Should AI-generated content be protected by copyright?

Copyright law arose to protect human-centred creativity from being copied. The law gives the creators of literary, dramatic, musical, audio-visual and artistic works the right to control how it is used or reproduced. The IPO has recently consulted on the relationship between copyright and artificial intelligence (AI) and whether AI-generated content should be afforded the same protections.

The consultation could be interpreted as proposing two fundamental questions – does copyright create obstacles for AI development; and does copyright provide an incentive for AI investment in the UK?

Fundamentally, copyright does not create any unreasonable obstacles to the development of AI. Copyright ensures that the creators whose works are used to develop AI capabilities are appropriately remunerated. There should be protections that keep copyright works from being used without authorisation.

“These protections are not obstacles. They are necessary. Copyright underpins the value of the information that is being used to develop AI.”

The AI development process

The process undertaken to develop the AI model typically involves the ingestion of a large number of works. Let’s take the example of music, AI may enable the machine to learn which melodic patterns are most likely to lead to a popular composition. The holders of the rights in the musical works and sound recordings used and reproduced in this way should be remunerated, and have a right to know their works have been used in such a way.

Legal licensing framework

The current legal framework is focused on a licensing regime. Licensing provides certainty for those investing in AI as it enables developers to control the risk of infringement by embedding the recognition of intellectual property rights within the AI code itself.

Therefore, AI developers should have to maintain auditable records of what data has been used. Then, if input data contains copyright works, it would be possible to determine whether a license is required, and the original creator would be rewarded.

This approach is also desirable from a wider public interest perspective. A legal requirement to maintain auditable records will enable developers and operators of those systems to demonstrate that they have used “good data” that is less likely to lead to discriminatory or biased outcomes.

“This would help to instil trust in AI systems.”

Should copyright exception be broadened for AI?

Some have argued that exceptions to copyright (i.e. instances where copyright works can be used without a license from the copyright owner) should be broadened for the purposes of AI development.

Broadening exceptions for developing AI would prejudice rights holders. Significant changes in copyright exceptions to allow unfettered, free access to UK content for AI purposes could merely result in the transfer of value from the UK creative industries to predominantly large technology firms.

“That is not to say there has to be trade-off between human-centred creativity and AI. Far from it. The creative industries are significant investors in – and users of – AI.”

For example, ‘Loyal AI’ is a suite of editorial assistants that deploy machine learning to suggest sources of inspiration for new content ideas. Image library websites use AI image recognition tools to assist both in the upload and appropriate tagging of image content to support customers in finding relevant images.

These operate through a licensing model. This demonstrates that licensing models have shown themselves to be sufficiently agile to adapt to advances in technology and demonstrates that broader exceptions are not required to facilitate investment in AI.

Conclusion

Ultimately, copyright protects the intellectual output of human beings, anything AI-generated should not be protected through copyright. There are alternative related rights and intellectual property rights that should be explored.

The works that are used to develop AI, or contribute to AI-generated works, already have copyright protection, therefore any ‘doubling-up’ of protection for the same underlying content does not make sense.

That is not to say that investment in AI applications does not deserve to be protected and rewarded, but it should be distinct from copyright protections for original works.

Whilst it is my intention that these views reflect the views of our members this should not be read as the official BCC position – this can be obtained in our response to the IPO’s call for views.

Share this story