Context

Recently, a new bill has been introduced in the US Congress aims to force AI companies to disclose their use of copyrighted material in training their generative AI models. 

Generative AI Copyright Infringements Around the World 

• An AI service provider in China recently has been found liable for copyright infringement and ordered to pay damages over those generative AI images.
• The New York Times, the world’s largest newspaper by number of subscribers, is suing OpenAI and Microsoft for infringing the copyright of articles. 
• Universal Music Group, the largest record company, is suing Anthropic for using its song lyrics without permission. 
• Getty, one of the biggest image libraries, is suing Stability AI for copying its images (as well as misusing its trademark). 

Enroll now for UPSC Online Course

Copyright Infringement Challenges in Generative AI Content

• Copyright Infringement: 

  • Using copyrighted content and material without permission can be considered copyright infringement, even if for training purposes.
    • It includes copyrighted material such as images, texts and music.

• Used for Training: 

  • AIs are trained on vast quantities of human-made work, from novels to photos and songs. It requires large amounts of data to train their algorithms effectively.
    • These training data are broken down into “tokens” (numerical representations of bits of text, image or sound) and the model learns by trial and error how tokens are normally combined. 
    • Following a prompt from a user, a trained model can then make creations of its own. More and better training data means better outputs.

• Arguments in Favour for Use of Copyrighted Content:

  • Applicability of Fair Use Doctrine: Fair use as a doctrine goes back to 1841, to a case about copying the writings of George Washington. It is a US concept that permits limited unauthorized use of copyrighted content. 
  • Balancing Factors: Fair Use involves balancing four factors:
    • Purpose and character of use
    • Nature of copyrighted content
    • Amount used
    • Impact on market value of original
  • Transformative Use: Transformative uses (meaning it adds something new or different to the original work) more likely to be fair use.
    • OpenAI argues training is transformative and does not replace New York Times.

• Arguments in Against for Use of Copyrighted Content:

  • Legality & Ethical Implications: AI technologies can be used to replicate or mimic existing copyrighted works. The algorithms can analyze and generate content that closely resembles protected works, raising questions about the legality and ethical implications of such replication.
    • Many AI companies have become reluctant to provide information about what data their models are trained on, citing competitive confidentiality. 
      • Example: OpenAI’s past disclosures show that its GPT-3 model was trained on sources including the Common Crawl, a scraping of the open internet which includes masses of copyrighted data.
  • Lack of Control: Copyright holders might lose control over their works if they are freely used for training AI models.
  • Lead to Inequality: Large tech companies with access to vast datasets benefit more from this practice, potentially creating unfair competition.
  • Copyrights & Copywrongs: While the lawyers sharpen their arguments, deals are being done. In some cases, suing is being used as leverage and lawsuits are negotiated by other means. 
    • Even once trained, AIs need ongoing access to human-made content to stay up-to-date, and some rights-holders have done deals to keep them supplied with fresh material. 
    • Most rights-holders are privately pessimistic. A survey of media executives in 56 countries by the Reuters Institute found that 48% expected there to be “very little” money from AI licensing deals. 

• Regulation in India On AI Copy Infringement:

• • The Indian Copyright Act, 1957: It grants exclusive rights to creators (authors, artists, etc.) over their original works for a specific duration. 
    • Using a copyrighted work without the owner’s permission, constitutes copyright infringement. 
  • Copyright on Generated AI: In India, there is no specific litigation in the context of text and data mining by AI. 
    • In the Copyright Act of 1957 in India, there is no way a non-human can be granted copyright protection. 
  • The Patents Act, 1970: It provides specific provisions for fair dealing and enumerated exceptions to copyright infringement.
    • It grants patent rights to discoveries that cover innovative and innovative processes, products, or articles of manufacturing.
    • It also doesn’t have provision with respect to copyright laws to safeguard any creation that is wholly generated by AI.
  • Digital Personal Data Protection (DPDP) Act, 2023: It plays a pivotal role in addressing data protection concerns. However, it exhibits certain gaps, such as legitimizing data scraping by AI companies when data is publicly available.

About Generative AI (Artificial Intelligence)

• It is a type of AI technology that can produce various types of content, including text, imagery, audio and synthetic data.
• It utilizes deep learning, neural networks, and machine learning techniques to enable computers to produce content that closely resembles human-created output autonomously. 
  • Examples: ChatGPT, DALL-E and Bard.

Enroll now for UPSC Online Classes

Significance of Generative AI

• Creativity and Novelty: 

  • It enables the creation of new and unique content, whether it’s images, music, or text.
    • Example: Automated Journalism: Generative AI can produce written content for news outlets, summarizing events, financial reports, and sports matches quickly and accurately, freeing human journalists to focus on in-depth stories and analysis.

• Automation and Efficiency: 

  • It automates the process of content creation, saving time and resources.
    • Example: Graphic Design: AI can create logos, marketing materials, and other graphics, enabling designers to rapidly prototype ideas and iterate on concepts.
    • Architecture and Engineering: AI can generate building designs and urban plans, considering optimal use of space, environmental impact, and aesthetic appeal.

