Artificial intelligence technologies have granted unconventional extensions to creative representations, which have long been the domain of humans. They have opened unprecedented horizons to overcome the limitations of human learning through intelligent models capable of instantaneous generation of new patterns of visual expression, where algorithms intertwine with human sensibility. However, this development simultaneously raises serious concerns regarding numerous ethical, legal, security, and social dimensions.
This debate has intensified recently with the announcement by the American company “OpenAI” of its ability to generate digital images that emulate the style of Studio Ghibli, the Japanese animation studio. This update received widespread attention among users who hurried to utilize it to embody themselves, their families, and their friends as cartoon characters in fantastical worlds. This extension even included embodying celebrities and public figures, leading the White House to publish a cartoon image generated artificially showing the arrest of an undocumented resident accused of drug trafficking. This popularity has given a momentum to the employment of generative AI models in creating visual content for communication in the virtual space, raising questions and concerns about the blurred lines between what is human and what is machine, and the implications of these practices on digital culture and social standards in the future.
Intellectual Property Dilemma:
Studio Ghibli was founded in Japan in 1985 by directors Isao Takahata and Hayao Miyazaki to produce animated films that developed a distinctive style, achieving wide fame and winning international awards, including an Academy Award; this contributed to the international spread of its unique style, linking the company’s productions to the visual memory of millions of Japanese anime enthusiasts worldwide. This trend of generating images using its distinctive style has led to a record increase in “ChatGPT” users, with one million new users in just one hour, as AI-generated images flooded social media networks.
Some have viewed the Ghibli trend as a tool to test the creative capabilities of artificial intelligence, promoting discussions about it across social media in a widespread manner; this enhances familiarity with those tools and their efficient use, in addition to presenting endless possibilities for artistic generation that blend Ghibli’s style with personal moments using smart technologies in a notable intersection of technology and creativity.
Conversely, these practices raise complex ethical and legal dilemmas related to authorship, copyright, and usage licensing issues. Uploading images to the intelligent platform for processing transfers ownership of the newly generated content to the platform owner, raising questions about the rights of the original image owner, as well as the rights of the artistic style used, as is the case with Studio Ghibli, whose unique artistic style was emulated. This situation parallels unintentional similarities between content generated through a series of mechanical sequences and other content produced through authentic human effort.
In the case of content inspired by Studio Ghibli, AI models have been trained on thousands of frames taken from films, posters, and artworks without obtaining a license, making it challenging to claim originality for these generative creations. This implies that offering them for sale, for example, may violate copyright rights, in addition to being subject to removal for mimicking legally protected artistic styles. This amounts to a gross waste of the effort of artists who spent years mastering their craft and developing a unique artistic style, condensed by AI in mere seconds. Moreover, relying on specific artistic styles that are repeated and recycled through recommendation algorithms may lead to erasing visual cultural diversity in favor of patterns perpetuated by AI models.
The dilemma does not only pertain to the ethical and artistic dimensions; it raises another legal issue concerning the ability to prove the use of original artistic production as input for training the AI model and demonstrating that OpenAI trained its large language models on original films and television shows owned by Ghibli, not on fan-made works shared online. While the works themselves are protected by law, the artistic style as a means of visual expression does not carry the same protection. These arguments highlight the challenges legal frameworks face in protecting creativity and intellectual property against generative artificial content.
Machine Learning and Data Collection:
The methods used to calculate the volume of data required to train machine learning models vary; some estimation strategies are based on a tenfold rule, implying that a machine learning model requires at least ten examples for each feature or predictive variable, which can increase significantly for deep neural networks containing hundreds of thousands or millions of parameters, as well as for natural language processing models that require billions of textual examples to grasp the vast linguistic variations. This means the size of training data varies according to the task required from the model and its complexity; whether the task involves classifying images, analyzing sentiments, making predictions, and other data accuracy-related parameters, including sources, quality, and potential biases.
In this context, social media platforms and AI tools have become not just digital spaces for content creation and communication but vast data mines, where every post, image, video, or interaction shared by users is collected, accurately transformed, and used to refine sophisticated AI models, enhancing their performance. They help customize services, improve recommendation systems, and develop virtual assistants and chatbots. Although social media content isn’t the only source of datasets used in machine learning, it is a significant and qualitative source of abundant inputs for various purposes and complexities, not only for using AI but also for other models. In South Korea, private and public data derived from the internet and social media constituted 10.2% of the data sources the public sector used to train AI models.
