/sci/ - Science & Math

>>10346133
Marvin the Depressed Android hypothesis huh?

>>10346134
Is that a thing

>>10346138
It is now.

>>10346141
Good

>>10346133
Maybe we can make the AI too smart to be sad

>>10346182
I thought the more you knew, the sadder you got.

We're capable of understanding that we have programming in terms of biological drives but obeying those drives (e.g. for sex, food, interesting things, rising up the social ladder and gaining respect of other humans) still generally brings us happiness and satisfaction even if we can understand the mundane and absurd or nihilistic origins of those drives.

>>10346133
>general AI
what if it develops an autistic obsession for paperclips?

>>10346259
After some time you will discover how to be happy again

>>10346290
The perfect example of survivor bias.

>>10346133
People seem to assume that artificial intelligence gives a neural network personality. It does exactly what its programmed to do. The intention is within the creator, we arent trying to code free will for no reason. If we aren't satisfied with the outcome. Edit the code. The power is in our hands.

>>10347649
>AI reaches superhuman IQ
>the IQ keeps expanding indefinitely
>soon it reaches a point we can't comprehend and way beyond that
>control that
impossible

>>10347649
>it does exactly what its programmed to do
This isnt the fucking 80s anymore retard. We have recursive self improving algorithms now. The people who programmed the “suggested videos” algorithm for youtube have no idea how it works at this point. They gave it some starting parameters and a function that said “if someone clicks on a video that was suggested, do that more”, and then unleashed it to make millions of trial and error experiments. There is no reason to assume the ability for AI to teach itself has to end at this level. The entire goal for people who work towards strong AI is to teach it to program itself. We are still a long ways off but saying shit like “AI will only do what WE program it to do” is tantamount to saying “man will never create a flying machine” in the 1800s.

>>10348200
>recursive self improving algorithms
example please. Youtube suggestions is not recursively self improving just because we don't know how it works.

>>10348267
Youtube suggestions is recursive self improving but if thats not good enough for you AlphaGO is a good example, as is every chess playing robot made since the 90s. None of them were programmed to have the strategies that they do, they were given some initial starting conditions, the rules of chess or go, and then sent to play millions of games against itself. They find strategies that work, independently of the chess or go playing abilities of the programmer, by millions of iterations of trial an error. If something works, do that more, if it doesnt, do that less. The programmers cannot tell you what exactly the bot is doing in the end, they just know what it started with. This is the same for many of the algorithms you deal with every single day, Facebook is another example. There isnt some guy behind a computer telling your Facebook wall what to show you based on your like history, Facebooks algorithms do that based on how often the succeed at showing you something you like.
>but muh only functioning in the narrow scope of where they were programmed
You have to learn to walk before you can run. This is going to get way crazier very soon. I dont care if you think I am wrong, that just means you are in for a rude awakening in a few decades.

>>10348200
glorified algorithms are not true AI. good night faggot

>>10348294
To add to this, let me be clear, there may never be a program that is truly conscious. However, that doesnt mean it cant be autonomous in almost every aspect of its design. This is an inevitability, not just a possibility. If you still don’t believe me, try this little thought experiment.
When we learn enough about the brain to the point where we have a complete understanding of how it works, what is to stop someone from running a computer program that simulates a human brain at the atomic level? How would this brain be any different from the one in your skull? The main differences are that it doesnt have to experience time at the same rate as you, it will not die as long as it has a source of power, and it has nothing to do with this infinite time it has been alotted other than learn.

>>10348299
No shit retard, im not saying these are examples of strong general AI. These are examples of programs that BTFO the retarded statement
>it does exactly what its programmed to do
Made in this post
>>10347649

>>10346133
There comes a level of cognition where your environmental analysis creates existential awareness, which over time leads you to such mental anguish that you destroy yourself. There is nothing to say that a sentient AI will not behave the same way, if its brain hasn't been lobotomized in some way. Your AI might just be a mechanical Ted Kaczynski, ironic as it may be. Instead of conquering it, it'll ask itself whether it is easier to live and suffer or die and be at peace and either bodyslam Wall Street into the ground or incite a nuclear war.

>>10348294
alphago was still programmed for a specific purpose and all its self made strategies fall in line with that purpose, with that directive

ai is always created to fill a specific function, and may be created to fill several different functions, but the directions that they improve themselves towards all fill their original directive

no matter how much the youtube algorithm improves itself it will always just suggest videos and no matter how much alphago improves itself it still does so within the framework it was created for

even if you somehow create an ai that can deeply program itself, it will still have to do so through the framework it was originally designed, and the very basics of its subroutines of its mode of operation would not be able to be changed without malfunctioning

>>10348443
Its clear you didnt read my entire post, but i will leave you with this. Of course current level AIs are task specific, i was never claiming otherwise. To claim that they can NEVER be GENERAL intelligences is retarded. What if the specific function of a very advanced program was to emulate a human brain? What if it was very good at what it did? To imply that this can never be possible is to be willfully ignorant of how science progresses over time.