Ryujin
Legend
Agreed. It's a thing that requires proof, not obfuscation of source. When I went looking for listed sources or methods, I typically found nothing.
Not every piece of art you don't like was made by AI
It would be nice, but I don’t see that happening without a push by either legislators or their customers.
The photography forums I follow tend to be gear focused, but they seem more concerned about AI regarding image generation, both in terms of correct compensation for the original photographers, and on the amount/kind of manipulation you can apply to a photo before it stops being a photo.
I don’t see the same kind of concern for AI post processing tools, so I don’t think there is a lot of pressure on companies to be more transparent, and a corresponding perceived market benefit.
This also got me thinking if it even makes sense to unethically train this kind of software, as it needs to work on the RAW files (i.e. the actual 14-bit RGB sensor output, which requires dedicated conversion algorithms before it can be displayed as an image) while images on the web are mostly shared in compressed formats such as 8-bit JPEGs.
Adapting to new tech isn't instantaneous.
The EU just passed AI regulations, including the need to label AI-generated images or video content. So, regulation is not off the table - and it doesn't have to be US regulation, either. The EU is to large for big tech companies to ignore, and any multinational company will have to comply with this for anything they to in the EU.
That often means they'll comply here in the US, too, because otherwise they run a risk of accidental violation if any work goes from US and gets reused in the EU.
Ryujin
Legend
Very true. The fact that we now need to accept cookies explicitly, for example, was driven by EU privacy laws.
trappedslider
Legend
Aren't they called biscuits there?
Ryujin
Legend
Well there are cookies, and there are biscuits, and they're rather particular about what is called what.
Yes and that's a great thing, although my understanding is that the provisions on copyright protections for existing models apply only to modifications and refinements of the model and, given these regulations will enter into effect three years from now, my fear is that in the short term will see a rush to plunder as much data as possible.
To give a bit of wider context, last year EU laws was amended so that the default position on allowing the use of materials for data mining was changed to opt-out rather than opt-in, and last Tuesday (the day before the EU regulation passed) our Prime Minister announced a new 1 billion-euro initiative to find "the Italian path to AI".
I guess we are looking forward to a few years of Wild West. On a longer term regulations will probably lead to better protections, and the dust will settle and we'll be able to assess the damage.
Similar Threads