When a user grants an AI tool access to their images for required alterations, whether modifying or adding an artistic touch, or mixing them with unreal elements, the intelligent model does more than just analyze and modify images; it stores and analyzes them for continuous training, often including them in much larger datasets to bolster AI performance. Companies clearly announce this in their privacy policies and terms of use.
The same goes for social media networks, where “Meta” benefits from public posts to train chatbots and virtual assistants, and “LinkedIn” utilizes resumes and professional posts to improve job-matching algorithms. Meanwhile, the “X” platform shares user data with third parties to train AI unless users explicitly decide against sharing.
For instance, OpenAI’s privacy policies state that using its services implies permission for it to collect data from four main sources: first, the user themselves, who voluntarily provides data such as personal information and content pieces like audio, images, and files; second, data that are automatically received, such as log data and device location; along with data received from partners like marketing suppliers regarding potential customers; and finally, information from other sources, like publicly available data on the internet. The company also claims the right to retain, disclose, share, and use that data for various unspecified purposes to “enhance and develop services and conduct research,” which encompasses numerous areas of use, including model training.
Through carefully crafted privacy policies, major tech companies gain access to user content, analyzing it and including it in their ongoing training processes. While the user thinks they are interacting with a service, their data are recycled to enhance the machines’ performance and expand their capabilities—often condensed to general terms like “improving services,” allowing companies to own vast amounts of data without serious or genuine accountability.
Unexpected Uses:
Even though users share content on social media out of fun, curiosity, or social engagement, these interactions can be exploited in ways they hadn’t anticipated, often hiding ulterior motives different from the basic purpose of their sharing. This issue becomes pronounced when users are implicitly directed to contribute to these objectives without awareness or transparency; opening the door wide to misuse and unconscious participation in building systems capable of tracking, classifying, and controlling.
With OpenAI’s launch of the Ghibli trend and the potential benefit from it to enhance its algorithms or achieve greater profits (which caused a 6% increase in in-app purchase revenue), a prior trend introduced by Meta on Facebook, dubbed the “Ten-Year Challenge,” resurfaced. This challenge solicited users to share their current pictures alongside images from ten years ago. Multiple reports subsequently indicated that this was part of a project to train the company’s algorithms to recognize faces, estimate ages, and differentiate features across various races, colors, and genders, leveraging the vast geographic diversity of Facebook users worldwide.
This illustrates how some tech companies exploit trending phenomena for hidden purposes, such as data collection, or sharing that data with external parties, as exemplified by Amazon, which faced criticisms from the American Civil Liberties Union for selling facial recognition technology to the government, specifically to law enforcement agencies, like the police departments in Orlando and Washington County, Oregon.
Facial recognition technologies are among the most concerning applications reflecting a global issue given AI’s capabilities to identify and define objects, utilized by governments, major corporations, and hackers alike for surveillance, marketing, and security tracking. The scenarios of these unexpected applications increase when considering the risks of data leaks and breaches, as the vast amounts of data represent a valuable target for malicious activities, leading to potential sales or use in extortion, forgery, and attacks. When personal images are uploaded to AI systems for image generation, they become susceptible to unexpected uses that go beyond the user’s intent, such as being used to train AI systems capable of recognizing individuals or tracking them in real-world scenarios. Once an image is uploaded to the cloud, it enters a digital system accessible by multiple parties, intentional or otherwise.
In conclusion, the combination of generative AI capabilities and their popularity across social media platforms imposes challenges that necessitate crafting more specific and comprehensive data protection policies. These policies should ensure users understand how their data is collected, used, and stored while also securing it properly. The enjoyment, excitement, and humor accompanying social media trends veil a network of privacy risks, copyright issues, and ethical dilemmas that many users are unaware of, obligating technology companies to enhance transparency regarding data management practices and provide genuine alternatives for consent and data control. Especially given the prevalent ignorance among users about the terms of service they often agree to without sufficient awareness of the consequences in exchange for immediate access to seemingly free services. The severity of the matter compounds with the ongoing accumulation of vast amounts of data utilized in training AI models, making them tempting targets for hacking attacks.
