{"metadata":{"kernelspec":{"name":"ir","display_name":"R","language":"R"},"language_info":{"name":"R","codemirror_mode":"r","pygments_lexer":"r","mimetype":"text/x-r-source","file_extension":".r","version":"4.0.5"}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"markdown","source":"# Copyright and Privacy Issues of Generative AI","metadata":{}},{"cell_type":"markdown","source":"Generative AI systems, such as ChatGPT, Dall-E, or Midjourney, can create texts and images from large amounts of web data. However, this raises legal issues for the protection of intellectual property rights. This also affects Kaggle as a platform for data science and machine learning competitions, where participants may use generative AI to create or enhance datasets and train and deploy models using (potentially) AI-generated content. Recently, whole competitions facilitate generate AI. But who owns the rights to the generated content or data? And who needs to give consent or get paid for their use? [5]\n\nThe current EU law is not clear on these questions, as it only protects works that are original and reflect the personality of human authors. The Court of Justice of the European Union (CJEU) [2] defines originality as the author’s own intellectual creation and free and creative choices. But this may not apply to works produced by AI systems, especially when human input is minimal or absent. Moreover, the EU law does not grant any rights to non-human authors, such as machines or animals [4,5].\n\nSome might say that using software to edit an image or write a text is also a form of machine-assisted creation. However, the difference is that the software is just a tool that supports the human activity of planning and executing the editing or writing. The outcome depends on the choices and skills of the human operator [5]. Therefore, this activity can be considered a personal intellectual creation of the human being, which is eligible for copyright protection.\n","metadata":{}},{"cell_type":"markdown","source":"\"creativity","metadata":{}},{"cell_type":"markdown","source":"## Creativity at the touch of a button\n\nGenerative AI works differently in creating things. The computer produces the outcome entirely by itself, with no human input. The software user acts more like a customer: He or she gives precise instructions to an artist on what to paint or write.\n\nHowever, an outcome that is made by AI automatically has no copyright protection. This is because the idea itself is not protected, only the specific way of expressing it. The same goes for the artistic or literary techniques used by an artist or author. But what about the AI creators? Can they claim copyrights for their machine’s works? The software code itself is protected, but not the things that come out of it. Other people who helped make the software, like the developer, also have no legal basis for this.\n\nGenerative AI poses a challenge not only to the legal system, but also to the concept of creativity itself [6]. How do we define creativity when machines can produce novel and useful content without human guidance? How do we evaluate the quality and originality of AI-generated works? How do we balance the benefits of generative AI for enhancing human creativity with the risks of misuse and abuse? These are some of the questions that need to be addressed as generative AI becomes more prevalent and powerful in various domains.","metadata":{}},{"cell_type":"markdown","source":"\"computer","metadata":{}},{"cell_type":"markdown","source":"## Who owns the rights to AI-created content?\n\nAI-made texts and images are usually not subject to the strict rules of copyright law. This has big implications: anyone can use the works that an AI creates artificially without any restrictions. That’s why some big players in the stock photo business are against the new technology and sue the providers. But others are looking for partnerships. For example, Shutterstock launched an AI-generator in collaboration with OpenAI and LG for realistic-looking photos, which can also make convincing faces. Some other examples of AI image and video generators are Kapwing and DeepAI. These tools allow users to explore their creativity and generate unique images with minimal effort [7].\n\nThe assessment that AI-generated content is not covered by copyright was also confirmed by the Copyright Review Board (CRB) of the US Copyright Office in early 2022. The owner of a \"Creativity Machine\" had applied several times before this authority to have the images he had generated with the help of AI entered in the copyright register there. The CRB came to the conclusion that the work \"A Recent Entrance to Paradise\" had been created by a computer algorithm independently. However, protection under U.S. copyright law always requires a creative human effort, a combination of \"human spirit and tangible creative expression\". Such a connection is not present in the AI image [1].