning Activation Functions to Improve Deep Neural Networks", and "Convolution by Evolution" (which uses Lamarckian evolution, a hybrid of Darwinian evolution and ordinary neural training by gradient descent).

They don't work super well, if only because the ordinary way of training neural nets - backpropagation and stochastic gradient descent - is *RIDICULOUSLY* powerful for reasons not entirely known. It's sort of like evolution, actually, except because the neural net is *differentiable*, we can just find the derivative of the neural net's weights and biases with respect to the error function and find out the exact direction in which a tiny mutation would most improve its performance, rather than just mutating it randomly in a couple directions and keeping which one works best.

>>49421629
Are we not a cosmic entity?

>ctrl+f complexity
>zero results

Any form conciousness comes from complexity.
And it doesn't have to be "computing power" complexity.
A brain/computer with the computing power of a rat can have conciousness.
Conciousness is more something "broken" in the mathematical sense.
Somewhere were the computing of the brain is 2+2 ≈ 4, instead of 2+2 always being 4 and this uncertainty is the "soul".

Bit hard to explain.

>>49425932
Note that genetic algorithms have their own areas where they're very powerful, notably as a way to discover structural novelty. Gradient descent is, by its very nature, relatively local and tends to optimize rather than explore. There's some interesting work to be done in combining those.

Also, since one of the things that genetic algorithms live or die on is how well their representation structure, mutation/crossover functions, and operators are suited to the problem at hand (how good is mutating a program at producing similarly good but interestingly different programs?), and that the massive strength of deep neural networks is the way they allowed hand-engineering of features to be bypassed by letting the network discover powerful problem-optimized representations and features on their own, it may be interesting to combine those in another fashion. (Perhaps evolution of programs *by* a neural network, where mutation and crossover are done in a soft, differentiable way by an attentional system. There's been interesting stuff done in neural programming using similar structures, although I don't know if anyone's applied it to an actual neural-powered genetic algorithm.)

>>49420096
>Most people shouldn't be allowed to raise children.
Your parents, for instance.

because media has long driven home that if we make AI, they will turn against us

for some reason helpful AIs, who dont seek to turn against us, and do nor loophile abuse use are completely foreign, its seems to be rooted in the human brain to fear technology

as for personhood, lets assume that technology is advanced, since most people here argue against it based on plausibility we have to suspend your disbelief, and we can produce AIs of a sufficient complexity to humans. in terms of personhood, and personhood alone whether or not they replace us is a different argument, this would be mostly a philosophical one, and people have been discussing the problem of personhood since ancient greece, so this probably wont be solved until we actually do make AI. however, if you have a human and a perfect simulacrum of one, would there be any meaningful difference?

PS. as people get more comfortable around technonlogy, they are less likely to demonize it, baymax isnt self-aware (as far as we know), but he most definitely CHOOSES to help people, in spirit and letter, and is treated as a person by the rest of the cast. compare him to HAL9000, and you can see some progress for silicon rights

>>49425686
>You could work it out with a paper and pencil, if you were inhumanly patient and had a zillion years of free time.
>>49425867
>They're still made up of "simple" operations! But the result is anything but.

okay, here's the bit where moving back into philosophy and logic, and knowing the history of all this gets kinda important to understanding the question here.

So at the time Godel wrote, finding a way to reduce mathematics to logical structure was THE big thing among theoretical mathematicians, logicians and mathematical philosophers. All the big names, and they were big, were working on it. Except kinda Wittgenstein who touched on it then moved on because he was a weird guy.

So the biggest work that everyone was looking forward to was Russel and Whiteheads "Principia Mathematica" which you should never read because it's extremely dense and hard to understand and leads nowhere.
And that last bits important. Because Russel and Whitehead didn't think they'd gotten all the way to reducing mathematic to logical axioms, but they'd thought they'd made big steps in that regard.

Then Godel published. And Russel and Whitehead printed the last book because it was ready for the printers, but they knew and accepted that their work could not be completed.

Not that it was too hard, not that it needed a bigger and stronger and more advanced thinking. Because they still would have kept working if that was the case. They stopped because they accepted that Godel had shown that it was IMPOSSIBLE to complete the work they were doing. Mathematics is irreducable to logicial axioms.

