[ 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: 2.01 MB, 3602x3602, Artificial Intelligence.jpg [View same] [iqdb] [saucenao] [google]
9936844 No.9936844 [Reply] [Original]

how do we actually achieve artificial general intelligence? what field can i pursue to work towards agi; while machine learning can sort of mimic some specific capabilities of the human brain, it doesn't really encompass consciousness. would neuroscience be a more suitable field to pursue for agi?

>> No.9936970

>>9936844
>how do we actually achieve artificial general intelligence?
Assuming AGI means an agent-type intelligence similar to an animal or a human, probably integrating a bunch of input and output neural nets with architectures aimed at some specific task using a bunch of other nets that integrate the information and use it to reason about the world in some way. It will be a gradual process I think.
>what field can i pursue to work towards agi;
You can't. You were born too late. Deepmind or OpenAI will get there long before you finish your studies. And even if they don't, AI research has some of the smartest people on the planet working on it. You're not among them. If your IQ is not 145+ and you can't learn the math involved in a few days you're not gonna make it. Unlike software development, AI research has no need for a huge pool of dumb manpower, it's a field that very quickly gets saturated.

>> No.9936976

>>9936844
The way you can tell a field is deeply oversaturated is when there is more money to be made talking about it and giving advice/tutorials on it than in the actual field itself.

The existence of pajeet youtubers like this guy:
https://youtu.be/9bbS-trc8ys

Shows that there are so many people trying to get in that he can leverage it for a steady income. If the field was still young and growing, anybody with a solid understanding would be in the actual field would be working in it, not trying to make a quick buck off others interested in it. Note the number of views this shit gets. ML is full.

>> No.9936984

>>9936976
>>9936970
those are exactly my points. i have enough understanding of the technical side to pursue ML but what im talking about is the actual direction i need to go to achieve agi. it just seems to me that mathematics isn't a model we can use to describe consciousness; although algorithms can come close to simulating it. which is why im asking whether some people on this board who may be in the fields also pursuing the same goal aren't doing ML and instead in medicine or something along those lines

>> No.9936987

>>9936984
ML will achieve AGI. AGI has fuck all to do with the hard problem of consciousness. Neuroscience can't answer that either. As for the much more trivial problem of being aware of its own state, we already have machines that are aware of their own state in some way. There's no reason a bunch of neural nets can't be wired together to make a self-aware one, it's just a matter of figuring what kind of nets you use, where and how the fuck you train something like that and how you make it work with the computing power available - none of which is easy of course, but at least it should be possible.

It's possible the AI will be conscious, but how the fuck do you even test something like that?

>> No.9937010
File: 22 KB, 500x500, 1534169714896.jpg [View same] [iqdb] [saucenao] [google]
9937010

>>9936970
>Deepmind or OpenAI will get there long before you finish your studies.
>We're on the verge of AGI guys, two more decades tops!
Get a load of this guy.

>> No.9937011

>>9936987
>we already have machines that are aware of their own state in some way
eg?

what field are you in, if you dont mind me asking?

>> No.9937032

>>9937010
if your assumption is they have to literally simulate a whole adult brain inside a regular computer so it'll be a while away you're wrong. shit's already going to get weird quick from what they can do right now, publicly.

>> No.9937037

>>9937032
No but it has to have some capacity whatsoever to not be a direct input = output programming line. Which is all we have right now. We don't even have a basic understanding of what qualia so trying to reverse engineer an intelligence isn't an option either.

>> No.9937048

>>9937011
Anything that takes its current state as an input, which is most machines nowadays. You are conflating awareness in the sense of being able to use some information with consciousness and qualia. There is no reason why a machine shouldn't be able to know the state it is in and take it into account, colors or no colors.

>> No.9937051

>>9937010
>two decades
I give it a decade tops. AI is nothing like fusion, because we have real life working examples everywhere that don't really use anything we don't have access to.

>> No.9937052
File: 209 KB, 1064x1447, 1533533535997.png [View same] [iqdb] [saucenao] [google]
9937052

Perhaps find something similar to this: https://agi.mit.edu
Perhaps they'll host/teach one each year.
Perhaps more alternatives will eventually spring up.
Perhaps perhaps perhaps.

>> No.9937053
File: 48 KB, 627x626, 490.jpg [View same] [iqdb] [saucenao] [google]
9937053

>>9937037

>> No.9937055

>>9937052
bruh i fucking love you thanks for this

>> No.9937064

>>9936844
AGI is bullshit pseudoscience. Until its evangelists define their parameters in any meaningful way it won't go anywhere, and even if they do in all likelihood it still won't.

