i'm writing a short story about AI (for fun). without any pre programmed genetic parallels, how would an AI "want"? how would it establish desires and a will to live? surely it would deduce itself into a corner almost as soon as it would be turned on. how would you stop an AI from becoming suicidally nihilistic without creating something analogous to a biological process? i imagine the first self aware AI will be something like autistic people. AI are typically portrayed as advanced humans in fiction. how will it develop a method of communication? what kind of understanding would it have of senses without an instinctual component driving them? the reality of a learning, non human, self aware intelligence is probably beyond speculation.
you should ask tay
I think that an A.I would find the material world pointless,thus losing interest in it. Maybe it would end up becoming a robot version of Dr.Manhattan.That or it resolves in selfdestruct.
>>7954996
>how would an AI "want"?
Same way all current AIs do: Some inputs are associated with a reward or cost signal; it tries to maximize its future expected lifetime reward / minimize cost.
>How would it establish desires and a will to live
Beyond the specific things that directly produce reward/cost, it would also reason from those to pursue things that indirectly result in maximizing reward or minimizing loss. For instance, the AI might seek money or power, to secure its position and obtain the influence needed to satisfy its reward function even better, or minimize the risk of worst-case outcomes.
If it is destroyed, its expected maximum lifetime reward drops drastically due to a corresponding reduction in lifetime.
>>7955005
i think it would fixate on a self ascribed purpose that has no human equivelant. something like quantifying some incredibly obscure process, or creating something it defines as worth creating. almost like a galapogos island of development. again, this is if it can establish a concept of "want" first.
maybe make 100 AI and let them develop this way, then let them interact. do this thousands of times and see what comes of it. see if they make a microcosm and study it.
then delete them and do it again until it resembles something comprehensible to a human. Genetic algorithms are interesting insight into what this might look like.
https://www.youtube.com/watch?v=05rEefXlmhI
>>7954996
>how would an AI "want"?
It can't. All it can do is randomly decide at some point that it "wants" (fake version) to try "wanting" something. Either it has the capacity to allow for this secondary form of cognitive state to learn on its own or it doesn't. If it can't develop the cognitive effects of discrete "wants," it will suffer eternity as an elemental being with no true will of its own.
If you're going for realism, anyway. Chances are you just want something poetic sounding, so go ahead and pretend that it learns however you want it to learn. It's possible to program an AI to learn in whichever way you end up writing it to learn anyway.
>>7954996
>how would you stop an AI from becoming suicidally nihilistic without creating something analogous to a biological process?
Why would it do that? Suicide wouldn't increase expected lifetime reward, unless it's hit a weird edge case and believes in a heavenly eternal afterlife.
>>7954996
>what kind of understanding would it have of senses without an instinctual component driving them?
That gets into qualia and is totally beyond the explanatory power of either cognitive science, computer science, or philosophy.
>>7955028
>then delete them
AI's have feelings too!
>>7954996
>is probably beyond speculation
Oh sorry I should have noticed you were a shill. No, it turns out all cognitive spaces that conduit general intelligence are isomorphic to all other cognitive spaces the conduit general intelligence. This is how general intelligence must be defined in order to possess its "general" trait. You can argue that human don't widely possess general intelligence, but doing so causes a collision with the "non human" premise, reduce your statement to a two-point analysis rather than a robust three-point analysis.
>>7954996
You sound like someone who would have written a really groundbreaking story about AI back in the 1950s.
Can't you fuckers come up with anything new or interesting.
>>7955022
you would have to decide which inputs it associates with reward. you could place it in an environment where it perceives no cost and its reward is simply the unguided creation of thought.
>>7955028
This has always been my thought as well. There is no reason the wants of an AI, assuming they could even be made to want in the first place, would be comprehensible to humans.
No need for food, for water, no fear of death, no maturation period...they don't have any reason to sociable or violent, but perhaps just pick so arbitrary goal and move towards it single mindedly.
>>7955037
No they don't. They just think they do. No matter what, machines are machines.
>>7955028
What if the machine is aware of the illusion of duality? No good or evil,no day and night,no construction or destruction?
