Is training AI fair use?

Fair use is a legal doctrine that allows for limited use of copyrighted material without the need to acquire permission from the copyright holder. The four factors that are considered when determining whether a use is fair are:

1. The purpose and character of the use, including whether it is for commercial or educational purposes.
2. The nature of the copyrighted work.
3. The amount and substantiality of the portion used in relation to the copyrighted work as a whole.
4. The effect of the use on the potential market for or value of the copyrighted work.

In the case of training AI, the purpose and character of the use is to teach the AI to perform a particular task. The nature of the copyrighted work is the data that is used to train the AI. The amount and substantiality of the portion used is the amount of data that is used to train the AI. The effect of the use on the potential market for or value of the copyrighted work is the impact that the AI has on the market for the original work.

Training AI is fair use under U.S. copyright law. First, the purpose and character of the use is transformative. The AI is not simply reproducing the copyrighted work, but is instead using it to create something new. Second, the nature of the copyrighted work is often factual data, which is not as protected by copyright law as creative works. Third, the amount and substantiality of the portion used is often small, as the AI only needs to be trained on a representative sample of the data. Fourth, the effect of the use on the potential market for or value of the copyrighted work is often minimal, as the AI does not typically compete with the original work.

Of course, there are also arguments that can be made against considering training AI as fair use under U.S. copyright law. First, the purpose and character of the use could be seen as commercial, as the AI can be used to create products or services that are sold for profit. Second, the nature of the copyrighted work could be seen as creative, as the data may have been created by a human author. Third, the amount and substantiality of the portion used could be seen as large, as the AI may need to be trained on a large dataset of data. Fourth, the effect of the use on the potential market for or value of the copyrighted work could be seen as significant, as the AI could compete with the original work.

That said, the arguments in favor of considering training AI as fair use are strong, and it is likely that courts will continue to find that training AI is fair use in many cases.

The world’s biggest tech companies are in talks with leading media outlets to strike landmark deals over the use of news content to train artificial intelligence technology.

View referenced article