What do you remember about the technology hotspots of the past year? Have you ever been attracted by some of them, and even got into “quarrels” with your colleagues?
At the beginning of the new year, Guoke has prepared a few short stories for you, all about the technological phenomena that have aroused heated discussions in the whole society in the past year, and the arguments surrounding them. We join you as we review these stories and look forward to more tech fun in the new year.
Technology has more than two sides, and the attitude of each of us is precious, and “quarreling” makes it even better.
$120 vs 2 cents
Each of us is approaching the moment when we will be snatched alive by machines. For Zhang Wei, this moment has come.
Since graduating from university, Zhang Wei has been an independent painter for eight years. During the National Day in 2022, he was introduced to receive a job with a picture – to draw the head of the character for the novel.
This is a “volume” job, which does not require high technology and creativity. The price offered by the other party is also reasonable, 120 yuan per piece, 65 pieces in total. The main cost is communication. Because it was the first time to cooperate, the two parties spent a day matching requirements. During this period, Zhang Wei drew three drafts, and the other party found some cases for his reference.
Everything seemed to be progressing smoothly. After the communication, he handed in the first draft and passed it. The other party also paid the manuscript fee for the first painting. Until a few days later, he suddenly received a notification:
“It was decided at the meeting that the illustrations of the novel will be changed from illustrations of characters to illustrations of items, using AI to draw, and the cooperation will be interrupted.”
“I was snatched alive by AI.” He was a little dumbfounded. He asked the other party to see the results generated by AI, and the other party generously sent it to him. “Pretty good,” he felt. But the bigger gap is the cost: the other party told Zhang Wei that it only costs 2 cents to generate a picture by a machine.
Painter to be one of the jobs most threatened by AI in 2022｜Pixabay
Give the AI a hand
That’s a difference of nearly 860 times.
The managers of the company were moved first – which company started to cut art, which company started to study AI technology, and which company was already using AI to generate pictures. Similar rumors were circulating more and more frequently among the painters.
Liu Xiaoli’s game company is one of them. At the end of October 2022, a colleague from the marketing department held an AI drawing sharing meeting, and used the finished picture of the game project to generate relevant pictures on the spot.
Participants included Liu Xiaoli, an art post, as well as marketers, planners, and programmers. Most people reacted flatly, after all, they have seen cases on the Internet more or less in the past six months. Liu Xiaoli’s feelings are not amazing, “there is a lot of uncertainty, it is impossible to generate a good picture at one time.”
Only the producer was very excited, and asked the main artist happily: Can some skins in the game be made with AI in the future?
Previously, some work such as skin painting would be outsourced, and the fee for each manuscript was about five or six thousand yuan, which required an artist to paint for a week. Now using AI, it only takes two hours to generate a picture, and then it can be completed in one and a half hours after the original artist makes changes. It takes only four hours to produce a picture.
Character skins are hugely attractive to mobile gamers｜”Tomorrow’s Ark” official website
Although the boss has made a decision, adding AI to the workflow is still in the testing stage, and Liu Xiaoli’s work has not been affected too much.
In the face of many detailed and common Party A requirements, the randomness of AI is not dominant. It can’t understand some simple instructions, such as, “put your feet back”-maybe that’s why, after the publisher chose AI to paint, they changed the illustrations of the novel from portraits to objects.
But before the real threat came, people’s mentality changed first: a colleague in the same position as Liu Xiaoli felt a sense of crisis after the AI sharing session, and “here came up.”
“Before, my colleagues basically didn’t take out outsourced work, and spent weekends with their children. But in the past week, someone came to her, and she started working immediately without saying a word.”
Turning around and returning to the company, this colleague was not too happy to “help” AI, and “pushed his mouth” when revising the manuscript.
Crypko, an AI drawing platform, sells two-dimensional character portraits｜Screenshot of Crypko’s official website
feed the machine
It is now understood that machines did not learn to draw in a vacuum. Before this, the well-known AlphaGo has learned 160,000 human chess records to improve its “chess skills”; and a face recognition algorithm needs to be trained to pass, and the number of face pictures generated by commonly used data sets often reaches the order of millions .