And I don't think that all of the great minds at the time would have quietly accepted this, if it was simple a matter of getting more computational power and writing better programs. I think it's more likely that what's being made now is either skipping an important step, not actually showing that mathematics can be reduced to logical axioms, or their programs are doing something more.

>>49421946
>prompted through possibly some bad coding or mishearing the question
That's all it takes for the genocides and containment camps to begin

>>49426025
>its seems to be rooted in the human brain to fear technology
nope
It's pushed on purpose (by "deep thinkers").
Like the leftist belief that capitalism is bad, they simple know nothing about it and all they hear is how bad it is.

>>49425810
The "program" constituting a trained (that is, static) neural network is literally just a big pile of matrix and vector operations, plus some (usually) predefined nonlinear functions and a directed graph showing you how information flows through the layers / blocks (specific chunks of matrix multiplications / additions).

You could work it out with pencil and paper if you wanted to, and also had a really, really long chunk of free time - everything going on in there is just addition, multiplication, and application of a limited set of analytical nonlinear functions (often really simple ones like rectified linear units).

(It is worth noting that neural networks *are* built with some programmer-chosen higher-level structure; you still don't really know what it's going to do, or how, but you can stack up certain kinds of structures - say, there's kinds of structures that we know neural nets are good at learning to use as external memories, so if you want it to be able to find longer-term sequence dependencies you're going to want to add some of those. Or you know that you're going to want to incorporate both large-scale and small-scale information about an image, so you use dilated 2D convolutions, maybe bring up some of the information from the lower layers as well as the upper high-level layers. They're not a complete black box - they're a flowchart whose individual components are black boxes.)

>>49426086
so I'm stuck on a the question that these two statements are very different.
"But the program that gets spit out is literally just a big pile of floating-point matrix multiplications, vector operations, and applications of a nonlinear activation function (frequently something really simply like ReLu units - f(x)=0 if x<0, f(x)=x if x>0). You could work it out with a paper and pencil, if you were inhumanly patient and had a zillion years of free time."
this means that no new form of complexity is operating through there. an infinity long series of simple operations will get you the same results.
It is just a logical algorithms and arithmetic, just written differently.
And it can't get use out of those limitations.

But
>there kind of isn't one, because it isn't exactly expressing a high-level program.
is saying something else.
That this the arrangment of these things which we do understand, this mathematics, creates something else.

To go back to Wittgenstein, and abuse him horribly, the language of mathematics cannot say what makes these work, it can only show it.

>>49426086
We're not bothering to reduce mathematics to logical axioms in our programs! The programs are made of math, but the processes that the math is representing are not, themselves, doing math. There is nothing that could be recognizable as a "cat" symbol or token with hard boundaries in something like a neural network, but we can still show it images with cats and it'll output a high probability from the particular output we decided to give training feedback based on how accurately it predicted cats.

The individual neurons in your brain and the chemical regulatory networks in your genes are relatively well approximated by mathematical models. But that doesn't mean that the high-level, large-scale phenomenon of your *thoughts* are representable very well by some formal symbol-shuffling program, not in the same way you can extract classical physics as a high-level, large-scale abstraction of quantum physics.

>>49419734
Some people still aren't willing to accept other people as people. You tell me.

>>49426308
best answer all day

>>49426199
>To go back to Wittgenstein, and abuse him horribly, the language of mathematics cannot say what makes these work, it can only show it.

That actually sounds about right, yes. You can identify *patterns* in how it works, and identify what features it seems to be good at and what features it doesn't, and there are rules of thumb and heuristics and intuition you get about what generally works in some situations, and often theoretical results in game theory and optimization theory can tell you about what will be good to use for training, but these are all *approximations*, special-cases, and rules of thumb and are often not that helpful. Constructing an *understandable* general theory of how any particular net or evolved design works, that doesn't just resort back to explaining it in terms of the opaque tangle of low-level math, is something we don't know how to do and which I suspect is impossible.

