Imagine your Labrador’s smile on a lion
November 3, 2019 8:22 PM Subscribe
We've already seen how AI image-processing algorithms can be dumb and smart at the same time, and that's on full display with NVIDIA's latest interactive demo. "GANimal" lets you take a picture of your favorite pet and transfer its expression to any random animal or breed including pugs, hyenas, bears and lions. That creature will then "smile" or pose just like your Fluffy or Fido.
Never seen that character. Must rectify.
posted by Johnny Wallflower at 9:37 PM on November 3, 2019
posted by Johnny Wallflower at 9:37 PM on November 3, 2019
Never seen that character. Must rectify.
Ya TOLD me to ya told me to ya told me to!
posted by Greg_Ace at 10:16 PM on November 3, 2019 [2 favorites]
Ya TOLD me to ya told me to ya told me to!
posted by Greg_Ace at 10:16 PM on November 3, 2019 [2 favorites]
although to be honest none of this really has much to do with the original post but I couldn't resist
posted by Greg_Ace at 10:19 PM on November 3, 2019 [1 favorite]
posted by Greg_Ace at 10:19 PM on November 3, 2019 [1 favorite]
Great. How is this AI please?
posted by GallonOfAlan at 1:16 AM on November 4, 2019
posted by GallonOfAlan at 1:16 AM on November 4, 2019
The work is powered in part by generative adversarial networks (GANs), an emerging AI technique that pits one neural network against another.
posted by Johnny Wallflower at 6:22 AM on November 4, 2019
posted by Johnny Wallflower at 6:22 AM on November 4, 2019
I Have No Mouth, And I Must Smile
posted by roger ackroyd at 7:22 AM on November 4, 2019
posted by roger ackroyd at 7:22 AM on November 4, 2019
The work is powered in part by generative adversarial networks (GANs), an emerging AI technique that pits one neural network against another.
The (I think legitimate) question is whether that makes this an example of "artificial intelligence", or an example of "some neat-looking gunk spewed out of a neural network."
posted by Wolfdog at 8:09 AM on November 4, 2019
The (I think legitimate) question is whether that makes this an example of "artificial intelligence", or an example of "some neat-looking gunk spewed out of a neural network."
posted by Wolfdog at 8:09 AM on November 4, 2019
Is it working for anyone else? I can't seem to get it to function at all.
posted by GoblinHoney at 8:23 AM on November 4, 2019
posted by GoblinHoney at 8:23 AM on November 4, 2019
So last night I was listening to some quite enjoyable jazz covers of Mozart that were made by humans.
"Creating thing A in the style of thing B" is an intelligent act, right? No matter how bad the execution?
I really don't think calling something "AI" entails any grandiose claims. "Artificial" is still right there in the name. It's just a label that we have been using for this kind of mathematical model, nothing more.
Anyway, these images are making me think about the possibility of using GANs to illustrate a reboot of Animorphs.
posted by esker at 10:23 AM on November 4, 2019
"Creating thing A in the style of thing B" is an intelligent act, right? No matter how bad the execution?
I really don't think calling something "AI" entails any grandiose claims. "Artificial" is still right there in the name. It's just a label that we have been using for this kind of mathematical model, nothing more.
Anyway, these images are making me think about the possibility of using GANs to illustrate a reboot of Animorphs.
posted by esker at 10:23 AM on November 4, 2019
The (I think legitimate) question is whether that makes this an example of "artificial intelligence", or an example of "some neat-looking gunk spewed out of a neural network."
If some theories turn out to be correct, that might be an accurate description of so-called real intelligence!
/hamburger
Is it working for anyone else? I can't seem to get it to function at all.
I've definitely had days like that, too.
/cartoon vaudeville shepherd's crook-impelled exit stage left
posted by Drastic at 12:35 PM on November 4, 2019
If some theories turn out to be correct, that might be an accurate description of so-called real intelligence!
/hamburger
Is it working for anyone else? I can't seem to get it to function at all.
I've definitely had days like that, too.
/cartoon vaudeville shepherd's crook-impelled exit stage left
posted by Drastic at 12:35 PM on November 4, 2019
"The (I think legitimate) question is whether that makes this an example of "artificial intelligence", or an example of "some neat-looking gunk spewed out of a neural network."
And the legitimate answer is that GANs work by playing a game of 'create new looking images' and 'detect the fakes.' Creation and discrimination can be seen as tasks of intelligence, especially considering that:
1. there's no deterministically "correct" answer to the creation task
2. In the final stages of training, the discrimination task is hard for us humans too
Of course, there is virtually zero 'transfer' of this skill to other skills, but I'm told that transfer learning in humans is surprisingly low as well. It's not AGI, but that is probably centuries - millenia away if ever. And I still maintain that AI is basically all problems without a known algorithm to solve them. As soon as we find a good algorithm, well, that's just a Hidden Markov chain, or a rules based inference engine, or a neural network spewing out gunk, not "intelligence," however nebulously defined.
posted by pwnguin at 12:49 PM on November 4, 2019 [1 favorite]
And the legitimate answer is that GANs work by playing a game of 'create new looking images' and 'detect the fakes.' Creation and discrimination can be seen as tasks of intelligence, especially considering that:
1. there's no deterministically "correct" answer to the creation task
2. In the final stages of training, the discrimination task is hard for us humans too
Of course, there is virtually zero 'transfer' of this skill to other skills, but I'm told that transfer learning in humans is surprisingly low as well. It's not AGI, but that is probably centuries - millenia away if ever. And I still maintain that AI is basically all problems without a known algorithm to solve them. As soon as we find a good algorithm, well, that's just a Hidden Markov chain, or a rules based inference engine, or a neural network spewing out gunk, not "intelligence," however nebulously defined.
posted by pwnguin at 12:49 PM on November 4, 2019 [1 favorite]
« Older Bound For Nowhere | Dave’s scandal was ahead of the curve Newer »
This thread has been archived and is closed to new comments
needs a PetePuma tag
posted by Greg_Ace at 9:28 PM on November 3, 2019