• Personalization and Customization: 

  • Generative models can be trained on specific data or preferences, allowing for personalized recommendations, tailored content, and customized user experiences.
    • Example: Generative AI can create customized educational content, adapting to the learning pace and style of individual students, making education more accessible and effective.

• Exploration and Inspiration: 

  • Generative AI can provide inspiration to artists, designers, and writers by generating diverse variations, exploring creative possibilities, and serving as a starting point for further creative exploration.
    • Example: OpenAI, which makes perhaps the most advanced generative AI models, is valued at nearly $90bn; Microsoft, its partner, has become the world’s most valuable company, with a market capitalization of $3.2trn.

• Examples of Generative AI

• • ChatGPT, made by OpenAI, can be used for recreation of long newspaper articles that it appears to have memorized.
    • Claude, a chatbot made by Anthropic, can be made to repeat lyrics from well-known songs. 
    • Stable Diffusion, made by Stability AI, reproduces features of others’ images, including the watermark of Getty, on whose archive it was trained.

Concerns with Using Generative AI

• Ethical Concerns: 

  • Generative AI raises ethical concerns, particularly regarding the misuse of synthetic media, deep fakes, and potential infringement of intellectual property rights.
    • Example: In November 2023, a deep fake video of actress Rashmika Mandanna was circulated widely on social media platforms. 

• Dataset Bias and Generalization: 

  • Generative models heavily rely on the training data they are exposed to. If the training data is biased or limited, the generated outputs would also be biased.
    • This can lead to discrimination and reinforce existing societal biases, if such an input data is provided to the AI system.

• Computational Resources and Complexity: 

  • Training and deploying generative models can be computationally intensive and require significant resources, including high-performance hardware and substantial training times.
    • Implementing and maintaining these models can be complex and resource-demanding.

Enroll now for UPSC Online Course

• Quality and Coherence: 

  • While generative models have made significant progress, they may still struggle with producing outputs that consistently exhibit high quality, coherence, and contextual relevance.

• Job Augmentation: 

• • Most jobs and industries are only partially exposed to automation and are more likely to be complemented rather than substituted by AI.
    • Generative AI systems similar to GPT (Generative Pre-trained Transformer) are more likely to become productivity tools, supporting and speeding up the execution of some tasks within certain occupations.

• Fears Over Job Destruction: 

  • The surge in generative AI and its chatbot applications has sparked concerns about job destruction, akin to the concerns surrounding the introduction of the moving assembly line in the 1950s.
    • Technology can enhance job quality in the workplace by automating routine tasks, allowing more engaging work, but also limiting worker agency or increasing work intensity.

• Gendered Impact: 

  • Effects of automation are “highly gendered”, with more than double the share of women potentially affected by automation, due to their overrepresentation in clerical work, especially in high- and middle-income countries.
  • The recent increase in women’s labor market participation may be threatened by concentrated job losses in female-dominated occupations.

• Digital Divide: 

• • Generative AI technology is dependent on access and cost of broadband connectivity, as well as electricity. 
    • In 2022, one-third of the global population (around 2.7 billion people), still did not have access to the internet.

Way Forward to Using Generative AI Content

• All Stakeholder Approach: 

  • Consultation and negotiation is a need between various associated stakeholders to come out of this copyright issue.
    • Market-based Solutions: There is a need for a market-based solution, similar to the music industry’s response to peer-to-peer file sharing.
    • Regulated Policy: It should emphasize the importance of fine-tuning policies to promote creativity while addressing concerns about ownership in AI-generated content.
    • Clear Guidelines: Clear guidelines on AI use in copyright applications is desirable and required to ensure transparency.
    • Time for Judicial Recognition: It is a must for AI training as ‘research’ under fair dealing.

• Need to Expand Fair Use Approach: 

• • The use of copyrighted materials for training purposes should generally be considered fair use. 
    • Need for Specific exceptions for text/data mining: It should be ensured that if anyone is using copyrighted material for training, they do not seek copyright protection for the content generated by the AI concerned. 
    • Follow Four-factor Test: The four-factor test laid down by the Kerala High Court in the case of Civic Chandran vs C. Ammini Amma, 1996 can be useful in determining a considered fair use. These factors are the purpose of the use, the nature of copyrighted work, the amount and substantiality of the portion used and the impact of the use on the value of the copyrighted work.

• Regulation & Alignment with Existing Laws: 

  • There is a need to update intellectual property laws to align with the advancements in AI technology.
    • For oversight and compliance purposes, implement the data usage and governance policies for AI projects.
    • There is a need to mandate AI companies to appoint compliance officers responsible for copyright protection, audits, and assessments.

Enroll now for UPSC Online Classes

• Maintain a Balance: 

• • The rising issue of copyright infringement and AI can have an impact on the development of AI technology and its potential applications. It is necessary to strike a balance between protecting copyright owners’ rights and fostering innovation in AI for the growth and advancement of the field.