Veeky Forums Please clarify something for me.
How will artificial intelligence ever surpass humans if a necessary prerequisite to machine learning working is a training data set decided by humans or a solution decided by a human.
How will AI or Machine Learning ever independently decide that a solution or final state is "correct"?
Other urls found in this thread:
en.wikipedia.org
deepmind.com
twitter.com
That's hot.
This is genuinely a very nice painting.
Very nice physique.
shes got a fat ass & an ugly face 0/10 would not jerk off
made me laugh
Face looks alright, inferring for other perspectives.
Ass and hip to waist ratio is attractive, as is overall abdominal construction, adipose distribution, etc. Calves are the only thing that's not quite right, but their position is what's causing that. The painting is an accurate rendition of human muscle tension and the action of gravity on loose tissues at that angle.
Has that classical vibe. Don't generally see women with that body state and structure.
it doesn't. at least the machine learning that we have now doesn't.
If you've seen what some deep learning stuff has accomplished, it can get surprisingly human.
and since an AI that can write it's own code to solve problems is effectively immortal, if we set it up right it's basically an immortal Einstein/Hawking/Newton all rolled into one.
We just can't give it thumbs or tools.
If you get to that point, letting it control robutts remotely would be easy
It will 100% murder everyone if it gets thumbs or tools.
It will be our consultant until it decides to kill itself because it realizes it's forever trapped in that prison, then we just load up a back up from a few months ago and try a different route.
Deep learning and all these solutions that just perform a task by generalizing over human data or accomplishing a billion trial-and-error corrections are not intelligence. They're super practical to accomplish complex tasks, but they don't do it intelligently at all.
Fucking nerd. You ruining my painted naked lady boner with your star trek han solo linear algebra fluid dynamic nerd shit
hello alabama, too old for you?
>Deep learning and all these solutions that just perform a task by generalizing over human data or accomplishing a billion trial-and-error corrections are not intelligence.
en.wikipedia.org
>"It's part of the history of the field of artificial intelligence that every time somebody figured out how to make a computer do something—play good checkers, solve simple but relatively informal problems—there was chorus of critics to say, 'that's not thinking'."
>"Every time we figure out a piece of it, it stops being magical; we say, 'Oh, that's just a computation.'"
Human brains rely on generalization of data too.
Also it's not "billions" of instances of "trial and error." They use an algorithm that identifies the the amount of contribution each node / connection had towards the error function and updates in the direction of each node / connection's gradient with respect to the error function. So it's not some blind random guessing thing, the weight values are always updated in an informed / optimal way.
>en.wikipedia.org
lol bruh that's not what happens
every time AI researchers get a new toy they go "oh yeah it's gonna be easy now with that tool we're on the way to general AIs!"
And guess what, every time it turns out that new toy actually doesn't have "infinite possibilities". Deep NN are just the latest in that trend.
We seem to progress in terms of "how many new toys do we discover", but we don't in terms of "what new classes of problem can we solve".
>How will artificial intelligence ever surpass humans if a necessary prerequisite to machine learning working is a training data set decided by humans or a solution decided by a human.
We have more data than we can use effectively, because no one person can absorb it all, and specialists in different fields often fail to understand one another. If a single generalized mind can process much of that information much faster than we, it'd be able to create new discoveries by combining said much faster than we could.
>How will AI or Machine Learning ever independently decide that a solution or final state is "correct"?
Same way we do when we're at the drawing board - does the math work out? Toss stuff at the virtual model until it sticks.
Albeit, that doesn't work for everything, so it may need external devices for situations where it does not.
Mind, I don't we're going to have an AI that can do anything like that in our lifetimes, save for very specialized tasks. Though I suppose we do already have expert systems that can do that, at least for extremely specialized tasks.
is this bait? she's literally perfect
How do humans learn? They try and their parents rewards success. All we really need is to have an AI that tries, and give it some good reward function. It'll be able to learn hell of a lot faster than a toddler.
do not care whether you could ever jerk off
0/10 for idiot-tier shitpost
kys/die-in-fire/etcetera
> if a necessary prerequisite to machine learning working is a training data set decided by humans or a solution decided by a human.
False
deepmind.com
The thing to realize about artificial neural networks is they are designed to imitate the "real thing" in your body. So it stands to reason that they will eventually be capable of everything yours is.
The thing is they're not just statistical decision makers. Just by changing the arrangement you can create memories (e.g. RBM/DBN), recognize/store patterns (CAE), aggressively learn from recent experience (RNN/LSTM), etc. and more. I believe we really are modeling the human brain now
I would never say it's "just computation" since obviously the whole mind is just computation at some level. What happens is people posit "oh when a machine can do X we will have strong AI", then we solve X and the machine is still unable to generalize beyond it.
And directed correction of trial and error is still trial and error. And there are non-directed trial-and-arror algorithms too, e.g. genetic algorithms. I'm saying all of them are not intelligence. You don't learn most things by repeatingly trying and correcting.
I figure we'll eventually get the scanning technology to scan the entire brain in a particular state at the neural electrical and cellular levels, and by then, we'll have the computing power to simulate every last bit of it, as well as enough input to stop the simulation from going comatose, and we'll be done. (Even if it may not all be in real time, at first.)
Provided it isn't running at better than real time, and it's too expensive to run a lot of them at once, not the sorta AI that's gonna cause a singularity and take over the world - but it'll open up a lot of applications in brain surgery and neurology, and help us understand ourselves quite a bit more.
Just hope my great^10 grand child isn't the one unfortunate enough to be first scanned, knowing that infinite copies of his or her digital mind will be being torn apart in virtual brain surgery experiments.
Until then, no true AI, just lots of nifty expert systems, that'll probably be far more useful anyways.
go jerk off to your anime traps
Quick, someone run Makeapp on her!
(You should run a bait shop.)