As you will no doubt be aware, Ed Sheeran has been defending himself in the Manhattan Federal Court against claims that he copied Marvin Gaye’s classic Let’s Get it On when creating his 2014 hit Thinking Out Loud. Whilst the songwriter had a successful outcome on this occasion, meaning he will not be quitting music as suggested, it could be that the real challenge that lies ahead for artists and other creatives comes from the emergence of Artificial Intelligence.

Artificial Intelligence, or AI, has been all over the headlines in recent months. University students are allegedly submitting coursework that has been generated using ChatGPT and photographer Boris Eldagsen won first prize in the creative open category of the Sony World Photography Awards 2023 with an AI-generated image entitled “Pseudomensia: The Electrician” (readers will be pleased to note that Eldagsen did in fact come clean and refuse the prize after he was awarded first place).

Artificial Intelligence will likely have a big part to play in our everyday lives in the not-too-distant future. This view is supported by research from the World Economic Forum, which revealed that approximately 75% of companies surveyed said they will be looking to adopt Artificial Intelligence in some way, shape or form in the next 5 years. However, the emergence of Artificial Intelligence will inevitably come with challenges. One key talking point, particularly amongst the creative industries, is how Artificial Intelligence will interact with the law, specifically in relation to work subject to copyright.

The Risks of Using AI-Generated Work

To understand the legal risks of generative AI, we must first explore how it works.

Firstly, we need to understand “text and data mining”. Text and data mining is a technique that involves analysing substantial “datasets” to identify patterns or trends. The product of text and data mining has a number of uses, but one of the most prominent at the moment is use as input and training data for AI systems and models.

The datasets used for text and data mining are vast and whilst they can be populated with whatever the creator of the dataset desires, they are often full of data that either is unlicenced (meaning permission has not been obtained from the owner of copyright in the data) or that has been “scraped” from the internet using a process known as “web scraping”. Web scraping is a data collection method that operates by searching the internet (including selected websites) and then extracting data that is deemed useful in some way. The extracted data can then be copied, retained in large datasets and used for things such as data analysis/mining.

One of the main issues in using generative AI is that ”end users” of AI software do not know how the AI software has arrived at the “end product” that is provided to them (and what data has been used to generate that end product). If the software has used unlicenced materials (which are subject to copyright) in their datasets and those datasets are then utilised to generate the end product, then the owner of the AI software is potentially committing copyright infringement. If the end user then went on to use the end product, they may also unknowingly be committing copyright infringement.

Only last month various streaming platforms (including streaming giants Spotify and Apple Music) removed a song that used AI interpretation of real-life performers Drake and The Weeknd on copyright grounds. The song went viral, but shortly afterwards the creator, who is known as @ghostwriter, said the song Heart On My Sleeve was created by AI that had been trained on the artists’ voices.

Furthermore, Getty Images has recently taken legal action against emerging text-to-image AI software company Stability AI. Text-to-image AI software allows users to enter instructions for an image to be created by the software, and Getty is arguing that millions of its images have been used to train AI systems without a suitable license. It is alleged that this amounts to an infringement of Getty’s copyright.

The allegation in the Getty case appears to be that web scraping was undertaken on work protected by copyright without the correct licences or consents being in place, and that the information/images scraped eventually formed part of datasets that Stability AI’s Artificial intelligence system Stable Diffusion was trained on.

Clearly the practice of using unlicenced material or undertaking web scraping to populate datasets that are later used to train Artificial Intelligence systems has the potential to cause issues in relation to copyright.

The Current Legal Position

The law in relation to Artificial Intelligence and Intellectual Property is, at this stage, unclear. This is not surprising, given that Artificial Intelligence is still in its early stages. It is expected that the Government will intervene in some way to provide clarity in the not-too-distant future (although how that is to be done has seemingly yet to be fully decided).

The UK Government’s initial response to a consultation on Intellectual Property and Artificial Intelligence (published on 28 June 2022) was to decide that the UK would seek to amend current laws to make it easier for “text and data mining” to take place without infringing copyright. This was subject to parliamentary debate on 01 February 2023, at which point the proposals were termed “misguided” by the Lords Communications and Digital Committee (LCDC). It was strongly suggested that they should be immediately dropped and following this, the proposals were indeed withdrawn. However, following a 15 March 2023 review by Sir Patrick Vallance, the Government has now confirmed that the Intellectual Property Office (IPO) will produce a code of practice to provide guidance and support to Artificial Intelligence firms in accessing work subject to copyright as an input into their models. It is hoped that the code of practice will ensure protection of generated output (for example by including labelling), and that it will strike a balance between rights holders and Artificial Intelligence operators.

In regard to ownership of AI generated content, it is not thought that there will be any changes made to the UK’s existing copyright laws relating to ownership of “computer-generated” works (as set out in s.9(3) of the Copyright, Designs and Patents Act 1988). Currently, the “author” and therefore owner of AI generated content will be deemed to be “the person by whom the arrangements necessary for the creation of the work are undertaken”. This is likely to be the creators of the AI software (for example, ChatGPT was developed and is owned by OpenAI, and therefore content generated by ChatGPT will belong to OpenAI in the first instance).The position on ownership can be displaced by an assignment in the platform operator’s terms, for example Midjourney AI (a text to image generative AI software developed by Midjourney, Inc.) has terms that state that the end user of the software owns all assets that they create through the AI platform. However, there are two potential issues with that. Firstly, the end user cannot be certain that Midjourney had the correct legal standing to give away ownership of intellectual property rights in the assets created via the platform to their users (for example, if they are committing copyright infringement during the “creation” process, then they likely cannot assign any intellectual property rights in the work to end users). Secondly, Midjourney’s terms state that the user grants back to Midjourney a perpetual and non-exclusive licence to use any assets created via the AI platform, and this means that whilst technically ownership of intellectual property rights in the assets may rest with the end user, Midjourney has a significant number of rights in relation to the assets that have been created (and this undermines the end user’s ownership in some ways).

Concluding

One of the promises of generative AI is to provide creative output for even the most un-creative. However, this itself is a controversial topic, and many who operate in the creative industries are rejecting and resisting the idea of generative AI on the basis that they feel it will put creative jobs at risk.

Beyond the copyright issues, it is also important to recognise that output from generative AI has been shown to be factually incorrect in many cases, or even fabricated. The quality of the training data and the way in which a request is framed can create very different results.

Ultimately, the jury is out on the bigger picture, as to whether AI is a glorious opportunity or a threat to traditional creativity. We’ve seen Elon Musk speak out against its wholesale adoption, and Geoffrey Hinton, the so-called godfather of AI, resigned from Google saying he now regretted his work.

For now, we will just have to wait and see how the law develops to accommodate Artificial Intelligence.

This is not legal advice; it is intended to provide information of general interest about current legal issues. You should seek specific legal advice before acting in reliance on any of the information provided.