Is AI overhyped or is it really the future?

is AI overhyped or is it really the future?

Attached: ai.jpg (721x960, 76K)

Both.

There is no AI and there is no real distinction between AI and any algorithm.
I have a superhuman pokerbot based on a paper by John Nash from the 50s. Looks intelligent, but it really isn't.

>There is no AI
Yet. Maybe in a few decades, but anything called AI today is just quasi-ai. A false meme.

Overhyped right now and I've been doing research with it since 2010. We need a better training algorithm backprop isn't all it's cracked up to be. You want to hurt ml guys see if a deep network can learn modulus.

Besides the training itself, there are alway underlying models or rules.
That's why AI is so good at games. Simple rules, many states. Basically Put a lot of computing power into it, bruteforce the endgame and find some approximation for the early game.
The ability to brute force the endgame alone is probably enough to beat the average player.
Most game AIs also only work well for 2 player games, add more players and the difficulty increases exponentially.

Then for many things there are no mathematical rules. How will an AI interact with a group of humans? What would the correct behaviour even be?

By giving it the means to try deduce or guess at random, based onthe other players initial movements the other players capabilities. Usually in a game you can see how good a player is just the first few seconds you see his movement, not even the fight itself.

t. NNlet

For automation? No. Artificial General Intelligence? Yes.

mine is really goood

>there is no Ai
>Quasi Ai

What I dont understand is why people keep creating their own bootleg definition of what Ai is based off science fiction. Instead of just learning how machine learning and deep learning works. And accepting that IS AI. AI isnt artificial sentience, a technological lifeform or a conscious mind. It's just machine learning based neural networks

So support vector machines aren't AI? Even if they perform better in a certain task?

A support vector machine is a support vector machine. Like I said people need to stop creating their own definitions of AI.

>support vector machine
Just googled it. It is Ai. Its a machine learning training model. Which is exactly what Ai is.

And besides SVMs, people act like there are only 3-4 AI algorithms that matter.
Poker was solved using CFRM, you probably never heard of it, but it solved a human domain game, not a toy game, a game that is bein played by people for thousands of dollars.
Is that not AI, because it's not a NN?

>Just googled it
Hot kek.

Then thousands of researchers are wrong, because SVMs are considered an AI algorithm.

Lol is the field really that immature? I began to figure out remainder tricks in the 3rd grade with no help whatsoever.

I misread your post.

NN and backprop algos are incapable of understanding rules. The only rules are the kernels and everything else is a weight that can only draw probabilistic decision lines.

So, put in a simple way, a modern "AI" would never know that a hot surface will burn, it will just have a high probability for a burn to bappen.

Well maybe that example isn't so concrete (leidenfrost effect, exposure time factors, etc.), but it serves for illustration.

Machine learning is over-hyped as shit. Sure there are plenty of applications but it isn't the end all be all for agent design.

Most of the issues surrounding AI are due to an over use of AI and automation as umbrella terms.

Yes

Lurking for Soren's input