Boing Boing: On Monday, October 14, 2024, Adobe launched a public beta for Firefly, the company's new generative AI service, which can be used to generate text-to-video prompts, enhance and expand backgrounds in Photoshop, and more. So far, so standard fare for generative AI software.
5 comments:
It’s nice to know that Adobe is at least thinking about these kinds of things, and has an actual protocol, which they have made public, to show that they are. The fact that their AI is only being trained on material they have rights to is definitely a great start. I’m still not crazy about generative AI as a concept and I’m not sure that this kind of hard line is even possible with all the complicated regulations surrounding rights and fair use online, but something is better than nothing, since a lot of AI has been running rampant and being trained on just about anything they can get their hands on, which is unfair to artists. The article makes a good point though that this doesn’t necessarily absolve Adobe as a brand of any ethical concerns. I too am still a little mad that they’ve made their entire model based on renting licenses so you can’t just purchase any software.
After reading about Adobe's new generative AI service, Firefly, I find the company's approach to ethical standards refreshing and necessary. Launched on October 14, 2024, Firefly not only offers typical generative AI capabilities, such as text-to-video prompts and background enhancements, but it also emphasizes responsible training practices. By using licensed and public domain content for its dataset, Adobe aims to respect copyright while providing fair compensation to contributors of Adobe Stock.The transparency regarding content credentials, which serve as a digital "nutrition label," is a significant step towards authenticity in AI-generated work. This initiative aligns well with discussions in other artistic fields about the importance of provenance and credit. While I appreciate Adobe's forward-thinking stance, I also remain cautious, given their subscription-based model that limits ownership. Overall, Firefly sets a potential precedent for ethical AI in creative industries, but it will be interesting to see how this plays out in practice.
I’m really glad to see this sort of thing start to pop up more and more with different companies like Adobe. With how rampant AI content is it's really important that at the very least companies like Adobe start trying to control the AI they use before eventually legislation comes about limiting it hopefully in the near future. Training the AI purely on stock images is a great idea for the sake of copyright, so that it can’t be stealing and studying human made artwork online and attempting to replicate it, it can replicate old stock images. The fact that Adobe did it without any prompting is surprising to me. There was no copyright lawsuit about how the AI was being trained, there was no other legal motivation to do so aside from the moral ethics of it, and doing things for the morality of it isn’t something that I would expect a company as big as Adobe to do.
I've used Firefly for awhile while it’s been in photoshop and I really appreciate Adobe's efforts to make it ethical. I don’t use it for text-to-image stuff, but within Photoshop it can do some cool things. You can ask it to expand your frame, crop something out, add something in, etc. I think this is probably the best way to do AI. AI is a useful tool and there’s no way to avoid it as much as we may want to. These efforts to make it ethical are important and if every company did this, then I think AI wouldn’t be so controversial. The end is fair… Adobe’s subscription service is definitely annoying and it’s frustrating that if you want to use an ethical AI, you have to pay a crazy amount of money for it.
I have to say, this is the most responsible I’ve seen a company use AI thus far, aside from just not using it in the first place. Adobe does seem to be perfectly equipped to train an AI model since they’re an existing company with their own stock photos and existing community. What stood out to me was that user-generated content wasn’t going to be used as training for the AI. It’s obvious that Adobe is really making an effort to control their AI; they’re the only ones who can train their model and they’re being careful about it too. Content Credentials is a really interesting idea but I don’t think it will work. It’s going to be incredibly easy for people to find their way around it. I’m excited to see what it can do once it gets out of beta. It looks promising but given how these things usually go, I wouldn’t be surprised if it was misused or abused.
Post a Comment