I’m not worried about it. The pace at which models improved in 2022-2023 has slowed considerably. There’s simply not enough training data left for these models to continue to improve at the rate we once saw. People keep saying “eventually it will be impossible to tell” but I think that’s assuming too much.
People also said decades ago, "eventually we'll have flying cars." We can't know what tech will do in the future as much as we're tempted to think we can.
I truely hope we can always tell, I’ll be pretty bummed if it’s perfect. I’d be excited for flying cars except for the fact that the airspace is regulated and we’d just end up in air traffic congestion that’s no different to current road traffic 😂
I work in the industry as well and the mostly we use AI for is for preproduction. It still has a long way to go when it comes to using it as main footage even for the VFX. Most of the time you are at the wim of what it gives you and not in creative control no matter how much you train it. There are a lot of issues as well when it comes to stabilization of the objects within the shots themselves. You can subconsciously sense that there's something off and always get an eerie feeling about it. So I believe it would be used for certain things such as commercials and online content, but when it comes to long format it is too far away in my opinion.
The point you make that the simple fact of knowing that something is AI generated will impact on how that thing is received is more important than is commonly appreciated I think. The current view seems to be that the 'uncanny valley' problem is about a failure to achieve perfect realism- which is true. But beyond that I beleive lies a second 'uncanny valley' problem which is the potential cognitivie dissonance created by the conflict between being presented with a perfectly realistic fake human that you know is not a real person. If the point of a movie is to be manipulated into some kind of emotional response- which is why we go to see movies-then a movie populated by fake AI generated people presents an interesting dynamic. If such a thing emotes on the screen how do we respond to it? Do we allow it 'in'- do we empathise with it and implicitly therefore grant it the status of being human? Or does our intellectual knowledge that the 'person' on the screen is entirely fabricated derail this empathic response? Or-more likely- do we occilate between responding to it as a person and seeing it as an artifact? Also how comfortable are we going to be to be seen emotionaly responding to a fake person? The thing about real people is that they exert a kind of moral gravity- there are rules about how you respond to real people, especialy in pubic situations. For example-as a thought experiment- imagine you are in a public gathering and some new prototype humanoid robot offers to shake you hand, introducing itself as 'Robbie'. Do you shake it's hand? If you do then you are placing yourself in the odd posiiton of granting human status to machine- but if you don't then the thing is just human enough to place you in the position of appearing rude and graceless. While this may seem a trivial issue I think the dillema it presents will become increasingly problematic as fake humans begin to propagate through the culture. Anything that seems sufficiantly human in appearance and action will exert a moral pressure to be treated as human- yet it will not actually be human. This generates a 'dobuble bind' sceanrio in which there is no correct response- the resulting psychic tension will create a new form of the 'uncanny valley' effect that I think will make it hard to integrate fake humans into society. The notion that AI generated content is entirely fungible with human made content ignores this problem and simply assumes that if the fake people are realistic enough they will simply be assimilated without issue. Time will tell I guess- but if I am right then a lot of money is being expended developing a technology that might never really work- we may find that we are not willing to engage with things that look and sound exactly like us precisely because of this similarity- their simple presence on our screens or in our spaces may just be too disturbing.
I had to read that a few times to take it all in. The part about shaking a robots hand is so fascinating, I really hadn’t considered that before. It will be very interesting to see which valley we all end up in! Thanks heaps for taking the time to watch!
I’m not worried about it. The pace at which models improved in 2022-2023 has slowed considerably. There’s simply not enough training data left for these models to continue to improve at the rate we once saw. People keep saying “eventually it will be impossible to tell” but I think that’s assuming too much.
People also said decades ago, "eventually we'll have flying cars." We can't know what tech will do in the future as much as we're tempted to think we can.
I truely hope we can always tell, I’ll be pretty bummed if it’s perfect. I’d be excited for flying cars except for the fact that the airspace is regulated and we’d just end up in air traffic congestion that’s no different to current road traffic 😂
I work in the industry as well and the mostly we use AI for is for preproduction. It still has a long way to go when it comes to using it as main footage even for the VFX. Most of the time you are at the wim of what it gives you and not in creative control no matter how much you train it. There are a lot of issues as well when it comes to stabilization of the objects within the shots themselves. You can subconsciously sense that there's something off and always get an eerie feeling about it. So I believe it would be used for certain things such as commercials and online content, but when it comes to long format it is too far away in my opinion.
Yeah that actually a great point, there isn’t much choice when trying to refine it atm. Just is what it is. Thanks heaps!
The point you make that the simple fact of knowing that something is AI generated will impact on how that thing is received is more important than is commonly appreciated I think.
The current view seems to be that the 'uncanny valley' problem is about a failure to achieve perfect realism- which is true. But beyond that I beleive lies a second 'uncanny valley' problem which is the potential cognitivie dissonance created by the conflict between being presented with a perfectly realistic fake human that you know is not a real person.
If the point of a movie is to be manipulated into some kind of emotional response- which is why we go to see movies-then a movie populated by fake AI generated people presents an interesting dynamic. If such a thing emotes on the screen how do we respond to it? Do we allow it 'in'- do we empathise with it and implicitly therefore grant it the status of being human? Or does our intellectual knowledge that the 'person' on the screen is entirely fabricated derail this empathic response?
Or-more likely- do we occilate between responding to it as a person and seeing it as an artifact? Also how comfortable are we going to be to be seen emotionaly responding to a fake person?
The thing about real people is that they exert a kind of moral gravity- there are rules about how you respond to real people, especialy in pubic situations. For example-as a thought experiment- imagine you are in a public gathering and some new prototype humanoid robot offers to shake you hand, introducing itself as 'Robbie'. Do you shake it's hand? If you do then you are placing yourself in the odd posiiton of granting human status to machine- but if you don't then the thing is just human enough to place you in the position of appearing rude and graceless.
While this may seem a trivial issue I think the dillema it presents will become increasingly problematic as fake humans begin to propagate through the culture. Anything that seems sufficiantly human in appearance and action will exert a moral pressure to be treated as human- yet it will not actually be human. This generates a 'dobuble bind' sceanrio in which there is no correct response- the resulting psychic tension will create a new form of the 'uncanny valley' effect that I think will make it hard to integrate fake humans into society.
The notion that AI generated content is entirely fungible with human made content ignores this problem and simply assumes that if the fake people are realistic enough they will simply be assimilated without issue. Time will tell I guess- but if I am right then a lot of money is being expended developing a technology that might never really work- we may find that we are not willing to engage with things that look and sound exactly like us precisely because of this similarity- their simple presence on our screens or in our spaces may just be too disturbing.
I had to read that a few times to take it all in. The part about shaking a robots hand is so fascinating, I really hadn’t considered that before. It will be very interesting to see which valley we all end up in! Thanks heaps for taking the time to watch!