Resolves to which ever option I agree with. I won't bet on this, so feel free to bring up any arguements you think will persuade me. Currently I'm kinda in the middle, so I'm not sure which I would choose right now.
Related questions
🏅 Top traders
# | Name | Total profit |
---|---|---|
1 | Ṁ30 | |
2 | Ṁ20 | |
3 | Ṁ8 | |
4 | Ṁ5 | |
5 | Ṁ1 |
If humanity is to survive and create a truly equitable society, we need to get out of this zero-sum mindset where automation is bad because it reduces scarcity and therefore some people's earning potential based on that scarcity. This is nothing new to art; these debates have been happening around automation of physical labor for decades. Image models are just bringing it to the people on Twitter who never cared about this until it was them personally facing a loss in profits and career opportunities.
In a just society, AI art would obviously be a good thing, as it increases the amount of beauty in the world and removes monetary barriers to accessing it. It remains to be seen whether actual humans will be able to set aside their capitalistic tendencies for long enough to rearrange society into a form that can accommodate the new lack of scarcity.
I think we are going to see AI generated art consolidate around a norm of training on public domain / creative commons art, and that will tampen down on most of the furor and see widespread adoption, and then we will see "taboo" models trained on just about anything (and especially NSFW material) consigned to the seedier corners of the internet.
-------------
My evidence:
The markets here think that there's going to be significant negative polarization around AI art in general:
https://manifold.markets/LarsDoucet/will-i-observe-significant-negative
HOWEVER, the markets are extremely certain we're going to see an all-public domain / creative commons AI art model trained within a year:
https://manifold.markets/LarsDoucet/will-an-ai-art-generator-be-trained
This eliminates the concern about training on artists' work without their permission.
How much does this matter?
I polled my own social media bubble (which has quite a few artists in it and a LOT of game developers).
Most of the artists were disapproving of AI art in general:
https://twitter.com/larsiusprime/status/1602661182535516162
And the reasoning they gave as their #1 concern was training on copyrighted works:
https://twitter.com/larsiusprime/status/1599487732224389121
But, for all those who were opposed to AI art, I asked if they would no longer be opposed to a model trained only on freely available images. The vast majority said they would now switch to being in favor of it:
https://twitter.com/larsiusprime/status/1599487055423471617
So I think we will see public "respectable" use shift to the public domain/creative commons trained (and largely SWF) models, and the hottest opposition goes away, and it becomes a tool in the toolbelt, and the unconstrained models become something that if you're outed as using you will probably lose friends.
@LarsDoucet I would expect new objections to arise to an AI trained on public domain images, if it was of similar quality.
@MartinRandall Yes I don’t think the debate would end. I do think the opposition would polarize slightly, because at least some people would drop their objections. But a bunch of people (at least half as a wild guess?) would still be mad.