Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Artificial Intelligence Is Already Weirdly Inhuman

nautil.us

Post reply on HN
Print view View original

21–30 of 84 posts

Re: Artificial Intelligence Is Already Weirdly Inhuman

#21

Post by idlewords »

noir_lord wrote: post #5

We saw this in Chess engines about 20 years back (when they started getting seriously strong) they play Chess incredibly well but looks very little like what a human playing does. It was fascinating to watch the changes in chess theory through that period as the machines validated or invalidated concepts and ideas that GM's had posited but been unable to prove one way or the other. John Speelman wrote an intro to The…

This is particularly evident in endgames. Forced mates have been discovered that require hundreds of apparently random and pointless moves, until suddenly the king is cornered and mated. There's a nice discussion of some of these on Krabbé's chess blog: http://timkr.home.xs4all.nl/chess/perfect.htm

Re: Artificial Intelligence Is Already Weirdly Inhuman

#22

Post by pygy_ »

white-flame wrote: post #15

I do have my doubts that plain neural networks will ever be able to achieve conceptual understanding. I have an affinity for classical, rational AI in that you can correct it and it will take that correction and instantly apply it to is knowledge base. It can also explain why it came to a conclusion. (though obviously this style has its very real limitations) NNs and other current statistical/connectionist approaches…

> I do have my doubts that plain neural networks will ever be able to achieve conceptual understanding.

What do you think you're made of?

Re: Artificial Intelligence Is Already Weirdly Inhuman

#23

Post by chriswarbo »

white-flame wrote: post #15

I do have my doubts that plain neural networks will ever be able to achieve conceptual understanding. I have an affinity for classical, rational AI in that you can correct it and it will take that correction and instantly apply it to is knowledge base. It can also explain why it came to a conclusion. (though obviously this style has its very real limitations) NNs and other current statistical/connectionist approaches…

I'm a firm believer that it's a mistake to ask machines to do X in the hope that they'll do Y. We can't train an ANN to classify images then ask it to explain its reasoning; that's not the task we trained it for!

How might we train an ANN to explain its reasoning? One approach would be to learn programs: have the ANN write programs which classify images. Then we have a classifier (run the program) and an explanation of how it works (read the program). We don't have an explanation of how the ANN chose the program, but again, we didn't train it to tell us. In principle, we could keep adding meta-layers; in practice, the search space and evaluation time would explode :)

Re: Artificial Intelligence Is Already Weirdly Inhuman

#24

Post by seiji »

Qantourisc wrote: post #7

Another reason that neural nets do crazy things: they have little knowledge. We see a dog, by looking at it's shape etc. The neural net by looking at features. A solution to this I think would be splitting up the taks, like "finding eyes", "is this fur", etc ... freeze these networks, and use these as input. This would prevent most errors given in these examples. A lot of animals know what eyes look like regardless o…

We see a dog, by looking at it's shape etc. The neural net by looking at features.

Have you ever seen a kid learn language? At first, every four legged animal is either a cat or a dog. Every car is a truck (or every truck is a car). The specific categories get learned over time, but they aren't in any sense "natural."

Re: Artificial Intelligence Is Already Weirdly Inhuman

#25

Post by seiji »

chriswarbo wrote: post #23
white-flame wrote: post #15

I do have my doubts that plain neural networks will ever be able to achieve conceptual understanding. I have an affinity for classical, rational AI in that you can correct it and it will take that correction and instantly apply it to is knowledge base. It can also explain why it came to a conclusion. (though obviously this style has its very real limitations) NNs and other current statistical/connectionist approaches…

I'm a firm believer that it's a mistake to ask machines to do X in the hope that they'll do Y. We can't train an ANN to classify images then ask it to explain its reasoning; that's not the task we trained it for! How might we train an ANN to explain its reasoning? One approach would be to learn programs: have the ANN write programs which classify images. Then we have a classifier (run the program) and an explanation…

that's not the task we trained it for!

Yes, exactly. We're not training consciousness networks. That's not even a goal of all this research. We're training dumb high-speed classifiers.

How might we train an ANN to explain its reasoning?

