Who owns intelligence? AI and intellectual property rights

Blog series Innovation and enterprise

Author Jeremy O’Hare, Research and Business Development Manager

Can you imagine a world without the internet? Or without social media or high-speed connectivity? That may well be what it’ll be like in twenty years, trying to imagine a world without artificial intelligence (AI).

The AI revolution is making waves in more areas of our work and everyday lives. The power of AI tools to increase productivity, create ‘new’ content, and solve complex problems is immense. It feels like we’re all on the cusp of something very significant, even if we haven’t yet grasped the enormity of it.

The technological changes are impacting all aspects of business, and it’s raised some deep questions around intellectual property (IP) rights. Such as, can copyrighted works be used to train AI models? And who owns the rights to a generative AI work? Can machines own IP or will it always be humans?

What are the intellectual property issues involved?

AI and IP rights can be grouped into three parts: inputs, outputs and foundation models. There are IP issues for all of them. The foundation models are like a brain. Millions of bits of data, such as text, images and sounds are inputted to create a neural network. Imagine your brain as a dataset of memories, experiences, feelings and facts. You can create things such as sing a song you heard yesterday or create a new work of art inspired by artwork you saw a years ago.

AI, artificial neural network (ANN) models are like this, incredibly sophisticated artificial brains and have an IP value in themselves. These kinds of neural networks allow for generative AI, that is platforms that create or ‘generate’ new work from user prompting.

There’s IP in the coding protected by copyright and trade secrets in the training methods, or its ‘model weights' to give another example. AI developers rely on their IP rights to create long term value in their platforms.

But there’s another side too.

What are the intellectual property issues in AI inputs?

There are IP rights in the inputs. That’s the content and data that gets fed into the neural networks. The question is where did this come from? Were copyrighted text, images or music used in the training of these AI models? If so, do the owners have a right to object on the grounds of copyright infringement?

A crucial case in determining this in the UK was Getty Images v Stability AI. This was heard at the High Court last year. A central point argued by Getty was that Stability AI’s platform Stable Diffusion contains copies of its images, effectively making it a secondary copyright infringement.

The question of primary copyright infringement was side stepped as the training for the foundation model could not be proved was done in the UK. The fact that IP rights are territorial by country adds another layer of legal complexity.

Stability argued that Stable Diffusion did not store or reproduce visual information from training images, and so cannot be infringing copies. The Court held that AI model weights are not a ‘copy’ as defined by existing copyright law in the UK. This is because an AI model has statistically trained parameters and not stored copies.

Given the novel nature of the legal circumstances, Getty has appealed this decision. A further ruling is not expected until the end of 2026 at the earliest.

Parliament are also looking at the issue, and a recent House of Lords report is recommending a licensing model for rights holders in allowing AI developers using their works as inputs.

What are the intellectual property issues in AI outputs?

There’s an important principle when it comes to ‘owning’ an output created by an AI platform. While many platforms don’t claim ownership of works generated, always check their terms and conditions. It’s also not a given that you own the copyright, in fact you may be responsible for infringement if you’re using an output that could be too similar to an existing copyrighted work.

Therefore, in practice it’s good to document everything you do in the creation of an AI work, including the prompts used.

Also be mindful that different countries and regions are adopting different laws. For instance, if you’re working in or with the EU in creating AI content, be aware of the implications of the EU AI Act.

It’s always advisable in all of these aspects to seek legal advice, especially as there is currently uncertainty over how these rights are played out in practice. Businesses that are increasingly reliant on using AI outputs, should especially get the legal guidance they need, as their business could be impacted by future laws or rulings.

What could the future hold?

Answering this question is as hard to comprehend as an AI algorithm. But we are starting to see how some of the difficult questions of rewarding innovation for AI developers while recognising the ownership rights of creators are beginning to be resolved. So, will we ever arrive at the right balance between human and machine? My ChatGPT says that this is one of the ‘biggest questions of our time – and the honest answer is: probably yes, but not automatically.’

So, it comes down to believing the bot, and it says ‘yes’.

Learn more about IP

If you’d like to learn more about IP, particularly why it matters to you and your business, check out our regular programme of workshops or check out our website to find out how we can support you today. If you have a new invention and would like to have a patent search done, our team are here to help.