An intellectual property attorney for OpenAI recently spoke at the Vanderbilt University Music Law Summit and discussed the challenges of negotiating individual licensing agreements for copyrighted works used to train large language models. This poses a significant barrier to the existence of programs like OpenAI’s ChatGPT. In response to this issue, the attorney proposed the implementation of an artificial intelligence tax specifically targeting large language models to ensure that developers and owners of these models internalize the externalities produced by using copyrighted training data.
Traditional intellectual property laws struggle to address the complexities of using copyrighted works to train AI models, particularly when the use of the work is to learn from it rather than to copy it. The attorney compared this scenario to an energy reactor generating electricity from recorded music, highlighting the challenges in determining what is owed to creators for the unforeseeable use of their works. The value of a work in the marketplace as a work of art may not align with its value to language models, complicating the question of fair compensation.
An AI tax targeting large language models based on their parameter size could offer a structured approach to ensure companies using protected materials contribute back to society and the creative community. This tax could fund compensation programs for cultural, educational, and technological initiatives, providing an alternative to impractical individual licensing agreements. By internalizing externalities, the AI industry would be forced to account for the broader impact of their technologies and balance innovation with social responsibility.
The implementation of an AI tax will require careful consideration of factors such as the scale of data used by AI, revenue generated by AI applications, and progressivity in tax rates based on entity size or data volume. The goal is to ensure that the tax is paid by AI developers and companies rather than passed on to consumers. Transparent allocation of funds, involving stakeholders from creative groups, will be crucial for the effectiveness of the tax policy. Consensus among stakeholders is essential to balance the compensation of society for the use of copyright materials with the encouragement of innovation in the AI sector.
In conclusion, an AI tax is viewed as a necessary policy measure to address the challenges posed by large language models using copyrighted material. By internalizing externalities and ensuring fair compensation to creators, this tax could help strike a balance between fostering innovation and social responsibility in the AI industry. Careful planning and collaboration among stakeholders will be vital to the success of implementing an AI tax that benefits both society and the creative community.