The same goes for AI mapping. The hot 2022 DALL-E 2, Stable Diffusion, etc., their training sets contain billions of parameters. Unless the style is directly specified (eg, Van Gogh, Monet), it is not easy to tell which artists a machine-generated map has “learned”.
But soon there were exceptions. At the end of August 2022, the AI drawing tool mimic once caused heated discussions among Japanese painters. This drawing tool can imitate the style of a designated cartoonist and output similar works. Japanese illustrators began to protest in groups, and the “use prohibited” tag became a hot topic on Twitter. Many illustrators said that their works would not be authorized to mimic for training and use.
mimic uses AI to imitate the Twitter artist in the official promotion｜mimic Official Twitter
Immediately afterwards, the situation spread to the circle of Chinese painters. Independent painter Hun Jun said: At the beginning, people just started using these tools with the mentality of “playing and watching”; then, some people who did not draw appeared to use the works of specific painters to train models, and then used the generated works in turn. Make fun of the artist.
On October 13, 2022, Soul Jun posted a Weibo, calling on the industry to pay attention to the phenomenon of AI painting infringement. He mentioned: At present, “AI is not authorized to use the artist’s paintings as nourishment” and “unauthorized to use the artist’s drawings to wash manuscripts” is gradually rampant. He hopes that more colleagues can join in the voice and resist the infringement in the name of AI drawing. The protests were directed at a painting tool called NovalAI.
On November 29th, Weibo netizen @Sueno罗晓 voiced his voice to boycott Nijijourney, the latest AI painting program based on Midjourney.
These controversial tools have one thing in common, they are all born for the creation of two-dimensional paintings. No matter what keywords you enter, the finished product will be in an anime style.
Compared with the super-large model used by DALL-E 2 and Stable Diffusion, the material pool of the second dimension is much smaller. When the input requirements are detailed to specific characters and traits, the “shadow” of a specific artist often emerges.
The resisted mimic and Noval AI platforms are both regarded as “washing” machines because they can produce works that are extremely close to a specific artist’s style-this behavior is so rampant that the artist’s original work, Sometimes it is identified as “AI-made” instead.
Image source: Screenshot of Weibo
New technology amplifies an old problem: plagiarism.
There is an unwritten habit in the circle of painters: Fan lovers will collect their favorite pictures on the Internet and upload them to some public platforms; while fledgling painters use this as a way to promote themselves and communicate with their peers.
In order to facilitate the search, they often put very detailed tags on the pictures. Take the following fanart of the game as an example. The uploader will not only mark the character name, the source of the game, but also specify the character’s hairstyle, hair color, posture and Clothing and other details.
The orange box on the left is the tag of this picture ｜ danbooru
As a result, these platforms have become natural AI material libraries. As long as the crawler technology is used to capture pictures and entries, they can be directly used for training, saving even the cost of manual labeling. Unconsciously, painters have also contributed to the progress of AI.
When using pictures as material for training, most AI platforms will not seek the consent of the original author. The profits from the commercialization of the platform will not go into the pockets of the original authors—some painters believe that this has actually constituted infringement.
ArtStation Artists Collective Against AI Drawing｜Independent artist Joyce Da Silva
Human Strikes Back
As a traditional role of Party B, it is difficult to say to what extent the attitude of the artist can influence Party A’s choice, let alone change the rut of technological advancement. But they have not since surrendered.
Mr. Hun put forward an idea in his proposal: In the future, when painters publish their works on the platform, they can cover a large area of watermark on the works, destroying the integrity of the works, so as to prevent them from being used to train AI and wash manuscripts. Some painters responded and began to share watermark resource packs. Unlike small signature-like watermarks, these watermarks are huge and try to cover the entire frame evenly.
The artist began to share self-made watermarks publicly and freely｜Screenshot of Weibo
Artists are also forced to prove that their work was not done by AI.
While uploading their works to social platforms, they began to actively post drafts, videos, etc. of the drawing process. Some netizens even said that this practice has become a new “posting etiquette”; Works to prevent being used for AI training.
In September 2022, an artist group called Spawning launched the website Have I Been Trained? (Have I been used for training?), helping artists understand themselves by searching the largest open source image databases LAION-5B and Laion-400M Whether the work is used for the training of AI tools.
