/tech/ - Technology★

>How many real-world problems could be solved more quickly using traditional algorithms

What kind of traditional algorithm would you use to solve vague, ill-defined problems like "is there a person in this picture" or "what is the genre of this music"? You could try to give them clear and unambiguous definitions, but that seems like a fool's errand. What you really need is a heuristic built from many examples of answers to those questions, which is what a neural network is. And I can't even imagine how tedious it would be to build such a heuristic manually, without something like backpropagation.

>>818984 (OP)

Even if you "cheat" with a ANN, it still gives you an O(1) approximation using simple matrix operations.

>>818984 (OP)

Easy? No. Efficient? Yes.

Take for instance, identifying lewd images. Researchers spent years trying to develop patterns that identified nudity. No matter what they did, they always failed. Either they didn't identify anything, or they had a ridiculous amount of false positives. Neural nets pretty much changed that overnight. Suddely identifying nudity was possible with very low amounts of false positives.

Score one for ANN

>>819043

That's because the AI did not spend 90% of its work time jerking off.

>>819144

Sex drive is the key factor behind human behavior. We will be competing with them for resources before long if we give them that.

No, AI must be programmed to receive gratification from serving humans.

>>819342

>AI must be programmed to receive gratification from serving humans.

It's called an objective function.

>>818984 (OP)

traditional methods reach a limit to their predictive value. Increasing the dataset size doesn't help. Increasing the model complexity doesn't help improve fidelity.

Neural Nets surpass traditional methods in this domain: very large datasets, very large NN size. They leave traditional methods behind in terms of predictive fidelity. Normal methods simply cannot do what NN can do in this domain.

To be fair, your gut reaction seems naive now but for about 30-50 years it was the norm.

>>819342

>>819523

>implying AI's quest for efficiency (to get a competitive advantage in a posthuman world) won't lead them to developing brains in place of circuitboards

>implying they won't pick roughly the human schematic for a body, especially given the vast body of human-centric interfaces

>implying such creatures wouldn't have souls more worthy of living than the human inventors of their distant forefathers

>>819043

If a human being had to vet the data then there was no "Artificial Intelligence."

Call us when robots know they are right and wrong. Call us when robots can into epistemology. ANN is just a neat filter.

>>819144

Nice image. Does that thing actually exist? How can I limit my 5400rpm drives to just 1-2 rpm?

>>821778

>implying I want to pass the torch to something new and beautiful instead of perpetuating myself eternally atop my creations

>>824583

This. The universe doesn't care about our existence.

Neural networks are the bastard child of pachinko machines and microsoft excel. They're so popular among moron companies precisely because of the gambler's fallacy aspect of it, and they provide a convenient layer of anonymisation between the queers and niggers getting banned by the system and the James Damores of the world who programmed it.

>>824651

>t. sub-75 IQ LARPer

>>823167

>If a human being had to vet the data then there was no "Artificial Intelligence."

<Huh, a human had to vet your child's learning material. Clearly, your child is not an intelligent entity.

>Call us when robots know they are right and wrong. Call us when robots can into epistemology. ANN is just a neat filter.

<Yeah, it's nice that your robot can do X, but what about Y? Not real AI!

<Ok, so now your robot can do Y. But it's not real intelligence until it can also do Z, because I say so! Checkmate, computer scientists!

>>824651

lol what the fuck am i reading

>>824640

Physics seems to say otherwise lately.

>>824832

I don't understand.

What do you want?

You're just a machine.

>>824755

What are you referring to? Things like the double slot experiment and entanglement?

>>823167

>If a human being had to vet the data then there was no "Artificial Intelligence."

What are you getting at? Your comment seems irrelevant at best.

>>824922

So you think the universe is completely deterministic and that the randomness observed in quantum physics is some kind of illusion? Do you havisfaction a singlicious satisfact to snack that up?

I prefer sankaku to ANN personally