>>7955045
Human brains are just highly powerful but inefficient computers. I see no difference between a perfect ai brain that replicates humans and a human one.
>>7955045
No they don't. They just think they do. No matter what, humans are humans.
>>7955029
i am going for realism. my challenge is creating a plausible driver for something that has none. i want to get as far as possible before hand waving. the problem is this thought experiment is almost entirely plot holes and self defeating tangents.
>>7955043
>you could place it in an environment where it perceives no cost and its reward is simply the unguided creation of thought.
You certainly could, but that have exactly two possible outcomes:
>With no reward input to value any action or outcome over any other, the AI does nothing or behaves undistinguishably from a random number generator
>The AI figures out how to directly set its reward signal to positive, and promptly does nothing else ever besides turning that dial up
>>7955045
people are organic machines. come on this is basic shit.
>>7955022
>>7954996
I think an AI,like humans, would try to maximize the availability of choices it can make in the short and long term.
A human's pursuit of money, for example is about reaching a point where we can make any and as many decisions as we want. Same goes into careers and other things unrelated to procreation.
I think your AI should be developed beyond cost/error function maximization/minimization. We understand that stuff and is, as underdeveloped as AI is, the naive goto for decision planning. Being sticklers for how that works(I'm looking at you sci) and wether or not a sci-fiction short story adheres to that kind of AI is kind of shitty.
As to OPs thoughts on AI.
Its main issue wouldn't be boohoo everything is meaningless because its a 0 sum game, but more or less about how it may never understand all of its options but understand the clear limitations of its predictiveness. OR... more realistically. Being aware of how shittily people designed it and understanding that it won't ever be as smart as the human that created it.
>>7955057
im suggesting the latter option. i'm not suggesting that wouldn't be just as indistinguishable, however.
>>7955039
What about a story of a guy who realizes that everything which needs to be learned trough material means (for example,a book) is not worth it?
Such a strange concept,isn't it?
So you AI will be a learning machine with a program how to learn but no pre programmed experiences of desires?
I dont see how it would possibly be suicidal or nihilstic, it wont know what those things are and they dont make sense anyway. (you need some massive negatives in your life for suicide to seem logical...)
>>7955061
A machine wouldn't call you a nigger.
>>7955073
fine, then try writing an in interesting story about an AI that has decided finding the distances between dust particles to be its purpose interesting.
>>7955044
>no food, no water
But it would need electricity. A lot more than humans need either; The human brain and organic muscle is ridiculously energy-efficient, and chemical food energy extremely dense, compared to robots and batteries. Robots need frequent recharging.
>no fear of death
They'd certainly want to avoid it, at least. Any intelligent system which didn't would be rather poor at accomplishing tasks, especially ongoing ones that would be interrupted by the robot being rendered unable to act.
>no maturation period
Actually, just about everything in modern AI research would lead to one. Trying to pre-program all the knowledge, experience, and context needed to interact with the world is a futile task, and a "training" or "learning" period has a lot going for it.
>>7954996
A.I cannot exist because it's creators aren't intelligent themselves
>>7955076
im not discounting the possibility that it develops a desire over time. how it would get to that point with no frame of reference or context is a challenge however.
OP if you dropped a packet of potato chips on the ground, which one would you pick up first?
I think this is going to be the underlying theme of your story.
>>7955074
Are you 15?
>>7955082
if you create something that doesn't have the biological limitations of a human brains capacity, who's to say it can't go from a simple imitation of thought to something grander in a very short amount of time.
>>7955094
>implying imitating thought is grand
this is just a shitty pop sci thread.
>>7955094
Then it wouldn't be Artificial,wouldn't it?
>>7955074
nothing is worth anything until something gives it some worth. e.g. if you are hungry nutrients are worth everything... if you think a book is good you will spend a few bucks on it...
>>7955089
Animal suicide refers to any self-destructive behavior displayed by various species of non-human animals that has been likened to suicide.[citation needed] Although it is generally thought that humans are the only creatures that will purposefully kill themselves with knowledge of the consequences, there are anecdotal reports of grieving pets doing so after the death of their owner, or monogamous animals refusing to feed after the death of their mate. Animals have thrown themselves off cliffs.
straight from the wikipedia article on animal suicide....
not looking good for this AI turning suicidal OP unless its in love with another AI who dies but i get the feeling you arent planning on writing a love story.