Some traditional picture platforms have also shown their attitudes. Image library sites such as Getty and Shutterstock have successively deleted some images on the platform that were clearly marked as AI-generated; Fur Affinity, a community of Furui (orc) enthusiasts, has prohibited AI works from appearing on the platform on the grounds of protecting human creators.
The trend of AI prohibition is spreading on various platforms｜Screenshot of Weibo
Of course, some people use legal means to fight back. On November 3, 2022, Microsoft received a class action lawsuit in which its AI programming assistant GitHub Copilot was accused of using code released by users in public warehouses for training, which violated the legal rights of a large number of creators. The “plaintiffs” behind the lawsuit are “millions of GitHub users,” the open letter said.
This lawsuit has attracted a lot of attention, because not only Copilot, but most of the current artificial intelligence generation tools, including AI mapping, are behind the same set of operating logic. Matthew Butterick, a lawyer involved in the lawsuit, said in an interview with the media: We are in the “Napster era” of artificial intelligence – in the early days of the emergence of digital music, the judgment of the “Napster case” laid the foundation for the development of copyrighted music and determined that people are now in the “Napster era”. The habit of listening to music on the platform rather than private downloads.
In the past year, the tools for generating images from text, music and even video from text have been refurbished and surpassed each other; while creators have spent a year settling down on a problem: how to protect their rights and interests in the face of technology?
Rick&Morty creators use AI graphics in new game to “create a sense of absurdity in parallel universes” ｜ Reddit
In the face of increasingly omnipotent AI, the human community has also begun to split.
Hunjun has added a lot of chat groups discussing AI drawing, there are groups formed solely by painters, and there are also groups with more complex members including technical researchers, company bosses, etc. He observed that those who study technology and are bosses are usually more excited and optimistic when talking about AI.
As for the painters, Mr. Soul concluded: “There are pessimists who think that AI will replace humans sooner or later; there are optimists who think that AI is just a tool and think that some new jobs may be generated around AI in the future; There was an extreme technologist who had no brains to support AI. He thought that human learning is not an infringement, so AI learning is not considered infringement. As a result, many people hated it. There are also painters who think that the emergence of AI is a good thing, and it can eliminate the so-called ‘Low-end’ painter.”
More and more signs show that this is not a short-lived quarrel, but a core proposition that cannot be avoided on the road to the popularization of AIGC technology.
Under AI painting-related content, people always spontaneously push the discussion about infringement to the top; Stable Diffusion announced in December 2022 that it will cooperate with the HaveIBeenTrained website to allow artists to search and delete their own works in the training set.
The well-known up master of station B deleted the video using AI painting｜Screenshot of station B
Soul Jun himself has tried to use AI to make things. As early as 2017 to 2018, he came into contact with the “automatic coloring” tool that was considered AI at the time, which was quite crude and unobtrusive. It wasn’t until the emergence of Disco Diffusion at the end of 2021 that he noticed AI drawing again-the energy of technology is no longer what it used to be.
After hesitating again and again, he still chose not to use it: “I’m still worried about copyright infringement.”
Zhang Wei, who was robbed of his illustration business by AI, feels more directly: AI is a large plagiarism machine. He believes that the current level of pursuit of AI will soon lead to varying degrees of unemployment for painters, and AI will eventually become an existence above humans.
Liu Xiaoli, an artist of the game company, believes that her competitors are not AI from the beginning: “The opponent is the post-95 generation, those younger new original painters, (I) already have a sense of crisis, what is AI?”
What about plagiarism in AI? Liu Xiaoli immediately asked back straightforwardly: “Don’t people copy?”
(Zhang Wei and Liu Xiaoli are pseudonyms in the article)
references
[1] https://ift.tt/faGbFgN
[2] https://ift.tt/KyebQ5n
[3] https://ift.tt/I8WuCqm
[4] https://ift.tt/8ir9G7V
[5] https://ift.tt/kHGuXfh
[6] https://ift.tt/Bi4uN7D
[7] https://ift.tt/cCtkZq3
Author: Guo Hengyu
Editor: Weng Yang
This article comes from Guoke, and shall not be reproduced without authorization.
If necessary, please contact [email protected]
This article is transferred from: http://www.guokr.com/article/463343/
This site is only for collection, and the copyright belongs to the original author.