The AI art situation

[robscallon]

We are very lucky to have Andrew Huang everybody.

[OctaveIndustries]

The worst receiving end of all of this anyone who has continued working "by hand" in their own unique style and are wrongly accused of integrating or replacing their entire process with AI. I cannot imagine how someone who has spent their entire lives building up a traditional drawing or illustration skillset would feel in this situation.

[atalhlla]

I miss when AI in graphics just meant Adobe Illustrator.

[MagnitudeReviews]

As an artist (a photographer to be more specific), I am pro-AI tools, but anti-AI generation.

What I mean by this is for AI tool I include things such as AI sharpening, enlarging, denoising. I also include AI tools such as auto masking and object removal. This also includes AI text generation in the form of Alt-Text, which is a massive time sync to do. Or tools that can help cut down time when editing a multi-cut

What I am against is the creation of AI art. I am against using a text prompt to generate in a whale or a new sky or a clock, etc.

[hitherescotty]

I’m not an artist, I’m a musician. But my main takeaway from this video is that Andrew is a really insanely nice guy. What a good dude. He described the situation, presented what evidence he had, defended his decision in choosing an artist, apologized for how it offended some people, and turned this small instance into a deeper discussion about how the world interacts with art and media in general. Most people would have put up an obligatory twitlonger and moved on. This wasn’t even on my radar and yet Andrew did a 20+ min video about it. Kudos man. I like you even more now

[spencertilton]

Artist here. I'm not inherently against AI and I agree that it creates a lot of potential for significant quality of life improvements. AI that can color frames of a 2D animation or draw in between frames could be extremely beneficial and allow single individuals to undertake large scale animation projects.

The issue I have with AI is it's current implementation and intention behind it. Currently, AI image generation is designed to completely replace the artistic process. It's not about giving artists tools to empower them, it's about big businesses literally stealing billions of copyrighted images so they can sell their product that acts as a replacement to the artists that created said copyrighted images. The reason i believe that it is indeed copyright infringement is because it's copyright infringement to use samples of copyrighted music, even if you're not making money from it. You can take a sample and modify it and cut it up so it appears different, but it's still copyright infringement. Even if the original image isn't needed after the AI has 'learned' from it, comparing machine learning to how humans learn isn't a good analogy. When a human artist learns from another artist, there is essentially a 0% chance they would recreate a copy unless they copied the same thing many many times, at which point they'd essentially be training to create a copy and even so, if they did create a copy, it would be intentional. AI has no intention and has been known to overfit (create a replication of it's training data). The way AI learns is very different than the way humans learn. In my opinion, what's happening right now is art theft on an massive scale.

I do agree that there are good things about AI and I acknowledge that AI image generation will become very good eventually even with ethically sourced training data, but even so, it is my assumption that most artists will not use AI image generation. Simply because, painters want to paint. Writers want to write. Voice actors want to voice act. Even if AI can do these things for us, we do these things because we enjoy the process. I still often paint with physical paints even tho I can create higher quality illustrations with digital art programs, because I enjoy the experience of using the physical paints. I don't want someone to paint for me, I want to paint, myself. Current AI image generation is not like generating samples to use for your music production, it's like pressing a button and having the computer write a song for you. Wouldn't it always be better to make it yourself or just listen to music from your favorite artists, where every aspect of the song is intentional, rather than it just being an amalgamation of all the popular songs? The only use case i can think of for an artist to use AI image generation is to act as a sort of random image generator to find inspiration, or for photobashing like the artist you shared in the video.

[yakkocmn]

I appreciate you making this video and I respect the nuance in "two things can be true at the same time" and wanting to be optimistic about the future of tech, but I think where the conversation (and some of your points) get muddy is in the nebulous terminology of the word "AI" - which is a fault of both internet debates and aggressive corporate branding.

I think the productivity tools you mentioned for work like rotoscoping & subtitling can assist editors without taking control or agency away, and it's always exciting to see new software speed up a mundane task or assist with the process, but despite these tools sometimes needing to be "trained, " they are in a vastly different realm from the "generative AI" that people take issue with in the Book of Chances (and in general). As you mentioned with current court cases, it is a moral issue at the moment rather than a legal one, but when the core value of these tools in the eyes of their developers and corporate users is to train off of other peoples' works to eventually replace the human effort and touch, they are not tools that many artists are interested in engaging with or supporting. You're right that we've never seen anything like this before, but that's also why the conversation is so heated - many technological advancements you mentioned in music either aren't related to the actual "creation" (digital streaming services) or have made creativity more accessible... they don't cut artists out of the process. As a video editor, rotoscoping helps me make a cool scene in After Effects, but I have little interest in loading a script into one of the new "online AI editors" and having it spit out a finished product with an AI voice and amalgamated motion graphics. This isn't to say generative AI has zero possible positive uses, but with the way it is currently being implemented, it's hard to give it the benefit of the doubt.

