Shulou(Shulou.com)11/24 Report--

The influence of AI on human career and the waves of public opinion caused by it are unimaginable. Remember that in 2017, AI was a standard new thing, and one of our core tasks was to explain to our readers that simultaneous interpretation, comic book writers, hosts and other human work could not be replaced by AI in the short term. This year, however, a discussion about the crisis of human painters caused by AI painting has been staged on social media at home and abroad.

DALL-E 2, Stable Diffusion, Midjourney, DreamBooth AI, Wombo Dream, Make-A-Video, Novel AI. These AI painting tools that generate images from text appear in large numbers and become more and more popular. More and more paintings generated by AI are on the market, constantly refreshing the public's understanding of the boundaries of AI capabilities, a large number of disputes around AI painting and worries about AI replacing human painters are gushing out.

At the time of the upheaval of AIGC, the legal and ethical issues about AI painting have become more and more important. where is this man-machine war going? We try to explain the beginning and end of the whole incident in one article.

Qi: the technology that caused public anger and the art paintings generated by AI have been around for a long time. In 2017, AI painting became the guest of honor in Christie's, Sotheby's and other high-end auction venues, with a high price of millions of dollars. apart from causing some analysis and saddening by people in the industry, the public opinion field has not caused any waves.

However, the condemnation of AI's painting has reached a peak in social media in recent months.

For one thing, a large number of painters announced a boycott of AI. On August 29, an AI painting website called mimic launched a beta version, allowing users to upload 15 to 100 images for AI to learn, and then output AI paintings with the same style. However, some unauthorized paintings have also been uploaded for study, selling paintings generated by AI as their own creations. Then there was the emergence of someone uploading the works of the recently deceased painter who was still under copyright protection to AI for learning, which led to the dissatisfaction of a large number of original authors, thus boycotting "AI learning".

Second, the crisis of public opinion affects almost all mainstream AI drawing tools. The mimic incident directly aroused the dissatisfaction of a large number of painters and readers against AI. Some people deleted their public paintings, some people began to resort to legal provisions, and at a time, disputes over data copyright and technical ethics also began to spread to well-known AI generation tools such as DALL-E 2, Midjourney and Stable Diffusion. It has been found that some popular tools such as NovelAI and the data source sites used for training and learning also upload unauthorized pictures.

According to a statement posted on the website Danbooru, generators like NovelAI are trained on billions of images on thousands of sites, including artist sites such as Pixiv, Twitter, DeviantArt and Tumblr, as well as sites such as Reddit and Pinterest, so removing works from the Danbooru platform will not stop AI from continuing to use artists' works on other sites. In other words, as long as it is open on the Internet, it may become learning data. Apart from asking the developer of the model, there is no other way to prevent his work from being studied by AI. Unless the painter deletes all his paintings, it is difficult to find out whether they have been used for learning one by one.

Third, there is a lot of controversy over the use of AI to paint. With the lowering of the development threshold, a large number of individual AI developers use open source tools, such as some people use AI to generate illegal content such as yhsq. On September 6, the "Space Opera House" generated by AI drawing tool Midjourney won the first prize in the digital category at the Colorado Exposition Art Competition, which also caused a lot of controversy, because the author's work is to enter description text, AI will create the painting on demand, and netizens think that it is "witnessing the death of art" that "meaningless and soulless". Artist Genel Jumalon said more bluntly. It's a damn thing to win the first prize in the field of art with an AI painting.

("Space Opera House") fear that AI will replace human painters and regard AI painting as a disparagement of human ability has become a contagious emotion.

As mentioned earlier, the generation of painting by AI is nothing new, so why has it suddenly caused a splash among the public recently, causing public anger? Of course, people are not crushed by the last straw.

Undertake: the road of AI painting in the volume, it is necessary to have a brief understanding of how AI painting has grown from the minority art concerned by auction houses and collectors to the mainstream standard of mass creation in a few years.

Computer vision has always been one of the main tasks of deep learning, including target recognition, target tracking, image segmentation, graphics processing and so on. Specific to the field of image generation, around 2015, AI art creation is mainly based on convolution neural network transfer learning to convert image style, first semantic segmentation of image content, and then the content and scene through line bending, style transfer and other means, converted into the specified art style, similar to the "filter" function of beauty software. Obviously, the artistic value of this method of generation is not high, and people who can click on the p-map do not think of themselves as great artists, and often appear very creepy works, and the pictures processed by Google's deep learning painting system DeepDream are very weird.

AI generation really began to show artistic value and comparable human level, it began with the popularity of the generation against the network GAN (Generative Adversarial Nets) in 2016, the principle of the GAN model is to make the generator network and discriminator network against each other, thus creating new images with higher realism and accuracy. During this period, a variety of XXGAN image generators appeared, giving birth to a large number of works of art. In October 2018, Christie's sold a portrait of Edmund Bellamy by AI for $432500, making it the first AI art auction in human history.