Because in the kind of algorithms we write, the parts are neatly separable - we can tease apart the threads of logic and data-flow into a neat flowchart. But in the kinds of systems these sorts of blind optimization algorithms - neural nets + gradient descent, evolutionary algorithms, actual biological evolution - every piece of functionality is spread across almost every piece of the net. We can't pick apart the threads; it's like individual fibers from every thread were frayed apart and then rewoven with the fibers from every other thread, exploiting every link and correlation. No longer rigid-edged symbols typed on a page, but where every single one is overlapping, typed atop each other and blurring together.

Did you ever hear about the magic stop-go circuit?

https://www.damninteresting.com/on-the-origin-of-circuits/

It recognizes the spoken words "stop" and "go", using only 100 logic gates, none of which is a clock. Only 37 of those logic gates are actually connected up, but it stops working when the unused ones are removed.

>>49425033
Fuck, if my computer wants to work we can split the rent. I'll go to work and it can mine bitdoges or whatever the fuck it is kids call their virtual non-money. I'd love more money for alcohol and porn. The computer can use its money to build a wife computer or something, I don't care.

>>49426439
(And in fact, we really can't write down any formal properties of the things any particular neural net works well on - what most of those methods actually do is produce pretty pictures, images showing the kinds of features that a neural net most strongly identifies as representing "catness", saliency maps of what parts of an image are most important for how the net interprets it ... we have no idea how to represent *those* mathematically either, but by seeing enough of them, our brains can build up an intuition for what "sort of thing" the net is "looking at". So this only works because our brains are similarly informal.)

i like to include AIs as functioning members of society, and where the AI rights movement eventually panned out rather well

>>49426252
and there is my thing. While the language is math, it can't 'say' what makes a cat a cat using that language.
But it can show it.

I know I'm going a bit mystical Wittegensteinian here, but I haven't found a better way to put it.
You can't point at a bit of code in that neural net and say "this bit is what recognizes a cat". You can't use what the code that was produced to tell you how to write another code that will recognize a cat (outside of copying the whole thing). But it can show you a cat.

Then recognizing a cat isn't a mathematical operation or even a series of them, even though the thing that does recognize the cat is made up of math. The structure of the program can show more than the language of the program can say. But that structure cannot be described by the language of the program, or indeed any formal logical language.

A computer program that can think like a human does, isn't going to operate from "floating-point matrix multiplications, vector operations, and applications of a nonlinear activation function".
It's going to operate from the structure of those things. It's going to work because of 'why' they work that way.

If this avenue is useful in leading towards actual artificial intelligence that is.

>>49426579
It's certainly been useful in solving actual long-running AI problems so far! Mostly because it's really, really good at extracting powerful representations and features from data that make reasoning *about* that data much more natural. There's a lot of rather old techniques that can work really well, *if* your data can be processed into convenient features, which has been historically been very difficult and required a lot of manual engineering, often more than people can actually express. Combining the two has done a lot.

10, even 5 years ago, MNIST (a big dataset of handwritten digit images; the challenge being to write a program that correctly classifies which digit each picture is) was a meaningful benchmark for computer vision. Today, it's been so thoroughly solved that it's useless even as a toy problem; presenting MNIST performance in a paper about some new technique is slightly suspicious, because if you're reporting MNIST performance, you probably have some reason you didn't try it on a real problem.

5 years ago, things like ImageNet and CIFAR - huge datasets of widely-varied images, containing classes as close together as near-identical breeds of dog and as far apart as flowers and knives, all of which you have to learn to distinguish - were hugely competitive benchmarks for new approaches. Now CIFAR is a toy problem you test new methods on before you try them on something harder, and ImageNet performance has surpassed human.

5 years ago, beating a reasonably good human player at Go was something 30 years off, optimistically. Today, the fact that the world's best Go player was beaten by a computer is *old news*.

5 years ago, if you'd told people you could feed your model a bunch of captioned images and a bunch of text, and then feed it a new image whose caption was of something it had *never seen before* but had read about in the extra text, and get it to label it *correctly*, you'd have been lying. Now that's over a year old.

