Reply to thread

Message: <blockquote data-quote="Abstruse" data-source="post: 9087616" data-attributes="member: 6669048">There are two main ethical concerns and two major practical concerns when it comes to the use of generative algorithms (aka "AI"):Ethical Concern #1: Generative algorithms harm the ability of writers and artists to earn a living through their work.Corporations in particular are using algorithmically generated content to do work previously done by human writers and artists. This is their clear stated goal in the use of these algorithms. In practice, the material produced by generative algorithms is not up to the standards required and likely never will be (see below under practical concerns). Writers and artists are then hired to "edit" or "touch up" the works at a rate far below their standard freelance commission rates - I have artist friends state they've been offered at most 25% of their normal commission rate to as low as 5% for this work. The issue is that it's not just a quick edit as there are fundamental flaws in technique and structure (both in graphical art and it writing) that makes it more work to "fix" than it would be to have just created it from scratch in the first place. This means that rather than making the work easier, generative algorithms are making more work for significantly lower pay.Ethical Concern #2: The databases used to "train" generative algorithms are almost exclusively sourced unethically and there are currently no ways to verify the source of works in those databases.To my knowledge, no software company making generative algorithms can claim that their database is entirely ethically sourced. The majority of them have scraped copyrighted works without the permission or even in many cases the knowledge of the creators of the work including results from Google Image Search, text from social media and blogs, online art portfolio/gallery sites like DeviantArt and Imgur, and fanfic archives like AO3. This includes all of the most popular ones like Midjourney, Canva, DALL-E, ChatGPT, and others. The closest so far is Adobe's attempt to create a database from licensed stock images and public domain art and photographs, but even that is considered unethical by many creators as the artwork license for the copyrighted stock images were often signed before generative algorithms existed and thus the artists were not compensated for their use in that manner (ie they were paid for the stock art/images as though they would be used in news articles, websites, blogs, videos, etc. and not be used to create imitations of their work and style via algorithmic generation).There are currently hearings by the European Union on algorithmic generation, and the biggest sticking point is that none of these companies can ensure that their algorithms are using images, art, and text that they have clear legal rights to use, that the creators in question have been compensated for their works' use in this manner, and that those creators can be properly attributed and credited. Many companies have stated that they will pull out of the EU market entirely if forced to disclose the source of their databases, either because of an unwillingness to disclose they have not ethically sourced their databases or because they literally can't verify that because they don't know where their algorithms are sourcing their databases from.The latter is a concern because the terms of service of many of these algorithmic generators state that anything entered in as a prompt is added to the databases to further "refine" the training of the algorithm. This applies to one of the alleged situations with the artwork for Bigby's Big Book of Big Bois where one artist's work was allegedly used as a prompt to "enhance" it via algorithmic generation without that artist's consent or knowledge. Meaning that artist's work may now part of the algorithm for the tools used by the other freelance artist who "enhanced" the work and, because these companies do not disclose where they source their database material, there is no way to prove this one way or the other. This endangers that artist's ability to make a living because their "style" is now part of the algorithm and can be "copied" without compensating them.Practical Concern #1: Generative algorithms are becoming worse, not better, over time due to their databases becoming corrupted by content made by other generative algorithms.The most obvious example of this is ChatGPT's text generation algorithm becoming corrupted as it is feeding off of content created by ChatGPT. Previously, ChatGPT's accuracy rate for simple math just a few months ago was 98%. <a href="https://finance.yahoo.com/news/over-just-few-months-chatgpt-232905189.html" target="_blank">It's now fallen to 2%</a>. The same database corruption is happening in graphical generative algorithms as well where the algorithm is training itself on material created by other algorithms or by that same algorithm, meaning it is reinforcing its own bad habits and mistakes. The joke online is that the generative algorithms have created their own Habsburg Jaw as the minor mistakes and issues become compounded as the algorithm "sees" more examples of it in the wild and thinks that's how it should be.For example, the "count the fingers, count the teeth" mantra for detecting algorithmically generated art was at one point getting better as the algorithms "figured out" that hands have four fingers and a thumb, each with only three joints on them including the knuckles. However, so much art was created where hands had six or eight fingers with two or four or more joints on them that those images are being fed back into the databases and the algorithms "think" that hands might have more fingers than that. So the algorithms are regressing to generating more images with too many fingers or too many teeth like previous.Aside: I use quotation marks around the "thinking" of the algorithm because it's not really thinking. It's just an algorithm, it can't think. For an analogy of how it actually works and how the algorithm is refined, <a href="https://www.tumblr.com/lumsel/712577333350023168/chinese-room-2" target="_blank">the Tumblr post "Chinese Room 2" is a good explanation</a>.Practical Concern #2: Content created primarily by algorithmic generation cannot be protected by copyright under US law.This doesn't apply in the situation with Wizards of the Coast, but it is relevant (and a few people online have brought it up in relation to this), so it should be addressed. Any art, text, or other creative work normally protected by copyright must have a human creator. This was established in law after PETA attempted to sue to allow a monkey to claim copyright on a selfie the monkey took of itself after stealing a photographer's camera and the court determined that neither the photographer nor the monkey owned the copyright because the work was not created by a human.This interpretation of copyright law has been upheld by the US Copyright Office, who denied a copyright to the artwork of a graphic novel that was algorithmically generated. Specifically, the entire graphic novel is protected by copyright because the artwork was placed sequentially by a human and the text was added by a human meaning the final work is the work of a human creative effort, but the individual art pieces are not protected and are in the public domain because they were not primarily the work of a human creative effort. It would be the same if I created a book made entirely of classical Renaissance art with funny captions written on them - I own the copyright to the book itself because of my creative use of juxtaposing images and text, but I would not own the copyright on the Mona Lisa.This doesn't apply in situations like the artwork confirmed to have algorithmic "enhancement" in Bigy Presents Glory of the Giants because the artwork in question was created by a human and just had additional elements added to it by algorithmic generation, and is treated the same way using a blur filter or color correction algorithm would be. However, there is not a strict line between how much of a work must be "hand-created" versus how much is algorithmically generated yet in legal precedent. And there likely won't be anytime soon because legislation lags so far behind technology. This creates issues around the use of generative algorithms because it's unclear if, in the situation I described above where a freelancer is hired to "edit" algorithmically generated work, that creative effort is enough to allow for copyright protection.These four issues must ALL be taken into consideration.All four of these concerns around algorithmic generation cannot be taken in a vacuum. This is one of the things that happens a lot in these discussions as supporters of algorithmic generation will focus on refuting claims about just one of them rather than taking them as a whole.And that's not even bringing in the other concerns about the use of generative algorithms such as the environmental impacts (the servers running the algorithms are getting bigger with far higher power requirements and other resource consumption), the impact on consumer computer hardware markets (some algorithmic generation companies are mass buying consumer PC graphics cards once again, which could become another arm's race similar to the height of the cryptocurrency/NFT bubble between consumers who just want to play video games and corporations who are buying the cards in bulk preventing them from ever reaching the market), the use of algorithmic generation to create "deep fake" style content for propaganda and harassment purposes (fake photos, videos, or audio of political figures, creating and releasing sexual content using the likenesses of people without their consent, etc.), and many others.I don't think a forum on tabletop roleplaying games is going to solve those problems, but they're all things to be considered when discussing the use of algorithmic generation.</blockquote>

Verification