/sci/ - Science & Math

[Anonymous No.8856202]

What's the bare minimum range of operations an AI should be able to perform?
I'm exploring all sorts of different ways an evolving AI could work, but I'm not sure which ones are even possible to compute with.

[Anonymous No.8856219]

>>8856202
Pleasing my dick

[Anonymous No.8856228]

Pain is nature's teachings.

[Anonymous No.8856242]

>>8856202
That's one cute snake.

[Anonymous No.8856249]

>>8856202
You aren't exploring anything. Get over it.

[Anonymous No.8856258]

>>8856249
Would you like to be explored?

[Anonymous No.8856281]

>>8856202
turing completeness or something close to it

Now get the fuck out of here and read pic related, which I have linked here:
http://gen.lib.rus.ec/book/index.php?md5=C37CE53726FB431E5815F9B1E573BFD6

[Anonymous No.8856286]

>>8856281
EE student with an interest in CS. Is this book any good or just a meme?

[Anonymous No.8856299]

>>8856281
>suggesting a book on AI
>when current AI is always either just database-absorbing or trial-and-error mutation

[Anonymous No.8856311]

>>8856202
A useful AI? Whatever you design it for.
A real AI? Everything a human can.
A skynet AI? Everything a human can and more.

[Anonymous No.8856508]

>>8856311
I mean fundamental, internal operations.

[Anonymous No.8856526]

>>8856508
It needs to be turing complete.

[Anonymous No.8856582]

You should look into Neural Turing Machines OP

[Anonymous No.8856830]

>>8856219
Not gunna lie... but this is part of my "Touring" (my body) test for AI.

[Anonymous No.8857016]

>>8856508
Again it depends. If you're trying a human like AI, nuerons simply connect each other, so all you need is to be able to create pointers and derefrence them. Of course nobody is gonna try to make an AI that literally has every neuron a human does, so simplifications are neded.

Think of it as how we simplify 3D graphics using vertex and polygons instead of trying to render every single athom/molecule. The results are an aproximation, but they are good enough to fake aur eyes.

So, I thught about that and I've come with 6 diferent types of linking aginst entire structerues (ideas) as a possible simplification (if wich I have no idea if it's going to work) but here you have it:

#define Idea2 std::pair<Idea*, Idea*>
lnk_op<Idea*> gen = lnk_op<Idea*>(this); //A is generalization of B
lnk_op<Idea*> elm = lnk_op<Idea*>(this); //A is element of B
lnk_op<Idea*> syn = lnk_op<Idea*>(this); //A is a synonim for B
lnk_op<Idea*> rel = lnk_op<Idea*>(this); //A has a relation to B
lnk_op<Idea2> cmp = lnk_op<Idea2>(this); //A is B compared to C
lnk_op<Idea2> act = lnk_op<Idea2>(this); //A is doing C in time B
#undef Idea2


I would not usually give my ideas for free, so please if you're gonna use this idea make your project open source.

[Anonymous No.8857018]

>>8857016
Ohh seems code tags don't work on /sci/.

[Anonymous No.8857029]

>>8856242
Anon, it clearly has legs. Here's an actual cute snake though.

[Anonymous No.8857569]

>>8856299
Niggah, current AI ain't like that at all.
>> database absorbing
That means absolutely nothing.
>> trial-and-error mutation
deep learning does not even remotely work like that, neuro-evolution is fucking old

This book is a damn good place to start to learn this shit

[Anonymous No.8857618]

>>8856202
The understanding of how the world works for entities that are human-shaped and have human parts and stand on human bits to serve humans because humans humans humans do not squish no.

[Anonymous No.8857848]

>>8857016
Thanks!

[Anonymous No.8858013]

>>8857569
I also soo the book recomended in Amazon, so I may just as well buy it, it better not be about statistical correlations.