[ 3 / biz / cgl / ck / diy / fa / ic / jp / lit / sci / vr / vt ] [ index / top / reports ] [ become a patron ] [ status ]
2023-11: Warosu is now out of extended maintenance.

/sci/ - Science & Math

View post   
File: 953 KB, 3000x4000, 1387444567595.jpg [View same] [iqdb] [saucenao] [google]
6236180 No.6236180 [Reply] [Original]

why is artificial intelligence so difficult?

>> No.6236182

>>6236180
because it's non-biologists trying to synthesize biology.

>> No.6236193

because scientist of today think deeply instead of clearly. artificial intelligence is the wrong approach to take, since you are trying to create intelligence not artificial intelligence.

when computer scientist start realizing (like stephen wolfram) that the computer can easily program itself ones you allow it to look for patern, they'll make more progress in a year then in all previous years.

>> No.6236202

>>6236193
>the computer can easily program itself ones you allow it to look for patern

how do you allow a computer to look for patterns?

>> No.6236211

>>6236202
By giving it intelligence

>> No.6236284

>>6236180
psychologist can't into science and make arbitrary definitions of phenomena and ambiguously speak on the workings of thought

More simply

Its hard to turn psuedoscience into science

>> No.6236292

>>6236180
It used to be easier but then they started using C++ and Python instead of Lisp.

>> No.6236305

1) because the hard things are easy and the easy things are hard

2) because we dont know enough about the human brain to simulate it properly.

We can make artificial neural networks that can act in the manner of simple animals, but the complex systems of the frontal cortex are very far off.

Also I will take questions concerning artificial neural networks.

>> No.6236315

>>6236193

>thinking deeply instead of thinking clearly

why cant I hold all these buzzwords?

>> No.6236361

it's not, we already have it.
it just speaks a different language. but it can solve any problem, as long as we're able to describe the problem in its language.
the only difference right now is that computers only act because we tell them to.

>> No.6236367
File: 1.17 MB, 3537x2924, 1387459869633.jpg [View same] [iqdb] [saucenao] [google]
6236367

>>6236180

It depends on what you want.

You can create great algorithms for exact, small problems. You can even beat a human with many tasks.

But what makes our intelligence so outstanding is it's multifunctionality. There are quite a lot of systems working in our brain which are autonomous and related to each other.

Teaching a computer what is "funny" is a hell lot of work, because it requires a lot of pattern matching and probabilities as well as creativity.


"Hey, what's up man?"
"The sky and the ceiling."

You can maybe build computers that recognize this as a joke, but it's a different story to recognize "the dude is now telling me a joke because I told him I'm upset" in a regular conversation:

"knock knock whos there?"
"my uncle died"
"oh shit, I'm sorry."
"no, just kidding"
"what's wrong with you, man! that's not funny"

this conversation requires:
a) understanding that someone's death is more important than finishing the joke
b) reconfiguring your "emotional setup" to react adequately
c) understanding how and why this is not funny

under slightly differnt circumstances the whole conversation would have had a complety differnt meaning. You can't just make up a rule, it's more like a very complex ruleset that changes with every moment.

>> No.6236372

>>6236367
the problem here is that you have a poor understanding of human humor, you don't know why we think it's funny
the machine has nothing to do with it

>> No.6236375

>>6236372
"humor" was just an example.
(like the faces in the jpg).

And there are a lot of different styles of humor - I could elaborate that..


The point of interest is that you can "compute" all these data simoultaneously. A computer that could be called "intelligent" would have be supposed to do the same thing - which is highly probelmatical.

>> No.6236377
File: 504 B, 32x32, 1387460897661.png [View same] [iqdb] [saucenao] [google]
6236377

>>6236367
that's cool, but my mind doesn't need a programmer to teach me algorithms

>> No.6236382

>>6236377

I'm not sure I understood what you were just trying to told me.

>> No.6236386

Because human intelligence is nothing more than a massive amount of learned information over decades.

There have been efforts to create ontology databases (the categorization of all common knowledge), but it took decades to create one that contains barely a percent of what an average teenager knows. You cannot understand speech without knowing everything that an average human knows, because you would not understand the concept.

Ironically, the problem of artificial intelligence is not creating the right program or algorithm. We got that down 20 years ago. The problem is speed and storage, and also the lack of digitized information to teach.

If you wanted to create an artificial human, you would need to create an artificial neural network consisting of 10^11 nodes, 10^14 connections, make it fast enough that it can calculate the whole network 20 times a second, and then immerse it in our world for 20 years like you would immerse a human. Of course, even if you could do this, it would miss information from all the senses of a human, and the emotions generated by hormones and shit. Humans are rewarded by endorphine and punished by pain and sadness. This has a massive effect on configuring our neurons's "weight function" to seek reward and try to minimize punishment. Without this, an artificial network would have problems learning and understanding how a human thinks.

>> No.6236416

>>6236386

the problems of scale arent too bad.

100 trillion connection weights is 100 terabytes, if you are careful with space.

I'm sure a 100 terabyte RAM supercomputer isnt out of the question.