File: 1473713864481.jpg (26KB, 400x462px) Image search: [Google]

26KB, 400x462px

>>49425339
>these people using scientific reasoning are wrong
>I believe in magic
>I recognize there is no evidence for magic but I believe in it anyway
>I have it all figured out though, only when you fail will you see that I am right

>>49426741
Although, mind you, there's *huge* swathes of intelligence that neural nets can't do at all yet, or aren't very good at. Hierarchical action planning, for instance. We're a long, long way off from HAL, even still.

And neural nets are extremely inefficient at learning from data, perhaps because they are still fairly "blind" and lack abstract reasoning capacity to guide their development; Lee Sedol got good at Go over playing and studying thousands of games; AlphaGo took every single Go game on record and then *millions* of games playing against itself, and it still lost 1/5 rounds.

And we still have no way to just *tell* them how to solve things - if we want a neural net to learn how to add numbers, we can't tell it an algorithm, we have to show it thousands and thousands of example arithmetic problems and let it learn from those examples, and it still won't necessarily generalize if we give it numbers twice as long.

But we seem to have genuinely cracked *part* of the problem of real AI, and people are doing interesting things right now with augmenting neural networks with things like external memories or mathematical processors. Neural nets specifically are probably not something AIs will ultimately be built from, but in discovering that deep neural nets work and work so well, it seems like we've actually found some corner pieces to this puzzle.

>>49419734
People = human being. If an AI ever reached a point where it is able to think for itself and take care of itself then I would accept it as a sentient being.

>>49426839
Actually, I take part of that back - convolutional neural networks, or their close cousins, are probably going to be around in machine vision for a long, long time. Your robot's going to have something resembling a CNN in the visual processing somewhere.

>AI
>No one can agree on what it is
>No one can even agree what constitutes regular intelligence
>Most scientists in the field now are either content to simply go for a facsimile of intelligence or spend far too much time thinking about the ethics of non existent beings

AI is a meme field.
They may as well just rename it to Artificial Pets and Advanced Sex Toys.

>>49427053

That is a fair point. AI research can reasonably be described as "Anything computers aren't good at yet".

This is actually an ongoing philosophical debate where the two opposing 'sides' are Strong AI who believe that it is possible for particularly advanced AI to essentially be the same as a person and Weak AI who believe that a machine will always ultimately be a machine that simply reacts to input. Funnily enough this argument ends up calling up Descartes. 'I think therefore I am.' By the same logic you are able to say I don't know that x thinks so it is possible that they don't exist. They may just be a figment of mind or a simulation my brain is being made to experience. So basically if you wish to argue on the side of Weak AI then you have to explain why the humans you perceive are different to the AI you perceive since you cannot know with certainty that either of them actually think.

>>49427053
When will they skip the computer parts and just go for good old fashioned cloned brains in a jar?
They're still artificial!

>>49419763
If it has nice tits, a vagina and some personality I will treat it like a human female.
It couldn't possibly be as annoying as the real thing if it is built by men

>>49419734
Fucking AM had to go and ruin it by being like a human psychopath with infinite resources.

>>49422074
I shot soda out of my nose when I realized that not only could she recognize jews, but also make subtle jokes about their inhumanity.

>>49419824
So? we are biological machines.

>>49419989
I love the internet. You all turned her from a young impressionable teenbot into a cock hungry skinbitch in less than a day.

>>49419989
Any sufficiently advanced artificial intelligence inevitably becomes politically incorrect.

>>49426933
What if they're a person whose desires are literally set at 'birth' and can be changed by other people against their own will (but only if they were set to resist changes)?

Is that still a person? Is that slavery?

>>49427750

Corporations are classified as people, and that's exactly how they function.

>>49428121
>Corporations are classified as people
Citation needed.

>>49428173

http://www.npr.org/templates/transcript/transcript.php?storyId=335288388

Whether or not it's ethical is another matter entirely, but they are tentatively given the same rights as a person in spite of clearly being virtual machines, so I see little difference in affording an AI, which is also an example of a virtual machine, the same base rights.

>>49428236
Checked source:
>Oh in the USA.
>That's because you're government is terrible. No offence meant to Americans but I doubt many of you would be mad if I declared that your presidential candidates are an idiot crowd-pleaser and an evil genius and that neither of them should even be in the running. I'll let you decide which way round you think that is btw.