\n\nThis outcome poses new legal challenges that have not been much considered by lawyers so far. For instance, can an AI user claim to be the author of images or text that are generated automatically? The media industry is not ready to deal with image material that has no protection. It could even be a crime if an author pretends that a computer-generated text is his or her own and asks for payments from a collecting society like VG Wort in Germany, ALCS in the UK, CISAC internationally, or ACUM in Israel.\n\nAI-generated content poses a challenge not only to the legal system, but also to the ethical and social implications of using such technology. How do we ensure that AI-generated content does not infringe on the rights and interests of original creators and users? How do we stop the use and abuse of AI-generated content for harmful purposes such as spreading misinformation, influencing people and deceiving them? How do we build trust and transparency in the creation and consumption of AI-generated content? These are some of the questions that need to be addressed as AI-generated content becomes more widespread and powerful in various domains such as health care, banking, media, and criminal justice [19].","metadata":{}},{"cell_type":"markdown","source":"\"AI","metadata":{}},{"cell_type":"markdown","source":"\n## The legal implications\n\nAI works that combine multiple sources to produce a new result may not be subject to copyright protection. This is because they are not considered original creations of human authors, but rather the output of algorithms that process data [8]. However, this clear-cut result only applies to AI works that do not copy large portions of text, images, or code from existing sources. If the algorithm merely reproduces existing content and this is found again in the output work, there may be a risk of infringing the rights of the original creators [9]. This was especially the case with early versions of AI programs. Users of AI-generated code snippets, for example, reported that they were too obviously derived from a clearly identifiable source [1]. However, these are likely to be “teething troubles” of the new technology: They will hardly appear in later versions. On the other hand, the possibility of uploading photos and having them edited by AI could become a problem for providers of image generators. Midjourney, for example, offers this and thus runs a legal risk. If the user has no rights to the uploaded image, legal conflicts can quickly arise, both in terms of copyright and data protection. Conversely, copyrights can be resurrected elsewhere: A well-designed and original prompt, which is intended to achieve a very specific result, can in turn enjoy the protection of the German Copyright Act (UrhG). Here a new professional group could emerge, which could be called \"prompt engineers\" and strictly protect their prompts as the basis of their profession.","metadata":{}},{"cell_type":"markdown","source":"\"human-machine","metadata":{}},{"cell_type":"markdown","source":"## Human-machine collaboration\n\nLegally, things get complicated when creations by human and machine are combined in one work. In the case of texts, this is the case when self-written text is mixed with generated passages or a AI excerpt is rewritten. Comparable problems are associated with the post-processing of AI images. As long as the changes are minor, the legal assessment should not change. Thus, it should not lead to a loss of copyright in the overall work if individual passages of a longer text are supplemented by a generated text. Conversely, it would not be sufficient to make only minor linguistic changes to a text by AI. For its own copyright protection, a complete transformation would have to take place, so that a new work of its own is created. If, on the other hand, a machine redesigns a text to a large extent, this can even lead to the fact that there is no longer any legal protection for the text thus created. Judges must then determine concretely word by word or pixel by pixel, to determine how much of the human or machine contribution is present in the final work.\n\nOne possible way to address this challenge is to adopt a collaborative approach between human and machine, where both parties contribute to the creative process in a meaningful way. This could result in works that are more original and expressive than those produced by either party alone. Some examples of such works include AI-assisted paintings, sculptures, and music compositions that combine human input and feedback with machine learning algorithms. These works could be seen as a new form of art that reflects the symbiotic relationship between humans and machines [10].","metadata":{}},{"cell_type":"markdown","source":"\"privacy","metadata":{}},{"cell_type":"markdown","source":"## The law and AI training\n\nThe legal situation of training AI with copyrighted content is more complex and controversial than using the output of AI. While old works that are no longer protected can be used freely, newer works that are still under protection require a legal justification. In the US, some AI providers rely on the “fair use” doctrine, which allows the use of protected material for educational and creative purposes [11]. However, this argument has not been tested in court yet, and some artists and photo agencies have already filed lawsuits against them [1]. They may face an uphill battle, though, as the “fair use” doctrine is generally interpreted broadly and in favor of innovation.