"No ethical consumption under capitalism" also doesn't feel like a fair way to dismiss concerns about these tools. Just because products that have become integral to our daily lives are entangled with unethical practices doesn't mean concerns about new technologies should be ignored. And none of this has even delved into the environmental cost, which is... a separate can of worms.

[caoenqi]

Hey Andrew, longtime viewer here! I honestly don't know if you'll ever see this comment, but as someone who is pretty familiar with these machine learning/AI tools (I am a grad student studying robotics so these pop up a lot in my field of work), and is also starting to dip my toes into music production, I think I might be able to bring a unique perspective to things here. I have massive respect for the way you have handled the situation, and honestly I'm really writing this comment to excise these thoughts from my brain as I have been thinking about this issue a LOT.

There are a few key points that I think are being missed when it comes to ethical creation and usage of these tools:

1) With regards to copyright protection, saying that no images are actually stored within these models is really only "technically" correct; it's sort of analogous to the wine bricks that were being produced during the American Prohibition era (for context, during the period of time in US history where alcohol was completely banned, manufacturers would produce these bricks made out of dehydrated grapes, and include instructions that said "whatever you do, DON'T put this in a tub of water and allow it to ferment *wink* *wink*", thereby bypassing that restriction). Sure, there isn't any raw copyrighted material that is directly stored in these models, but it's almost trivial to get these models to produce such a thing.

Like you said, if someone uses it to replicate something that is protected, that is almost certainly infringement, but historically, who has the power to actually seek justice in these situations? Established people or those with studio backing may be fine, but anyone who is independent now has orders of magnitude more trouble making sure their work is safe. Not to mention, that's just from people who are doing it on purpose; there have been instances where a model spat out a more-or-less carbon copy of a still from a film despite only being given a generic description. So not only do creators now have to be 2-3x more vigilant in protecting their work, anyone using these tools has to be 100% SURE that the thing the model created is safe, which if it is trained on copyrighted material, is nigh impossible to guarantee; the tools are literally trained to emulate this material!

2) One point in particular that you mention is "if the artist couldn't find what he was looking for, he wouldn't have hired another artist, he would've changed his idea". I think that's a little reductive; they could have also licensed a stock image or used any of the massive free clipart libraries that artists have put together. Yes, they wouldn't have directly hired another artist, but the human beings who created the assets they're working with would still be properly compensated, whether directly or indirectly. The point is accessibility to creative tools and works for the purpose of transforming them IS IMPORTANT, and there are people who are working to make that happen while making sure the people keeping that creativity alive are not pushed out of the process! AI completely ignores that second part. Heck, the sample library Splice (which is a platform you yourself have endorsed) literally works off of this principle by compensating producers for the samples they create and then making them accessible to users!

3) You mention other tools that you yourself have used in the music creation process that indeed would have been "trained" on expert material, but the fundamental difference there is that those tools were created IN COOPERATION with producers whose end goal was to actually use these tools to make their jobs easier, while also still giving them control over the finer details if they wanted to make changes. Currently popular Generative AI tools were not designed with this in mind. Painters don't want a canvas that they can talk to and have it paint everything for them, they want to paint because that's the part where their creativity comes to life! And I'm not discounting the artist you hired either, they clearly are also very creative in their own right and they happen to work with the kind of medium (photobashing) that these tools can assist with. However, I don't think that justifies the existence of these tools; one, because the artist again has multiple other avenues to achieve that result (see point 2), and two, because the pushback against generative AI is due to a more holistic view of the technology.

There absolutely can exist cases where the tools are created ethically and used in a creative manner, but given the way these tools have largely been used by people to pump out low-effort content (see any content farm page on any social media) to straight up disinformation (remember that whole picture of the Pope in a puffy jacket?) and by corporations to not pay artists (see Nicki Minaj promoting her album with unmodified AI art), the concern is that the technology as a whole will lead to more negatives than positives given the way it is currently being developed. Hence, the pushback against these tools as a whole; sure, some people can play nice with it, but most people (especially those in power) probably won't. It's sort of like saying, "hey, it turns out nukes will be really good at terraforming Mars for construction!" Like, sure maybe, but everyone knows they weren't created for that reason.