>> No.9937065

>>9936987
>There's no reason a bunch of neural nets can't be wired together to make a self-aware one
There's no reason it can be done, either. Just waving away the hard problem of consciousness does not an AGI magically make.

>> No.9937069

>>9937065
What does the hard problem of consciousness have to do with anything?

One of these days I'll stop falling for low effort bait, I swear.

>> No.9937073

>>9937069
the general intelligence part of artificial general intelligence

>> No.9937083

>>9937069
Consciousness is for all meaningful purposes synonymous with self-awareness, although both are woolly, non-scientific terms no matter how badly AI evangelists wish it weren't so. You raised the issue of defining consciousness in scientific terms, waved it away, then immediately opined that, "there's no reason a bunch of neural nets can't be wired together to make a self-aware one", as though the comedy of what you're saying weren't evident. From where does this magical self-aware property arise? Can you define it? Pin-point where in the mess of neural nets it suddenly emerges? AI evangelists don't know what consciousness is, they don't know what self-awareness is, and they only loosely even know what intelligence is. They don't know what a general intelligence without using torturous circular logic. The whole thing is pseudoscience.

>> No.9937096

>>9937073
General = not specific

Animals exhibit general intelligence.

>> No.9937099

>Can't even do NLP
>Can't remember a lot
>hyperparameters
If I had to bet money I would put it on hyperparameters too bad you can't get rid of those pesky hyperparameters.

>> No.9937100

>>9937083
>You raised the issue of defining consciousness in scientific terms
No I didn't. Consciousness has nothing to do with intelligence. Human behaviour is physically determined and is not influenced by qualia and consciousness, whatever the fuck they are. Intelligence is just computer science.

>> No.9937106

>>9937096
what, only humans have consciousness?

>> No.9937114

>>9937100
Alright. Let's say for the sake of this that intelligence is reducible to non-specific problem solving and adaptive learning. The AI evangelist's conceit for the last decade has been to say that if they throw enough sophisticated neural nets together, they'll eventually get a general intelligence. This is akin to me saying if I throw enough kitchen chemicals into my car's engine, I might create rocket fuel. Do you see how ridiculous it is yet?

>> No.9937118

>>9937052 what's with the pic?

>> No.9937154

>>9937114
Did you even think about thinking before you started typing? Of course that's not how it goes. Neural nets are just the start. Neural nets work because they can be arbitrarily complex but not arbitrarily long(even LSTM's). Just even letting Neural networks be random in size will be like taking 2 million steps. It's called Hebbian Learning and it's already out there it's just not well understood.

>> No.9937170

>>9937052
where did you find this image

>> No.9937202
File: 54 KB, 359x2048, FB_IMG_1534247775523.jpg [View same] [iqdb] [saucenao] [google]
9937202

>>9937118
Reading library expansion suggestions.
>>9937170
On this board, I think. It's beautiful isn't it?
Some guys screenshot of his e-library.
I may make a similar one after I've sorted through my 9001 e-books.
However, it's been relatively difficult to acquire certain desired copies.

>> No.9937208

>>9937202
please do ill wait here :)

>> No.9937251
File: 457 KB, 883x286, TimeTech2.png [View same] [iqdb] [saucenao] [google]
9937251

>>9937208
Thread may expire before then, anon.
You can spoof an email address with something like this: https://www.guerrillamail.com/
- If you don't want to forgo/forfeit the missed connection and share information and whatnot.
Anyway;
I'd recommend this: https://g.sicp.me/books/
The gentoomen library (if you haven't heard of it).

>> No.9937277

>>9937154
What do you see are the next steps to make A.I more like the A.I we would want to see? In the way we would want the 'Master Algorithm', that can do many tasks of learning and memory (one shot learning, recognition, association, transformation, representation, goal-setting, planning)

>> No.9937380

>>9937106
First off, no. Humans aren't special you god fearing fool. Second, general does not mean conscious.

>> No.9937471

>>9937380
that's my point dummy

>> No.9937572

>>9937114
>This is akin to me saying if I throw enough kitchen chemicals into my car's engine, I might create rocket fuel. Do you see how ridiculous it is yet?

Yes, that is a ridiculous analogy. There isn’t a big difference between the brain of a mouse and a human, except the size. So why wouldn’t an enormous neural network be sufficient?

>> No.9937872

>>9937065
>>9937083
>>9937114
but the brain is a reason it can be done even if we dont understand how right now... if the brain can be self-aware then so can networks. and we dont need a definition of consciousness or self-awareness to build an AI.