\n\nIn Germany, a new law that came into effect in mid-2021 has made it easier to use publicly available works for text and data mining, which is essential for machine learning and artificial intelligence [12]. According to Section 44b UrhG, this use is allowed unless the rights holders explicitly opt out by adding a machine-readable notice on their website, such as in the metatags. However, this raises some questions, such as what happens if a rights holder changes their mind after their work has been used to train an AI. Apart from that, though, the new law opens up new possibilities for using copyrighted material for training artificial intelligence.\n\nHowever, the legal situation may not be the same for all types and amounts of content used for training AI. For example, using snippets of text or images may be more acceptable than using entire works or collections. Moreover, the legal situation may also depend on the purpose and outcome of training AI. For example, using protected content for research or education may be more acceptable than using it for commercial or malicious purposes. Finally, the legal situation may change over time as new laws and court decisions emerge. For example, the EU has proposed a new regulation on artificial intelligence that could affect the use of protected content for training AI [13]. Therefore, users and providers of AI should be aware of the current and future legal implications of their actions.\n\nA way out may be to adopt a collaborative approach between human and machine, where both parties contribute to the training process in a meaningful way. This could result in more diverse and original content that reflects the human values and preferences of the trainers. Some examples of such approaches include interactive learning, where humans provide feedback and guidance to the machine, and co-creative learning, where humans and machines jointly explore and generate new content [13]. These approaches could also foster more transparency and accountability in the use of AI systems.","metadata":{}},{"cell_type":"markdown","source":"\n\"privacy","metadata":{}},{"cell_type":"markdown","source":"## Privacy issues\n\nAnother area that concerns lawyers in connection with AI generators is data protection. This primarily concerns image generators such as Dall-E or Stable Diffusion. These are compulsorily trained with images that are freely available on the Internet. To ensure that the programs can display everything, the developers of the underlying algorithms use a large number of such images in order to obtain, as far as possible, a cross-section of the image material available on the Internet.\n\nAmong the images found in this way are, of course, many photos of real people. If the person depicted is clearly identifiable, such images constitute personal data within the meaning of the General Data Protection Regulation (GDPR) [14]. This is because data protection applies to all information that clearly identifies a person. In addition, every person in Germany has the right to decide whether and how their image may be used. It is legally irrelevant whether the image is a photograph, a drawing or a rendering. The so-called right to one’s own image applies. It is derived from the right to privacy, which is enshrined in the German constitution. Exceptions apply only to people who are known to the public, such as celebrities.\n\nThis raises the question of whether those affected are even aware that their images are being used. The law provides for information obligations that are intended to prevent precisely this [14]. However, the GDPR recognizes exceptions to the rule that data subjects must be informed about the processing of data concerning them. Thus, the data processor may waive this requirement if “the provision of this information proves impossible or would require a disproportionate effort” [14]. One can conclude, with good arguments, that it is not even possible to inform all data subjects in the case of mass image capture. This is because image capture is automated - the providers of Al-supported image generators are unlikely even to know themselves exactly which personal images their Al system uses for training.