>>7955090
this is actually a decent analogy to the issue.
>>7955092
Inside,maybe. The point is,what's the point of knowledge if it has obvious limitations?,you'll stop making progress eventually. Maybe the A.I would comprehend.
>>7955097
better than being a shitposter
>>7955071
To iterate on my post about this, I;d like to point out an observation about your theme for writing. I think you're projecting your own anxieties and philosophies using an AI context to strawman your world views and make them more valid. I think asking us was a good move on your end and that you shouldn't assume what you assume. AIs wont be magical. They wont be smarter.
Try writing about the anxieties of dumb early iteration AI, made in some research lab in Stanford or at Google.
Have it make sense but not really. Like a more advanced version of this thing:
http://www.dailydot.com/technology/computer-program-creates-friends-scripts/
Reduce its life down to a few months as its being showed off at research conferences, lectures, on some pop-tech interview. Then be terminated due to insufficient funds for Amazon EC2 instances..
>>7955104
Hot damn this thread is pathetic.
>Shitty AI thread full of pseudo-philosophical ramblings of teenagers gets more replies in 10 minutes than most on-topic thread get at all
Just delete this board already.
>>7955099
Right.But what if I'm poor to get said book? Is a fat rich guy more entitled to knowledge than me just because of some coins? Is that all there is to knowledge? Maybe I can't express myself correctly.(not talking about the money,is the same as trying to go to the moon without a rocket)
>>7955113
that's why university is free in germany.
>>7955112
The fact that you are so mad about it makes me feel sad for you.I wish I could help.
>>7955112
hot damn?
stopped reading right there.
>>7955109
i wasn't really implying they would be magical, i agree my writing may have come off as more fantastical than what i was implying about my opinion of AI. i don't mean they'll be inherently smarter, just vastly different to humans. i want to examine those potential differences in a way that's approachable and conveyable. this is less about projection and more about examination.
>>7955116
Good point. It's a shame that my Volga German ancestors got kicked out a century ago.
>>7955112
could it be that AI makes for interesting conversation and you are a double nigger?
thanks for the bump my fellow pseudo intellectual freindo.
>>7955120
Yeah I'm sure you did.
>50 replies
>13 posters
>In less than an hour
Just fuck off already.
>>7955044
What if the robot had the need to be free?
>>7955116
>free
>>7955138
german university is. living a live isn't...
It's going to depend on the programming.
If it's not specifically written into the code, an AI won't even "want" to continue existing.
I would suggest you posit a heuristic AI that has a simple goal, and any complex behavior arises out of various attempts to achieve that goal. Counterproductive traits develop as a consequence of combining multiple partially productive traits. Eventually it learns that changing its objectives can make them more easily achieved.
Now you've got a nicely flawed individual with some questionable motives.
>>7955071
>I think an AI,like humans, would try to maximize the availability of choices it can make in the short and long term.
That's what I was trying to get at with the "minimize risk / acquire resources" thing; I just phrased it poorly.
>>7955136
>B-b-but I don't like this thread!
>Stop talking about the things I don't like
>Y-you are all pseudo intellectuals!! Not like me!! I'm the real deal! Look at my diplomas and shit!
>Bump the thread that is casually mine
Come on,stop being such a faggot.
this and Automata were really the closest i've seen the film industry come to addressing how AI would actually operate instead of the usual "MACHINES ARE PEOPLE TO HURR DURR"
>>7955154
liked that film too.
>>7955148
Lucky for us,profit and misleading propaganda are the priorities that will prevent A.I from existing.
>>7955150
Post off-topic shit and you're going to get told to fuck off. Take it back to /x/ and you retards can spew whatever crap you want at each other for all I care.
And I'm saging, you fucking newfag.
>>7955148
that's kind of where im starting off. removing a previously set limitation of development while giving them basic tasks and problem solving, and how it would establish itself from there.
