Debunking myths about AI and copyright – Astana International Forum

AI as an ally, not a threat to creativity

Meta’s Director of Public Policy, Sarim Aziz, called for a realistic understanding of AI’s capabilities and emphasized the importance of preserving the balance between innovation and creators' rights. He stressed that modern language models are merely assistive tools, not independent creators.

“AI doesn’t understand the physical world, it has no long-term memory, no ability to plan, and it can’t solve fundamentally new problems. It’s not a god, as some might think. It’s just a tool,” he said.

Aziz then addressed the legal dimension, challenging the notion that training AI on publicly available data infringes on copyright. He explained that AI models do not memorize or reproduce specific works, but instead learn patterns and abstract structures from large datasets.

“The training process is about extracting patterns, not reproducing expression. The models don’t store protected content or provide direct access to it,” Aziz explained.

He advocated for relying on existing flexible legal frameworks, such as fair use and exceptions for text and data mining, which have long been in place in AI-leading countries like the U.S., Japan, and Singapore. According to him, these tools help strike a balance between copyright protection and the need for technological progress.

Aziz also highlighted the importance of transparency. He acknowledged that full disclosure of all data sources used to train a given model is technically unrealistic. For example, the Llama 4 model was trained on 30 trillion tokens of text, images, and videos, making it impossible to track the copyright status of each element.

He further argued that the market is capable of finding its own solutions without regulatory intervention. As an example, he cited Meta’s agreement with Reuters, which allows the company to use up-to-date news content with official source attribution in Meta AI.

“This is an example of how the market can resolve issues independently—without regulators stepping in. We license content for specific use cases, such as news,” he said.

Concluding his remarks, Aziz shared an inspiring example from Kazakhstan, which he described as a case of how open technologies empower countries to develop their own AI models tailored to national languages and cultural contexts. Kazakhstan trained its own version of Llama on 150 billion tokens in Kazakh and Russian, and made it available to the local developer community.

Keeping humans at the heart of intellectual property

Daren Tang, Director General of the World Intellectual Property Organization (WIPO), emphasized that AI is just the latest technological shift - one the intellectual property (IP) system is already learning to manage.

“So my wish for all the member states is to make sure that whether it's AI or some new tool that will emerge, we don't know which way technology will turn, that all these tools are used to empower, enable, and support the human creator and the human innovator, and not to destroy, replace, or undermine the human creator or human innovator,” he said.

Tang noted that the IP system has been adapting to innovation for over 130 years, from the advent of airplanes to mobile phones, and that it must continue to evolve in the AI era while keeping the human at its core and updating legal frameworks as needed.

AI as a tool for a more efficient patent system

Jian Xu, First-level Inspector at the China National Intellectual Property Administration (CNIPA), focused on AI’s practical applications in the patent field. China is actively integrating AI to enhance procedures related to patent examination, data search, and classification.

He explained that the country is developing new use cases for AI in the patent system, including the adoption of large language models.

“At present, the China Intellectual Property Office is actively exploring the application of large language model technologies for a series of research and project initiatives. We hope to use these technologies to deepen semantic understanding, further improve efficiency and accuracy, assist examiners in better inventive step evaluation, and in drafting notices—continuously improving workflows and enhancing the quality and efficiency of patent examination,” said Jian Xu.

Earlier, Kazinform News Agency reported that President Gordana Siljanovska-Davkova of North Macedonia called for greater female participation in international politics.