However, GAN can not get rid of the problem of the traditional AI depth model: it cannot understand "logic" and "common sense". For example, AI can stack elements according to text keywords, but it often draws very "Cesuru" anti-common sense works because it cannot understand the logical relationship behind the natural language. The change comes from the rise of pre-training models.

The large model trained through large-scale data and violent computing shows strong robustness. It not only performs well in the field of machine vision, but also iterates out the ability of cross-modal generation, which promotes the generation of AI from language to vision. This wave of "text-to-image" painting tools can produce fake and real paintings, and the "brainpower" behind them is generally supported by the basic technology of large models. Google, OpenAI and Baidu, which launch technical competitions on AI drawing tools, are all leaders in large model technology and promoters of AIGC.

(KREA one-click generation of Andy Warhol-style "cyber punk brain polar body") it is not difficult to understand the technical heritage of AI painting, it is not difficult to understand why the struggle for AI creation has been fermented at this moment.

First of all, there are so many giants that the technological progress is beyond imagination. AI painting brings together a large number of technology giants "bright muscle", including Google, OpenAI, Meta, Microsoft, Baidu, Tencent and other AI capabilities, making the technology by leaps and bounds, the number of tools at an unprecedented rate of explosive growth, AI painting (image generation) ability has also reached an astonishing level. A year after the launch of DALL-E in January 2021, OpenAI launched the latest DALL ·E 2, which has a four-fold increase in resolution and can create realistic images from natural language descriptions. Google has rolled out a number of AI drawing tools, including Imagen, Parti and so on, while the AI drawing platform launched by Microsoft is named NUWA Nuwa, which shows that it is ambitious.

The excellent generation effect of the large model makes AI painting have certain practicability and commercial potential, including text generation illustration, original sample display of creative work, automatic completion of repeated mechanical work, and so on. AIGC naturally faces the suspicion of traditional practitioners while moving towards the industry.

(DALL-E 's painting with "horseback riding, astronaut" as the key word.) secondly, the active open source of large models has lowered the threshold for AI painting again and again.

This round of mainstream AI drawing tools have chosen open source to prosper the development ecology of the big model, reduce the threshold for the use of technology, and anyone can use these tools to create, some even for free. For example, StableDIffusion created by British startup Stability AI is completely open. OpenAI launched Dall-E 2 and offered a free trial on September 28, and NovelAI opened on October 3, which attracted a large number of users to use it.

Compared with the previous AIGC as a niche art and collection investment object, the large model-based AIGC allows more people to participate in training, development and use. Although various open source communities have clearly published regulations on intellectual property rights, including not to use adult content, hate or violent images, and to avoid the use of copyrighted materials. However, with the increase in the number of users and the lowering of the threshold, violations will inevitably occur once someone does not understand and abide by the intellectual property norms and protocols of the open source community.

(content policy of DALL E 2) in addition, the legal gap and lag make the original creators feel powerless to protect their rights.

It is difficult to avoid AI abuse by relying on the constraints of the open source community, so can we rely on special laws and regulations on digital works to protect the intellectual property rights of originators? The current situation is obviously disappointing, legislation, as a very serious matter, often has a certain lag. Compared with the rapid development of AI technology, as to whether the works generated by AI have intellectual property rights / copyright, and how to deal with illegal images generated by AI, the relevant laws and regulations are still blank and discussed all over the world. In 2019, Stephen Taylor (Stephen Thaler) hoped to name his DABUS artificial intelligence system "inventor" and obtain patents for related inventions, but it was rejected by courts in the United States, Britain, Europe, Australia, Germany and other places. At present, it seems that on a global scale, the intellectual property system pays more attention to human creation, and using AI and clicking "go" is not considered creative. For example, Shenzhen Nanshan Court has ruled that in a case, articles generated by AI-assisted collaboration system are protected by copyright law, and unauthorized reproduction and dissemination need to bear the corresponding civil liability.

Intellectual property rights are difficult to judge and safeguard, and the "absence" of legal protection of AI generated content aggravates the difficulty of safeguarding rights, and it is difficult to restrict illegal use, which makes human original painters face AI painting tools, and their intellectual property rights are also in a "vacuum" zone.

Technology, culture, ethics and law and other factors intertwined, constitute the beginning of the AI painting disputes and the origin of the storm, this is an interlinked story.

Turn and merge: the Creator's upheaval and New Machine so does the controversy mean that AI painting has come to a standstill? It's not!

In fact, after frequent discussion, a large number of original painters have a certain attempt and understanding of AI painting tools. From the feedback of some well-known painters, there are generally three attitudes towards AIGC applications such as AI painting.