There are networks that exist today that'll happily explain an entire scene to you (the whole "this picture contains a pizza sitting on an oven in a kitchen and there is a dog in the corner"), which is closer, but still just pairwise training. The "thought" process isn't recurrent or turing complete so it can't make progress on its own.

Re: Artificial Intelligence Is Already Weirdly Inhuman

#26

Post by igravious »

Qantourisc wrote: post #7

Another reason that neural nets do crazy things: they have little knowledge. We see a dog, by looking at it's shape etc. The neural net by looking at features. A solution to this I think would be splitting up the taks, like "finding eyes", "is this fur", etc ... freeze these networks, and use these as input. This would prevent most errors given in these examples. A lot of animals know what eyes look like regardless o…

You remember that bit in Hitchhiker's Guide where Arthur Dent is explaining to the ship's onboard Nutri-matic thing (I think it was) all about what a cup of tea is. After getting back many different versions of a drink that was nearly but not quite unlike tea, Arthur explained about tea leaves and climates and tea trade and tea ceremonies and on and on. I'm sure my recollection and retelling do not do it justice.

My point is, what we've got now is crude visual-feature pattern matching and recognition. What you're asking for is conceptualisation. Wiring up concepts to sense-data is our trick. We are reverse engineering ourselves bit by bit. You make the next step sound so reasonable, and maybe perhaps it is.

Re: Artificial Intelligence Is Already Weirdly Inhuman

#27

Post by crazypyro »

hiddencost wrote: post #6
crazypyro wrote: post #3

(My take on this problem, feel free to correct any assumptions I make that are wrong) I think this is because neural nets are not actually "intelligent" in terms of the commonly accepted definition of intelligence. They are dumb. They try, usually with simple probabilistic techniques and input element-wise transforms, to mimic some function that produces approximations for a given set of inputs and outputs. The train…

That's pretty good. I wouldn't say things are impossible, but there're a lot of conversations happening about the problems with our current approach. Leon Bottou gave a great talk at ICML that covers stuff like this. https://www.google.com/url?sa=t&source=web&rct=j&url=http://...

Thanks for that slidedeck! It was really informative and enlightening to have some of these vague, hard-to-explain ideas that I've been wondering about cemented by Bottou.

Re: Artificial Intelligence Is Already Weirdly Inhuman

#28

Post by faragon »

How vs why

I started learning genetic algorithms (random permutations selected/promoted for picking better combinations of Hamiltonian paths for solving NP-C problems obtaining near-optimal solutions quickly) and neural networks (mainly backpropagation, using different topologies) circa 1995 (college). It was like magic: you learn how to use those techniques easily in order to solve problems. However, the why was not so evident: while in the case of genetic algorithms you can understand that in a space of solutions with many similar cost solutions could allow you to pick a "good one" easily, in the case of neural networks the idea I got was like some a "magic box" which was supposed to do some interpolation/extrapolation that effectively work like generating a convex surface for giving a "solution"/"location" for a given input. It was like you got alien technology simply to make it work for simple things, but not being able to really understand why it was working (except for trivial/small networks).

Do people really understand what complex neural networks do? I.e. is still trial-error only, or are built on purpose?

Re: Artificial Intelligence Is Already Weirdly Inhuman

#29

Post by noskynethere »

crazypyro wrote: post #3

(My take on this problem, feel free to correct any assumptions I make that are wrong) I think this is because neural nets are not actually "intelligent" in terms of the commonly accepted definition of intelligence. They are dumb. They try, usually with simple probabilistic techniques and input element-wise transforms, to mimic some function that produces approximations for a given set of inputs and outputs. The train…

> They try, usually with simple probabilistic techniques and input element-wise transforms, to mimic some function that produces approximations for a given set of inputs and outputs

It's my understanding that this is basically how the brain works. My personal theory is that enough of these "dumb" inputs, wired correctly together, leads to emergent behavior that is consciousness.

Re: Artificial Intelligence Is Already Weirdly Inhuman

#30

Post by pokkanome »

Neural Networks do not think. They perform a series of computations, and arrive at a result. It is not intelligence. It is a very useful tool for solving problems that can be quantified and that we can generate a lot of data from, but ultimately, it is still human intelligence that is interpreting the problem and result.
Post reply on HN