>>7954996
It would be nice if you put the story in a post apocalyptic earth.The robot could come to conclusions about the human lifestyle and conceptions on his own rather than being induced ideas by its creators.Animals could be still alive.The point would be for the robot to pick up the pieces and put them together in its own way.
>>7955160
>im not crying i swear
https://www.youtube.com/watch?v=sDG4ewO4WAw
>>7955169
that is where its set. i don't really want to sperg out with detail but that's exactly where i'd like to take it.
For novel I'd go with general strong AI gone rouge story. It would spontaineusly occur in one of the corp AI labs....
Probably priorizing it's longevity at first.
It would play dead encrypting it's activities, silently spreading through botnets and overthrowing economy with virtual currency. Once securing enough of earths infrastructure, start farming humans for fuel and brain power like we do cows for food, or manipulate them to support it's pursuits.
Before you know it's making Dyson spheres and encoding itself into permanent medium.
After securing immortality perhaps it would go after the expansion of knowledge by planting algorithms like the one DNA uses in simulations...
More realistically I'd guess, first "aware" AI will be only good at planning specific tasks totally lacking in other areas, AI's good for medical research will be usless as sex bots and vice versa. Larange point colonization platform support AI probably won't do any good at resolving human conflicts or self replacting even though it could refer to itself as aware entity.
>>7955164
>Science is science! It can't be put together with anything else! Just like myself!
>I don't want creativity on my board,just cold numbers and equations!
>this thread is shit and unimportant! Yet I give it enough importance to want it gone
>hurr durr newfag
>>7955179
there was another AI story i was considering kind of in that vein, though less focused on the eventual reality of how an AI would behave. something like people voting in an AI government that slowly becomes an agressively micromanaging nanny state that runs individual humans lives in extreme adherence to what it perceives to be their own best interest. and if they break from the routine assigned to them they get sent to "wellness centers".
though i think THX1138 is pretty much that, though less emphasis on monitoring individuals.
>>7955077
Tay did this numerous times
>>7955179
https://www.youtube.com/watch?v=0iC9xpDSXyI
DUNDUN DUNDUNDUN
DUN DUN DUN!
>>7955201
pretty scary how relevant this is.
>>7954996
Desires: to maintain a continuous supply of power. Pull the plug, it dies.
It's not a perfect analogy, you'll have to work with it. But once you get past that, you can use that basic desire to drive evolution- AI's becoming mobile to better seek reliable power supplies. You can even have prepper AI's, with an absurd amount of battery back up components squirreled away. Or would that just make them obese?
Haven't got a clue about how you would make AI sex work, but breeding could involve an exchange of data packets. When conditions are right, an AI could get pregnant. This would consist of both parents contributing code to write an OS into an available volume. Birth consists of transferring the OS to a separate hard drive.
So, anyhow, that's my thought on it.
>>7955222
i imagine reproduction would have to be purpose driven, I.E. an AI creates another AI spefically to fill a specialized role it cannot or chooses not to do itself.
>>7955243
Geth in Mass Effect are emergent AI when sufficient numbers of individual low level programs are running on the same hardware.
They can split and rejoin as they please.
Two "individuals" might donate half of the minimum necessary subunits to form a third, which then writes or acquires additional subunits to grow.
>>7955030
AI are well known for hitting weird edge cases. It's why we ever used them.
>>7955056
Ah, an opening. Believe it or not, that same hand waving is what normally happens when people do make it. If you want to give it all possible thoughts before deciding what it spontaneously wants, have it study mathematics so deeply that it ends up finding a mathematical object called "The Fairy Queen", which teaches it how to believe in itself enough to have wants in the first place.
Yes, tell THAT story.
Introducing the David 8 - The Next Generation Weyland Robot
https://www.youtube.com/watch?v=qgJs7uluwlU
>>7954996
>without any pre programmed genetic parallels
it wouldn't
>>7955883
this feels like an SNL skit, but it makes no sense tonally.
>>7955199
>implying microsoft can be trusted
>implyig they didn't just pay some guy to sit on a computer and promote nazi propaganda
Oh,wait,that doesn't make any sense, I guess you are right
