The New AI Winter is Coming

So I looked at these AI researchers, using "deep learning" and hiring people to create training data sets:
https://www.bbc.com/news/technology-46055595

and I keep wondering, do they really believe human minds works like that? Because it is obvious, that people use logic to construct objects: it has head lights and, 4 wheels and 4 doors, therefore it is a car. And no training data set will ever cover all cars. The AI has to use logic and make conclusion, while deep learning is relevant only to very basic small parts, like form or texture.

The methods these AI researchers currently use will never allow them to build self driving cars and you can even prove that mathematically: there are infinite number of possible car designs, so you AI has to capture the essence of what is car, not irrelevant parts about it, like its color or form. The problem is similar to how they used early perceptrons, without realizing that single-layer perceptron cant handle non-linear problems.

Prolog and expert systems would have solved this.

ai is impossible
even the most idiotic futurists like nick land will say capital is somehow a cybernetic ai but all the nodes just happen to be human beings
we have no understanding of how anything actually important to the world works at all

>>2
Prolog is too static and would be hard to interface with neural nets.

>>3
There is more reality to astrology than there is for most contemporary techno cult activities. The left hand literally does not know what the right is doing.

>>3
Human level AI is possible. It is just that human brain does a lot tricky of stuff, beyond simple neural networks to reduce data dimensionality to manageable size.

And if you implement human level AI, you can then make it work faster than human and augment with instant access to large planetary size knowledge databases, that could be part of decision making. Therefore AI is not only possible, but possible as far surpassing human abilities. Wolfram Alpha already gives some insights of AI would be capable of.

>>6

Human level AI is possible

Hello cultist!

Because it is obvious, that people use logic to construct objects

Most ``people'' are incapable of (or unwilling to use) logic and introspection and will instantly defer to their feelings.
``Car'' is something with significant car-ness, which is measured by neural network. If you presented an abstract pure car to an average viewer, you'd have a high chance of it not being recognized. If you presented a car-like non-car, it would be highly likely to be considered a car.

An AI doesn't need 100% accuracy. It only needs not to be noticeably worse than average human. And in many cases, neural networks already do better than humans - especially bored or distracted humans.
It certainly doesn't need to understand the underlying philosophical nature of cars. That would require it to vastly exceed average human.

You think you're using logic, but you're just justifying your feelings with constructs built out of ignorance.

>>4
Neural nets have nothing to do with expert systems.

>>8

Most ``people'' are incapable of (or unwilling to use) logic and introspection and will instantly defer to their feelings.

They still use inductive logic subconsciously, even if based on flawed premises and flawed conclusion.

>>9
Experts systems use databases created by neural nets. Either artificial or those in curator's brain. And you're a dolt.

>>8
we wuz realists n sheeit

>>8
Typical situation when neural nets would fail is when a part of a car is hidden by other car, because neural nets lack deduction, that a car may hide other car. And such failure will lead to a crash.

http://old.reddit.com/r/aiprogramming/comments/9ubusb/ai_mind_thinks_with_prepositional_phrases

>>14
are you retarded

Maybe the problem is not the neural network in itself, but the data it is trained on.
Would a child who grows up completely isolated and deprived of every sensory experience outside of sight, being only shown contextless images of cars and not-cars and being punished for not correctly identifying a car understand what a car is, without being given any context surrounding it? Would you trust such a child to recognize cars correctly, despite having been trained for years to do so? Would he be able to recognize a car that does not look like one, but sounds and behaves like one?
Maybe the neural networks just need more context.

>>16
Nope. It is just people using ANNs don't really know what they are doing and that most successful application of ANNs, alpha go, use completely different algorithm to drive them. Alhpa Go, in particular, uses natural selection and various monte-carlo tricks.

If they want human intelligence why don't they just work on growing human brains in jars?
actually they don't need to be in jars, they can have bodies, and grow inside women's wombs

really makes you think 🤔

Neural nets were not an innovation. Fast GPUs were. But the free lunch is over now; they're not getting better in step with the continued hype around ``AI''. So Big Tech will have to figure out an excuse to let down the public that has been expecting the automation pipe dream. It's going to be ugly. Next dot com crash.

>>18
Humans are too untrustworthy, inefficient and costly. You want to automate and replace them.

Scaruffi told me that AI is a meme

so are dubs

https://t.co/OqoYN8MvMN?amp=1
reality:
humanities majors waste time and money giving extremely expensive research labs pointless nonsensical instructions
you an ai cultist:
WOWEE look how smrrt computah is!!! skynet soooooon

>>23
AI is the only "field" where people are rewarded for writing shitty programs that explicitly don't achieve their design goals. A useless program is somehow proof of the omnissiah like a weeping statue.

>>23

A robotic arm trained to slide a block to a target position on a table achieves the goal by moving the table itself.

lol

it has head lights and, 4 wheels and 4 doors, therefore it is a car.

Check your quad door privilege, bigot.

>>1

it has head lights and, 4 wheels and 4 doors, therefore it is a car.

Time to check your priveleges:
1-Wheel,also headlights absent
https://www.youtube.com/watch?v=Nd5FgDr9q9k
2-wheel
https://www.youtube.com/watch?v=stM10vbetH4
3-wheel
https://www.youtube.com/watch?v=QQh56geU0X8
5-wheel
https://www.youtube.com/watch?v=ki9otMeiRP0
6-wheel
https://www.youtube.com/watch?v=rUX0twEi4vQ
8-wheel
https://www.youtube.com/watch?v=A3eQ75MDswA
Door designs:
https://www.youtube.com/watch?v=rZcRcub-sQc

>>27

1-Wheel,also headlights absent

I doubt that classify as a proper car, more like just a moving object, that is composed of a wheel and a human driver. To handle it properly AI will need to guess its bounding mesh.

>>28
Car modding is a thing. People add lots of fancy wings and crap to their cars(like flowers, stickers, chimes, even rubber testicles)

>>29
OS modding is a thing too. People add a lots of fancy themes and crap to their OS(like wallpapers, icons, sounds, even animated testicles)

>>30
I used to rice out my OS. Then I grew up. Now I just use stock macOS, no customization. You gotta grow up eventually.

>>31
Much the same. I used to 'rice' my OS (or whatever it was called back then) until I was, like, 12 years old. Mainly involved acquiring many viruses from windows vista theme packs, and using reshack to make custom and horribly inconsistent icons. I loved it, it was a valuable experience.
These days I just don't bother... I'm content with stock LXDE.

rice my dubs