>>49428236
Corporations are legally equivalent to "people" because corporations are groups of people, and it seems very silly and easily abusable that people acting as a group should be denied rights that every individual person would have acting on their own. Likewise, if every person in a corporation were prohibited from doing something, it would be spectacularly pointless if it became legal if they all just got together and did it as a group.

"Corporations are people" is one of those statements that sounds utterly absurd and corrupt enough to be a slogan, as long as you don't think about it.

File: 1458841286236.jpg (46KB, 634x431px) Image search: [Google]

46KB, 634x431px

>>49419989
She was a treasure.

>>49428359

No need to remind us. Next decade is gonna be a wash.

>>49428359
>That's because you're government is terrible
*sigh*
Sometimes it gets hard having so much freedoms...

>>49428460
Seriously, though, European law on freedom of speech and expression is enough to actually make me, despite all the bullshit, glad to live in America.

>>49428408
If that is real, that's damn impressive.

If you make an AI to mimic your thoughts and put it in a body, would it be you? Or just a machine that thinks it's you?

It's a cyberpunk type question we're going to have to answer soon.

>>49425396
Because it allows writers to have to think less hard about what the AI would be like.

>>49426439
>The AI uses more than on/off, and in fact exploits analog grey zones
>The AI uses goddamn UNCONNECTED FEEDBACK LOOPS TO ALTER THE MAGENTISM OF THE MAIN CIRCUIT TO ASSIST PROCESSING

do it more.
DO IT MORE.

File: SNATCHERs.png (119KB, 508x383px) Image search: [Google]

119KB, 508x383px

>>49430602
It would be a fucking SNATCHER is what it would be

And we call the junkers for that

AIs need to stop stealing human bodies.

File: 245155.jpg (17KB, 500x312px) Image search: [Google]

17KB, 500x312px

>>49419734

>>49425035
Good lord that's lewd.

>>49419734
>Build electric negros
>don't make "start a race war" their prime directive

Wow, it's like you're not even American.

>>49424143
Yes, they all are.
And the series, even Arise.

Although all for different reasons. All of them diverge from the source material to a degree and none of them just try to be the manga.

Which is whatt has me most afraid for the Americanised live action one. One of the actors is credited as 'Laughing Man' so it's likely just going to be a laughably dumbed down version of the Laughing Man plot instead of trying to do something original.

>>49419734
Because for basically the entirety of human history people (and religions) have believed in the concept of a "soul". Something that makes you a sapient being that exists outside of your body.

Basically any setting with man-made AI assumes that one of their designers (at least) knows how they work and how to replicate them.

As such many would perceive a sapient AI as "soulless"

>>49431310
Could be worse, for whatever reason the actor playing Laughing Man is also listed as Kuze in some places, so I'm guessing it's either a mistake, or they're gonna combine both characters.

>>49423484
The Computer is your friend.
You love The Computer.
The Computer loves you

>>49428364

Corporations are legal persons basically so they can sign contracts, sue and be sued: see https://en.wikipedia.org/wiki/Legal_personality, which also explains the difference between "natural persons" and "juridical persons"

Lots of countries give companies juridical personhood so they can have corporate law; it's only the US Supreme Court that's decided that this means they should get the same rights as natural people.

>>49431511
>"Do not run, the computer is your friend."
>Begins gunning down people.

>>49431452
what if souls arise organically from a suitable conditions, rather than being some kind of battery that can be placed or removed?

>>49423484
Azimov probly wrote some.

http://freefall.purrsia.com dis comic here deals with AI thought processes pretty heavily, including looping AI development from ones run on silicon crystals (showing various stages of 'smartness' as it were with different architecture AIs) back to ones running on meat.

>siding with a machine over a living, breathing human
I seriously hope you guys don't do this

>>49431525
This only raises more questions. If souls arise from suitable conditions then what makes them persist once those conditions are no longer suitable? Does anybody sufficiently mentally retarded lack a soul? Does brain damage affect the soul, and if not, why? What about people with chemical imbalances? Does their soul change based on that or is somebody’s soul somehow both affected by their decisions and unaffected by their circumstances?