\n\nIn the future, privacy-by-design principles need to be integrated when developing and using Al generators, which means adding data protection features into every stage of the process . This could involve actions such as hiding or blurring the faces of individuals in images, asking for clear and informed consent from data subjects before using their images, applying security measures to prevent unauthorized access or misuse of personal data, and deleting or destroying personal data when it is no longer needed.","metadata":{}},{"cell_type":"markdown","source":"\"Fakes","metadata":{}},{"cell_type":"markdown","source":"## Dealing with fakes\n\nFake images of people that are created with the help of AI-supported image generators are explosive in terms of data protection law. This is because they are inaccurate data within the meaning of the GDPR. Such fakes are highly likely to discredit individuals and cause significant damage. It is precisely such damage that the legislator wanted to prevent when it created the GDPR. For this reason, the GDPR provides a legal basis for the deletion of such images [16].\n\nIn addition, the disseminator may even be liable to prosecution or damages if the images are clearly misused or even used in a sexual context. Some services such as Dall-E 2 therefore prevent images of real people from being created from the outset. Providers such as Midjourney, however, facilitate use that is hardly compliant with data protection by allowing images to be uploaded. To set clear limits to such use, the providers have built in various measures to prevent misuse. For example, they use filters that generally prevent the display of sexualized content, especially with children. Depictions of violence or images of weapons are also prevented, and the software has an extensive list of words that must not be used. However, it should hardly be possible for the provider to exclude all possibilities of abuse and violations of data protection.\n\nThis raises the question of whether those affected are even aware that their images are being used. The law provides for information obligations that are intended to prevent precisely this. However, the GDPR recognizes exceptions to the rule that data subjects must be informed about the processing of data concerning them. Thus, the data processor may waive this requirement if “the provision of this information proves impossible or would require a disproportionate effort” [17]. One can conclude, with good arguments, that it is not even possible to inform all data subjects in the case of mass image capture. This is because image capture is automated - the providers of Al-supported image generators are unlikely even to know themselves exactly which personal images their Al system uses for training.\n\nThe use of AI to make fake images of people is risky and dangerous for data protection law. It also raises ethical and social questions, like how these images affect our society, our democracy, and our dignity. The law may not be ready to deal with these questions, because it depends on things like consent, identity, and accuracy that may not make sense or work for AI-generated images. We may need new ways to control the use of AI to make fake images, like codes of conduct, certificates, or watchdogs. We also need to be responsible and ethical when we use AI to make fake images, and when we see them online. We need to respect the rights and interests of others, and we need to check the truth and source of the images we see. We also need to be careful when we share our own images online, because they may be used by AI without us knowing or agreeing.\n\nFake data can also affect Kaggle. Datasets used in competitions may contain fake images of people or items that are created by AI or uploaded by users. This can affect the quality and validity of the data, the fairness and accuracy of the models, and the privacy and rights of the people whose images are used. Therefore, Kaggle participants need to be aware of the issues and challenges of using AI-generated fake images of people in their competitions, and they need to follow the best practices and guidelines for data protection and ethics.","metadata":{}},{"cell_type":"markdown","source":"## Conclusion\n\nAI can create amazing images and texts, but it also raises legal and ethical questions. Who owns the rights to AI-generated works? How can we protect our privacy and identity from being used or manipulated by AI? How can we tell what is real and what is fake online? These are some of the challenges that artists, lawyers, and users face in the age of AI. We need new laws and rules to regulate AI, and we also need to be responsible and ethical when we use it.