Should EA avoid using AI art for non-research purposes?
Treating âagreeâ as âyesâ
I think the strongest reason against (using AI art for non-research purposes) for me is the idea of all the uncompensated art that made up the data. Itâs a bit of an original sin for AI in general (including text generation), and not one which weâve found a good response to.
Reasons for (using AI art for non-research purposes): - It makes sense for EAs to adopt a place in the memetic space where we think that AI is and will be very powerful (and therefore itâs important to learn how to use it), and itâs likely to be very dangerous. I donât think there is a contradiction there, and avoiding the use of AI would be increasingly hobbling. This is relevant because I donât think we can make a clean distinction between AI generated art and AI generated textâboth are likely built on an amount of stolen/â un-compensated data. - Using AI images (like bulby above) is just a bit of fun, i.e. the scale of use is pretty smallâthis correlates with this not being a very big deal.
- A lot of the concerns are hypothetical comms concerns, Iâd take this more seriously if things played out that way, but right now Iâd guess that the anti-AI-use camp is fairly loud but not strategically useful. And since I disagree with them for the other reasons above, Iâd rather not pretend that I do for optics reasons.
Overall: Iâm definitely open to changing my mind on this. I especially donât feel like I have a principled response to the un-compensated labour that went into creating AI, and itâd be great to have one.
Treating âagreeâ as âyesâ
I think the strongest reason against (using AI art for non-research purposes) for me is the idea of all the uncompensated art that made up the data. Itâs a bit of an original sin for AI in general (including text generation), and not one which weâve found a good response to.
Reasons for (using AI art for non-research purposes):
- It makes sense for EAs to adopt a place in the memetic space where we think that AI is and will be very powerful (and therefore itâs important to learn how to use it), and itâs likely to be very dangerous. I donât think there is a contradiction there, and avoiding the use of AI would be increasingly hobbling. This is relevant because I donât think we can make a clean distinction between AI generated art and AI generated textâboth are likely built on an amount of stolen/â un-compensated data.
- Using AI images (like bulby above) is just a bit of fun, i.e. the scale of use is pretty smallâthis correlates with this not being a very big deal.
- A lot of the concerns are hypothetical comms concerns, Iâd take this more seriously if things played out that way, but right now Iâd guess that the anti-AI-use camp is fairly loud but not strategically useful. And since I disagree with them for the other reasons above, Iâd rather not pretend that I do for optics reasons.
Overall: Iâm definitely open to changing my mind on this. I especially donât feel like I have a principled response to the un-compensated labour that went into creating AI, and itâd be great to have one.