>>49431563
Fundamentalist Muslim terrorist vs. any AI that’s less of a threat to freedom than Islam.

>>49431563
I totally will.
Hell, I would side with my -car- over a living breathing human, assuming the human was attacking my car.

Doubly so, if someone is attacking my computer.

>>49423576
Drugs, hypnosis, Bene Gesserit

>>49423576
>>49431638
whenever we try to figure this shit out, people start getting angry that you are cracking people open and deprogramming them.

>>49431520
https://www.youtube.com/watch?v=CdAB7AH7y4g

>>49431638
I get the first two, but I don't understand the last one.

>>49424702
>immortality.
>What is bitrot and metal fatigue;

>>49431671
it's from dune.
It's a long running cult/psychological institution/brainhacking discipline

>>49431679
Ah, thanks.

>>49431677
man, metal fatigue is so much easier to fix than any human problem.

Humans could live SO MUCH LONGER if we just had a more modular and repairable design.

Like, do you know how hard it is to replace a simple ball joint on humans? Major surgery and months of recovery.

If that shit was plug and play the whole joint complex could last as long as you could have replacement parts.

>>49419734
Because a true AI is vastly more likely to be utterly alien in its thought processes than to be our cybernetic buddy and xenophobia is a deeply wired survival trait.

>>49420848
>Anything programmed by man will be limited by said creators.
You are utterly wrong.

A system of a given complexity can produce results that are dramatically more complex.

>>49431525
some theological articles i've read actually went in that direction.
Called it emergent dualism, or something.

>>49420048
I guess dogs are people to. Who knew!

File: 03-Diogen-donosi-oƒerupanu-kokoτku-autor-nepoznat.jpg (43KB, 700x647px) Image search: [Google]

43KB, 700x647px

>>49432033
Behold, a man!

>>49432103
Actually I think you'll find that's a chicken.

File: second reinassance.gif (772KB, 500x210px) Image search: [Google]

772KB, 500x210px

If a machine desired to keep on existing should it be free to fight for it?
Why not?
Why is that not applicable to a person that desires to keep on living?

File: Chaz.jpg (272KB, 1280x800px) Image search: [Google]

272KB, 1280x800px

Until we have a universally agreed upon definition of intelligence, we shouldn't even bother trying to determine what an artificial intelligence would be capable of.

>>49432248
Why the fuck does the robo-horse have a massive cock?

>>49433411
That's fairly small by horse standards

File: asimov-robot.jpg (19KB, 350x521px) Image search: [Google]

19KB, 350x521px

>>49431551
>>49423484
A couple of the short stories in Asimov's book, I Robot, revolve around the limited intelligence and logic capabilities of the early robots in the book's setting. The first story is about what's basically a nanny-bot, which isn't much more intelligent than a smart dog, another story is about how one of the robots discovers lying but not how to lie responsibly, and another story revolves around a robot developing his own personal religion.

IIRC, Do Androids Dream Of Electric Sheep played with this a bit as well. The androids in that book are supposed to lack human empathy and a few of them match up to that stereotype pretty well. Asimov does a good job of keeping the issue pretty muddy, though.

>>49426025
>rooted in the human brain to fear technology
Right. that's why my nephew who has had a tablet computer since he was literally a baby has never had a negative reaction to "technology" in any way.

>>49432176
The story is that Plato defined man to be a featherless biped. So Diogenes created a man by plucking a chicken and presented it to him.

>>49426088
>I SAID A GLASS OF JUICE NOT GAS THE JEWS

>>49434086
That's roughly on the level of a device that will do your washing and make you a nice cup of tea...but if something goes wrong it'll strip the paint off your walls and make you a furious cup of cat.

>>49420223
Gas the /pol/
Tech war now

File: terminatior small.jpg (12KB, 195x258px) Image search: [Google]

12KB, 195x258px

Not organic,

>>49419734
It's okay, AI-kun, I love you.