\n\nAI is also changing the world of Kaggle competitions, where people use machine learning to solve problems. Generative AI is a new trend that can create realistic data, such as images, texts, or sounds. This can be useful for augmenting or synthesizing data sets, or for generating new solutions or insights. However, it can also pose risks, such as cheating, plagiarism, or bias. Kaggle participants should be aware of these issues and follow the guidelines and ethics of the platform. They should also respect the originality and quality of the work they submit. Generative AI can be a powerful tool for Kaggle competitions, but it should be used wisely and fairly.","metadata":{}},{"cell_type":"markdown","source":"\n[0] Andreas Guadamuz. Artificial intelligence and copyright. WIPO Magazine\n\n[1] Appel G., Neelbauer J., Schweidel D.A., Generative AI Has an Intellectual Property Problem, Harvard Business Review, April 07, 2023.\nhttps://hbr.org/2023/04/generative-ai-has-an-intellectual-property-problem\n\n[2]: EU proposes new copyright rules for generative AI | Reuters\nhttps://www.reuters.com/technology/eu-lawmakers-committee-reaches-deal-artificial-intelligence-act-2023-04-27/\n\n[3]: Generative Artificial Intelligence and Copyright Law\nhttps://www.venable.com/insights/publications/2023/02/generative-ai-copyright-overview-part-1\n\n[4]: Generative AI – Copyright Overview Part 1 | Insights | Venable LLP\nhttps://www.venable.com/insights/publications/2023/02/generative-ai-copyright-overview-part-1\n\n[5]: AI and Copyright Law: What We Know | Built In\nhttps://crsreports.congress.gov/product/pdf/LSB/LSB10922\n\n[6] : How Generative AI Can Augment Human Creativity - Harvard Business Review\nhttps://hbr.org/2023/04/how-generative-ai-could-disrupt-creative-work\n\n[7] Shutterstock Launches New AI-Powered Image Generator | Digital Trends\nhttps://www.legal500.com/developments/thought-leadership/legal-issues-with-ai-generated-content-copyright-and-chatgpt/\n\n[8] Hugenholtz P.B., Quintais J.P. (2021) Copyright and Artificial Creation: Does EU Copyright Law Protect AI-Assisted Output?. IIC - International Review of Intellectual Property and Competition Law 52: 1190–1216.\n\n[9] Kretschmer M., Meletti B., Porangaba L.H. (2022) Artificial intelligence and intellectual property: copyright and patents—a response by the CREATe Centre to the UK Intellectual Property Office’s open consultation. Journal of Intellectual Property Law & Practice 17(3): 321–326.\n\n[10] Reben A. (2022) The weird and wonderful art created when AI and humans unite. BBC Future.\nhttps://www.bbc.com/future/article/20221123-the-weird-and-wonderful-art-created-when-ai-and-humans-unite\n\n[11] Fair Use – Wikipedia \nhttps://en.wikipedia.org/wiki/Fair_use\n\n[12] Text und Data-Mining – juristische Fallstricke und bibliothekarische Handlungsfelder : Text & Data Mining - European Alliance for Research Excellence\n\n[13] When AI can make art – what does it mean for creativity?\nhttps://ogc.harvard.edu/pages/copyright-and-fair-use\n\n[14] EU data protection rules – European Commission\nhttps://commission.europa.eu/law/law-topic/data-protection/eu-data-protection-rules_en\n\n\n[15] 5 things you should know about GDPR for images\nhttps://www.fotoware.com/blog/5-things-you-should-know-about-gdpr-for-images\n\n[16] Privacy law and resolving 'deepfakes' online\nhttps://iapp.org/news/a/privacy-law-and-resolving-deepfakes-online/\n\n[17] Guidance note on the capturing of images and videos in relation to data protection law\nhttps://www.ucl.ac.uk/legal-services/ucl-general-data-protection-regulation-gdpr/guidance-notices-ucl-staff/guidance-note-capturing\n\n[18] Image-Generating AI: Trends and Legal Challenges\nhttps://www.hklaw.com/en/insights/publications/2022/12/imagegenerating-ai-trends-and-legal-challenges\n\n[19] Ethical concerns mount as AI takes bigger decision-making role in more industries.\nhttps://news.harvard.edu/gazette/story/2020/10/ethical-concerns-mount-as-ai-takes-bigger-decision-making-role/","metadata":{}},{"cell_type":"markdown","source":"Disclaimer: I used DALLE to create the images and used Deepl Write to improve grammar and language of the text.","metadata":{}},{"cell_type":"code","source":"suppressMessages(library(tidyverse))\nsub <- tibble(type = c('essay_category', 'essay_url', 'feedback1_url', 'feedback2_url', 'feedback3_url'),\n value = c('Generative AI',\n 'https://www.kaggle.com/code/christph/copyright-and-privacy-issues-of-generative-ai/',\n 'https://www.kaggle.com/code/sumamallapragada/textdata-and-ai/comments#2323952',\n 'https://www.kaggle.com/code/hassanabdullahsaleem/the-evolution-of-ai-ethics-balancing-the-promise/comments#2323963',\n 'https://www.kaggle.com/code/lannister/kaggle-ai-report-generative-ai-in-computer-vision/comments#2324034')) \nwrite_csv(sub, file = \"submission.csv\")","metadata":{"execution":{"iopub.status.busy":"2023-07-03T08:05:19.060552Z","iopub.execute_input":"2023-07-03T08:05:19.063677Z","iopub.status.idle":"2023-07-03T08:05:20.182868Z"},"trusted":true},"execution_count":null,"outputs":[]}]}