Ultimately, the way I see it, AI tools as they are now disproportionately hurt smaller creators and while it would be nice to live in a world where these tools are ethically created and used, we don't live in that world and it's dangerous to push for acceptance of these tools when they aren't in a state to be acceptable.

Anyway, apologies for the long read. I hope I was at least able to contribute to the discussion and provide some points that maybe aren't thought about as much. Wish you all the best moving forward!

[tonywatson]

honestly super refreshing to hear a mature response to conflict on the internet. Its oddly soothing.

[schoclatesouce]

I think a lot of people commenting here either don't want to hear a nuanced take or didn't watch the full video. As an artist myself who is against AI I think this response was mature and reasonable. Thank you for addressing it honestly.

[samlilymusic]

Can't believe Andrew Huang just started a video with "Hey everyone" instead of the usual "Hey it's Andrew Huang *hwpshh*"

[seanandernacht]

I'm very sympathetic to the issue of art being trained on without the artist's consent. However I think the thing the copyright discussion always misses is that the real reason copyright exists is so artists can make a living under capitalism by not having someone else profit off their creation. If we had our basic needs met and opportunity for a comfortable life as an artist, we wouldn't actually need copyright anymore. Then we could treat art within the economy the same way we treat art in our minds: as expression and inspiration of the human experience to build upon, not some hoarded property to be withheld for a fee. Either way, I agree that artists should be able to opt out of training algorithms whether as things are now or in that idealized future

[SotonyaAcckaya]

I recently posted a link to an art piece of a girl with a bike and ppl started to throw accusations of using AI to draw a bike since the bike design was very unconventional. Thing is, bike was 100% drawn by hand, in some parts rather lazy hand) and if you have even basic understanding how a CG painting works you'd see that this was done by hand, yet ppl see what they perceive as an unconventional and somewhat artificial - and they think of AI and start this recreational outrage. I had by fair share of fun posting photo of a real bike of this design and explaining how painting strokes work, yet i felt somewhat sad that ppl cant even understand what is AI and what is just mistake, art style or even just an weird design.

[andresito]

Andrew, we appreciate you. You're always so level-headed and can discuss difficult topics with nuance. You know how to listen, how to be respectful, and how to communicate your own point of view clearly and eloquently. Thank you for everything you do. We need more of your qualities in our society.

[willstarbailey]

I just realized he sounds like the boss from smiling friends

[JadeWave]

I hope everyone sees the end bit at 19:54 before commenting. It's a genuinely virtuous action Andrew, very class, and it means a lot.

[MechMK]

The Silicon Valley philosophy of "It's easier to ask for forgiveness than for permission" is quite horrendous when it's companies with billions in capital taking from people living hand to mouth from their own art. It's not so much the art itself that bothers me (it all looks very same-y and uninspired), but rather the fact that these companies stole from small artists and re-sell their art now with zero royalties or other compensation. Even when artists explicitly say "You are not allowed to use my art to train AI models", companies don't care either. Why? Because artists can't prove that it happened.

[johnboldt]

This is a very mature response. Out of all of the ways this could go, I think this is the most surprising and also the most positive one. 10/10 confidence restored :D

[IronxIx]

AI relationship status: It's complicated.

[joshundrwd]

The only place where I think you fell short a bit here was in your description of the copyright question and the issue of creators not wanting to opt into the AI algorithms. You mentioned that you understand that it feels unfair and scary that a computer should be able to easily do things that humans work so hard at.

But I don't think this is the problem. The idea that machine learning would get better and better over time doesn't bother me at all. The problem is that it's not computers arbitrarily getting better at a thing. It's powerful corporations scraping the web for content and then having full say over what is done with it. The problem is that it's anti-democratic, which feels not just anti-artist but also anti-worker.

The companies creating these pieces of software are owned by capitalists, controlled by CEOs, and funded by VCs who want a return on their investment. The problem is that, as artists, we don't trust those people to have anything other than primarily their own interests at heart. They want to OWN as much as they can... and then they want to tell us that we're overreacting by saying we want to OWN our creations and get to decide what is done with them. Many of us do not see why CEOs and VCs and owners should get to have their cake and eat it too, seemingly without ever having to beg or scrape or ask permission, while artists neither get to have their cake nor eat it. It begins to feel like "heads they win, tails you lose."

The technology isn't just "scary" or "unfair" for no reason. This is about whether a tiny corporate elite who thinks it should get to own everything should get to dictate how the future will work for everyone else. In my view, this is a form of de facto governance, and these people should not get to rule without any consent from the governed.

All that said, overall this video was interesting and very productive, and I'm convinced you're approaching this from a good place. Cheers.