And also we can make digital hardware mimic the parallelism of the brain via batch processing.

>> No.6236437

>>6236382
Every algorithm is designed. But the brain evolves, and wires itself up

>> No.6236460

>>6236377
>>6236437
But that's wrong, your brain uses mostly the Hebbian/anti-Hebbian learning algorithm.
If two neurons often fire at the same time, the connection between them will become stronger.

Your brain does change itself, but the change is dictated by a set of simple rules.

>> No.6236462

>>6236180
because its unpredictable if not done corectly

>> No.6236484

>>6236460
Then why can't you simply throw bunch of neurons on a pile and provide stimuli? They're trying: The number of virtual neurons goes up, consciousness is nowhere to be found.

>> No.6236497

>>6236484
Because of this >>6236386

>> No.6236514

>>6236497
Alright, if the answer is "speed", forget about sentience. Evolve me a primitive algorithm. Something easy but non-trivial. You have computers with cat's-brain-worth number of neurons, let it learn how to solve a problem. I'm a computer scientist, but I ain't seen nothing like that yet. All I've seen are genetic algorithms, but those are set up to evolve better ways of solving things they are designed for. No program can learn to solve new problems.

>> No.6236574

>>6236514
You know that a cat's brain number of neurons is the same order of magnitude as a human one?

Also, I think you are mistaken.

>> No.6236589

>>6236574
Yeah, I looked it up, it was hype. Still, there is absolutely no sign of Strong AI to be seen as of yet.

>> No.6236595

>>6236589
What do you call a Strong AI?

>> No.6236600

>>6236595
Let's start with this: Anything that transcends its design in that it learns to solve tasks it wasn't programmed to solve.

>> No.6236613

>>6236514
>I'm a computer scientist
>I'm taking undergrad computer science
>I'm in trade school for computer programming, but they call it "college" and "computer science" because this is 21st century Unitedstan of America where everyone is equal

>> No.6236614

>>6236600
Why would something like this be possible?
I mean YOU can't solve tasks you were not programmed to solve...

>> No.6236618

>>6236600
So "Strong AI" to you means machine learning (well established) plus serious bugs in the part that's supposed to keep it on-task.

>> No.6236620

>>6236613
>thinks ad hominems are valid arguments

>>6236614
When you have a new idea, are you not solving a problem you're not programmed for?

>> No.6236622

>>6236618
Bugs don't solve new problems.

>> No.6236625

>>6236620
>doesn't know what an ad hominem is
>doesn't know that 4chan isn't debate club
>doesn't know when he's being mocked and dismissed rather than argued with

>> No.6236628

>>6236625
>doesn't know what he accuses someone else of not knowing
>says he doesn't want to contribute to the conversation
>keeps making comments

>> No.6236629

>>6236620
>When you have a new idea, are you not solving a problem you're not programmed for?
How is the idea constructed if not by applying a (complicated) program?

>> No.6236636

>>6236629
I'm just saying that there is a discrepancy between Strong AI and Weak AI. You claim there isn't, so I demand proof. How did the burden of proof fall on me? It wasn't me who claimed to know how creatures learn or become sentient.

>> No.6236644
File: 118 KB, 567x411, 1387474527804.jpg [View same] [iqdb] [saucenao] [google]
6236644

>>6236620
>When you have a new idea, are you not solving a problem you're not programmed for?
The human brain is a computer. Hell, it even uses electricity to transmit information. Its programming is its ability to strengthen or weaken the links between neurons. When you find a good "neuron configuration", you have your solution to your problem.

Yeah, the brain does have a programming. The programming is this ability to change itself. The ability to solve a problem is merely a piece of memory, or experience that your neurons learned over the years. This is the reason kids are given toys like "insert the square shape into the square hole". They are designed to create the problem solving "circuits" in the kid's brain.

But the brain can do some other pretty amazing things just like rewiring neurons. Some crazy scientists cut off the eyes of a lab rat, and wired its optical nerve to the animal's auditory cortex. After a while its brain learned how to process the information, and became able to recognize shapes again. This is the programming of the brain, and we can emulate it.

>> No.6236656

>>6236628
Look, chum, an ad hominem is when you make a logical argument that someone else's logical argument is wrong specifically because of who made it.

It's not when you just insult or mock someone, it's not when you shoot down an appeal to authority, it's not when you call someone's qualifications into question in a general sense.

Almost no one ever makes a genuine ad hominem argument. An ad hominem is not just something that's frowned on arbitrarily as a bad practice, it's a piece of reasoning that's necessarily invalid due to a very particular type of fallacious reasoning.

With very few exceptions, people who whine about "ad hominems" on the internet are obnoxious idiots who are terrible at logic.

>> No.6236658

>>6236636
I was merely commenting on you distinguishing between "Strong AI" and "Week AI". I accept your definition, but I don't think "Strong intelligence", as you define it, exists.

All i'm saying is, it is possible to compute a "Weak AI" that would be equivalent in behaviour to a human being.

>> No.6236668

>>6236656
The assertion that someone is wrong need not be explicit for the (lack of) argument to be ad-hominem. You fail parrot.

