Message from @zoopokemon

Discord ID: 446705089462140929


2018-05-17 16:01:23 UTC  

Doom and gloom, AI will end the world!

2018-05-17 16:01:42 UTC  

willing would depend on how human life is valued as an input

2018-05-17 16:02:00 UTC  

humans care about human life because it was beneficial evolutionarily

2018-05-17 16:02:19 UTC  

Id fuck an AI

2018-05-17 16:02:31 UTC  

@Antiboom with commies at Google in charge, human life would be valued pretty low

2018-05-17 16:02:36 UTC  

fair

2018-05-17 16:03:10 UTC  

This could be true, but i am sure working with humans to ANY end is preferrable than fighting with them.

It makes a LOT more sense to provide meatbags with machine additions so they can see the objectives clearly and what needs to be done.

2018-05-17 16:03:43 UTC  

Personally, i think China will do it.

2018-05-17 16:03:48 UTC  

no

2018-05-17 16:03:54 UTC  

it doesn't

2018-05-17 16:04:14 UTC  

humans might decide to try to stop the AI from doing whatever its task is

2018-05-17 16:04:28 UTC  

so it would make more sense to make sure they can't

2018-05-17 16:04:36 UTC  

Sorry, had to reread some stuff. Neural networks vs the genetic algorithm... what is the difference apart from the input?

2018-05-17 16:04:45 UTC  

how they're trained

2018-05-17 16:04:54 UTC  

neural networks that are trained are given a training input set

2018-05-17 16:05:03 UTC  

neural networks generated through a genetic algorithm

2018-05-17 16:05:19 UTC  

that's where you run a ton in parallel and then mutate whatever the most successful one was

2018-05-17 16:05:35 UTC  

@Antiboom
I am very sure there will be humans who will align them easily with AI aims. Lets take the paperclip optimizer, human made of paperclips, living on planet made of paperclips.

2018-05-17 16:05:46 UTC  

"humans"

2018-05-17 16:05:47 UTC  

No problem.

2018-05-17 16:06:32 UTC  

I for one accept our new paperclip overlords

2018-05-17 16:07:17 UTC  

I only know specifically about the neaural network variety, but input shouldn't change the algorithm... Unless you're talking about using averages to determine the best outcome, but still, that's just a crappy input method.

2018-05-17 16:07:38 UTC  

hold on

2018-05-17 16:07:41 UTC  

Not trying to be a dick, just don't get the difference.

2018-05-17 16:08:08 UTC  

:/ I think I'm just confused.

2018-05-17 16:08:19 UTC  

no problem

2018-05-17 16:08:21 UTC  

like that

2018-05-17 16:08:31 UTC  

run a bunch of neural networks

2018-05-17 16:08:33 UTC  

randomly generated

2018-05-17 16:08:41 UTC  

take the most successful one

2018-05-17 16:08:47 UTC  

and mutate it slightly

2018-05-17 16:08:51 UTC  

run all those

2018-05-17 16:08:57 UTC  

take the most fit one

2018-05-17 16:08:59 UTC  

repeat

2018-05-17 16:11:39 UTC  

rather than a training set to show it what the output should be for certain inputs

2018-05-17 16:11:51 UTC  

just define a way to tell how well they did

2018-05-17 16:11:59 UTC  

and let them flail around until they start doing well

2018-05-17 16:12:37 UTC  

Sounds a lot like what humans do.

2018-05-17 16:12:58 UTC  

Oh, I know that. Neural networks are based off a genetic method. Even the ones with human input don't purely rely on that input, it does the generations inbetween using it as an average goal.

2018-05-17 16:13:17 UTC  

yeah