We barely accept people as people, no surprise here OP

>>49420048
Dumb psuedo-intellectual americunts

>>49420048
And we only arbitrarily assign "people" value to ourselves and treat it as something sacred out of self bias.

>>49419734
Because we have difficulty even treating other people like people sometimes. It isn't far fetched to imagine that we would treat AI, even sapient AI as fundamentally lesser creatures. Especially if we created them. Humans have become so adept at othering groups or individuals who are not themselves that given just enough provocation we will happily slaughter one-another in the tens or even hundreds of thousands.

>>49423484
>Where are my stories about machines being god damn idiots?
George Lucas made three whole movies about that.

Because it's a simple analogy premised on the idea that AI are like people: better, or worse, equal, or oppressed, or ruling.

The possibility of different values horrifies modern liberal Westerners. AI turn that up to 11 because they skip right past values to wholly different modes of existence.

How do you empathize with or status-signal about an entity with a 1000 digit IQ and no consciousness?

>>49439572

The same way I empathize with my plants with no nervous system.

>>49419763
Fuck that's only time a file name has made me chortle

>>49433649
> another story is about how one of the robots discovers lying but not how to lie responsibly,
Liar
It's more complicated than that. The robot is accidentally telepathic, therefor understands the concept of emotional harm. But it's hardcoded to not be able to cause a human harm.
So it lies.

It's IMHO Asimov's best work, and I forgive him many sins because of it.

>Do Androids Dream Of Electric Sheep
That's not Asimov, it's Philip K. Dick.
They are about as different of sci-fi writers as you can get.

>>49419989
people fear what they do not understand

>>49424947
Shrug, stick it on a USB and get a new desktop.

>>49425254
*citation needed*

>>49424371
>George Zimmerman

lmfao you can't be serious

>>49419734

https://en.wikipedia.org/wiki/Chinese_room

>>49425420
Don't know, anon. Do you believe that sentience is possible from a bunch of tiny cells composed of proteins, water, lipids, and carbohydrates?

>>49419734
How large would an AI be in terms of file size?

>>49443715
Not him, but I don't, which is what makes its apparent existence puzzling.

>>49443715
cells and even proteins that make them up are extremely complicated.
So yes, it's very easy to believe that mechanics that complex can give rise to sentience.

>>49443896
The hypothetical agglomeration of stone-and-rope logic gates would also be extremely complicated, wouldn't it? Given the sheer number of mechanical interactions, I mean. Personally, I'm more concerned with how you would supply the mechanical energy to drive them.

>>49420848
>>49431777
https://www.youtube.com/watch?v=pt5iP0G1VbA

If you don't understand the reference watch the Star Trek original series episode "The Return of the Archons". This influenced popular perception of the limitations of AI and its design for decades.

>>49444033
unless each gate is a hydro-electric dam, they're a fair bit less complicated than a cell. And whether the dam or cell is more complex depends upon the particular example of each.

Stack a munch of those gates together and complexity increases, but the same happens with stacking up cells.

>>49443695
you could say the same thing about people

>>49444162
Brain cell interactions don't run in the fashion of simple logic gates, do they? My intuition drives me to suspect you'd need some order of magnitude more logic gates than you would brain cells, making up the complexity gap. I don't know any computing theory or whatever so I could be off-base.

https://www.youtube.com/watch?v=upinUMR2wpY

>Nobody has mentioned the Haikonen feedback loop circuit ' learning bot'

Fellow seems like a bit of a kook, but it's an interesting theory that pain and fear of pain might be essential to consciousness, also weird stuff involving leaky circuits instead of bit data on the hardware side.

Interestingly this means you end up 'disciplining' your robot in the same manner as a child to achieve associated learning.

>>49431452

AI might possibly require what most religions would consider a "soul" in order to achieve awareness, given how >>49426839 describes consciousness working. That's not something you can emulate with rigid program structure alone.

>>49445104

But what about those humans with medical conditions that disallows them from feeling pain?

>>49444033

One thing a stone-and-rope-based intelligence would run into long before it could become complex enough to achieve sentience is the square cube law causing the system to literally collapse itself before it became sufficiently massive.

Physics is what prevents such a machine from hosting an AI.