>> No.6236674

>>6236658
>weak = strong
I get that you're saying that - that's precisely what I have an issue with. I see no proof of that.

>> No.6236676

>>6236180
Because we don't know how natural intelligence works.

>> No.6236683

>>6236676
But you just said it's no different than artificial intelligence

>> No.6236686

>>6236674
I'm not saying that, i'm just saying that, as far as I know, we don't have any evidence for the existence of "Strong AI".
I agree that, with your definition, weak != strong, but I believe "human intelligence = weak intelligence", as said by >>6236644

>> No.6236691

>>6236180
>why is artificial intelligence so difficult?
Because we don't even understand what 'consciousness' is on a scientific level, so how can you create something non-biological that does the same thing if you don't understand it in the first place?

>> No.6236693

>>6236377
what is a professor

>> No.6236697

>>6236683
I haven't posted in this thread before.

>> No.6236701

>>6236691
>so how can you create something non-biological that does the same thing if you don't understand it in the first place?
Science does that all the time. We don't completely understand half of the technology use use today.

>> No.6236704

>>6236701
Yes we do.

>> No.6236707

>>6236691
If you are able to create questions, make hypothesis, and then discover answers without any external influence(such as me now wondering what colour underwear you have on right now. I assume black) then you have consciousness.
That is really the only difference between man and machine.

>> No.6236717

>>6236668
>The assertion that someone is wrong need not be explicit
It does need to be specifically because of who's making the argument, though.

"We shouldn't bother listening to this guy because he's an idiot." isn't an ad hominem argument. "This guy is an idiot because his position is wrong." is the opposite of an ad hominem argument. So is "This guy is supporting such an obviously wrong position because he's an idiot."

It has to be some form of, "We know this position is wrong because of the person making it." It has to be an actually logically fallacious proposition, not just rudeness or unwillingness to respect and engage.

Just because you want the conversation to be a formal logical debate about a specific topic of your choice doesn't make it one, and pretending it is one when other people have chosen to mock and dismiss you instead just makes you more ridiculous.

>> No.6236722

>>6236693
external stimuli

>> No.6236724

In before transhumanism and people jerking off to LessWrong.

>> No.6236753

>>6236717
>"We shouldn't bother listening to this guy because he's an idiot." isn't an ad hominem argument.
There is no way that sentence could be a valid argument against what the person it pertains to claims. What you did certainly wasn't. Now, had you to explicitly said that you were simply aiming to simply to discredit my "credentials", it would be a different story. But without such context, the assumption is that it applies to the claim discussed, which was not my "credentials" - and therefore - you've guessed it - was ad hominem.

>It has to be an actually logically fallacious proposition, not just rudeness or unwillingness to respect and engage.
>>6236668 (succinct form of the above)

>> No.6236786

>>6236180
you need to trap light to make intelligence

>> No.6237283

>>6236180
It took evolution hundreds of millions of years to make it the first time. Making it the second time is going to be pretty damn hard.

>> No.6237302

>>6236753
>>"We shouldn't bother listening to this guy because he's an idiot." isn't an ad hominem argument.
>There is no way that sentence could be a valid argument against what the person it pertains to claims.
There's no reason it has to be. What's wrong with your brain that you can't understand that?

Just because one person takes a position doesn't obligate the other people around him to either agree or make well-formed logical arguments against him.

Mockery, dismissal, and abuse are also valid choices, and they're exactly what you should expect if your position is obviously idiotic or incoherent. They're also not ad hominem arguments.

>> No.6238198

>>6237283
Man-made would obviously have it sped up. Why would it need to wait the whole time?

>> No.6238214

>>6236180
>op's pic, strongsads ideal wife.

>> No.6238232

>>6238198
Because we don't even understand completely how our brains work to begin with, try to reproduce something you don't know for sure how it works.

>> No.6238440

>>6236707
No your "conscious" decision to say black was a combination of past events and randomness.
Computers can be programmed to do this.

>> No.6238448

>>6236180
it isn't

>> No.6239014

>>6236460
>your brain uses mostly the Hebbian/anti-Hebbian learning algorithm.
That's not the interesting point.
It's like saying "A house consists of bricks, so if we have enough bricks we can build any house."
The interesting point is how they work together, the emergent properties. Large-scale is quite different then small scale.

>>6236644
>The human brain is a computer. Hell, it even uses electricity to transmit information.
My telephone also uses electricity to transmit information. You're a hell of a pleb.
The "computer metaphor" is decades old and discussed broadly. It's still a metaphor, though.

>>6236707
>That is really the only difference between man and machine.
Ahahaha... You obviously know nothing about the human mind.

>> No.6239041
File: 8 KB, 200x200, 1387571966154.jpg [View same] [iqdb] [saucenao] [google]
6239041

Because intelligence is immaterial.

>> No.6239050

>retarded computer scientists think they can into biology

>> No.6239072

>>6236284
this, they really cant

>> No.6239091

who teaches professors how to profess?

how teaches professers how to teach professors how to profess?