It is too early for 1.AI to replace the human painter. After testing, a blogger found that "there are some pictures that can be seen, but there are basically none that match the description intention."AI painting is actually similar to the filling function of ps in the eyes of a graphic decorator." The reason is that there is still room for improvement in the content quality of AIGC. On the one hand, there is still a big gap between the natural language understanding ability of AI model and human beings, and the text generated by GPT-3 is equivalent to the level of primary school students. in this case, it is difficult to make AI drawing tools understand complex text and accurately express content ideas in images. On the other hand, in business, the art market praises highly unique works, and scarcity has always been an important pricing standard for works of art. Industrialized and mass-produced things are of no value to collectors. With the open source of AI painting tools, unique handmade products may become the art that people pursue. In fact, this trend has already appeared, some netizens think that the future contribution is likely to be "purely manual" and "zero AI" will become a selling point instead.

One creative blogger said bluntly: painters who boast that their pictures are hand-drawn may need to add extra value to their works if they want to be rich in the future-this way of life is already in operation, that is, artists.

So at least so far, AI painting tools can provide a lot of help, but still can not replace human painters.

2.AI painting is really useful. It should be noted that art creation and art collection is a relatively minority activity, and it is a very small group of people who can become artists. However, there is a lot of visual and design work in daily production and life. In these areas, AI has been able to play a very useful "drawing assistant" role, has become the trend of the times.

One is to reduce repetitive / risky work. For example, in visual creativity, communicating with customers takes a lot of time and energy, and there is a lot of uncertainty. Customers often say "it's better to have the first version" after painting, and AI can act as a painstaking party B to quickly generate a variety of AI creative pictures, avoiding heavy upfront investment and even rework. In addition, every holiday promotion activities, artists will inevitably encounter a large number of repetitive and boring work, such as making highly homogenous posters, can also be done by a high level of AI ability. Therefore, some designers named DALL E "stupid Party A Terminator".

The other is AI as an auxiliary tool to improve the work efficiency of designers. Many AI-based drawing tools have been developed as plug-ins for Figma, Photoshop, Blender and other tools to help creators render details, color with one click, and provide inspiration. For ordinary people who have not been trained in systematic painting, the threshold for artistic creation has also been greatly reduced. Some netizens used Midjourney to synthesize photos of their deceased grandmother in flowers, and some netizens used it to make illustrations for their own fan novels. Before the emergence of large models and AIGC, they could only be entrusted to professional painters after a long wait to complete, but now drawing with AI can be completed with one click.

Ordinary people can realize their brains locally at a low cost or even 100%. The freedom of creation supported by the AI model has just begun, so some creators say bluntly that "AI painting is for the benefit of mankind."

3.AI intellectual property protection needs to be accelerated. For creators, AI painting tools are undoubtedly an excellent assistant, at the same time, to avoid the barbaric growth of AIGC, because of some illegal painting operations into controversy and negative, but also become art circles, AI circles, legal circles together to discuss, speed up the legislation of the urgent task.

At present, the intellectual property debate of AI painting mainly focuses on three aspects: 1. Data copyright. AI model training requires the number and coverage of sample data, how to ensure the source of data copyright and protect it; 2. Creative copyright. Data is only the first step of infringement. "painting style plagiarism" is a very distressing infringement of painters. It is more difficult to judge whether or not plagiarism is generated by AI than to judge human plagiarism. Protection of interests. At present, a large number of AI drawing tools provide paid services, but once the AI generation technology is commercial, there is no effective solution and protection means for people as data sources or creators to get reasonable returns and benefits. So DALL-E 2 directly stipulates that its users cannot sell any work of art created with it. However, if there is a measure to accurately judge the contribution and distribute the benefits to contributors, such as federal learning technology, blockchain technology and so on, I believe many originators will be happy to participate in the industrialization process of AIGC.

With the protection of law and technology, AI painting can accelerate human-computer cooperation within the boundary, rather than separation and conflict.

In a word, the evolution of modern civilization is the disenchantment process of a world. The process of industrialization is to replace the mysteries of skilled craftsmen with the systematic application of science and technology such as steam engine and electrification, and this disenchantment process has also brought about mass production, high efficiency and low cost. the life of the general public is more convenient and affluent than at any time in the traditional era.

In this sense, AIGC is not a process of artistic disenchantment. As the creator of Midjourney said: beautiful stones come from rivers, but rivers are not creators. This AI system does not have the ability to create, but beauty can come from it. (Every beautiful stone comes from the river, but is the river creative? No, I don't think so. Is the system creative? No. Can beauty come out from it? Yes.)

Nothing can destroy the soul and creativity, just as photography does not disintegrate Van Gogh and Monet. In the era of intelligence, embracing AI has become a foregone conclusion. In this process, how to find the value and meaning of people will be a common issue for all of us.

This article comes from the official account of Wechat: brain polar body (ID:unity007), author: Tibetan fox

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.