Whose IP is it anyway? Possible policy changes to protect AI-generated works | Fieldfisher
Skip to main content

Whose IP is it anyway? Possible policy changes to protect AI-generated works



United Kingdom

Towards the end of last year, the UK Government launched a consultation, seeking views on whether the UK's current IP framework is still fit for purpose in respect of continued developments in Artificial Intelligence (AI). The consultation raised a number of important questions on whether an AI-generated work should be protectable by IP law and, if so, who the owner(s) of any such rights should be.

The government published a response to its call for views in March 2021, which highlighted the divergence of opinion about whether inventions or copyright works created by AI should benefit from IP protection in the same way as human creators, if at all. Notably, although perhaps unsurprisingly, this difference of opinion was particularly stark between respondents from the technology industry, and those respondents with an interest in the arts and other more traditionally creative industries.

The government has set about considering which aspects of the UK's IP framework might benefit from a revamp to redress the balance between (i) encouraging innovation in AI technology and (ii) protecting and promoting human creativity, which many respondents still consider the central role of UK IP law.

AI and IP policy change consultation

On 29 October 2021, the UKIPO launched a new consultation (see Press Release here and consultation here) on how the copyright and patent system should deal with AI. It is part of the Government's wider national Innovation Strategy (published in July 2021) and AI Strategy (published in September 2021) to maintain and cement the UK's status as one of the leading IP environments in the world, despite such exceptional challenges over the past 18 months.

The consultation focuses on three specific areas of possible policy change:

1. Copyright protection for computer-generated works (CGW) without a human author;
2. Licensing or exceptions to copyright for text and data mining (TDM); and
3. Patent protection for AI-devised inventions.

Copyright protection for CGW

Unlike most other countries, CGW are currently protected in the UK for a period of 50 years – already significantly shorter than the usual life-plus-70-years afforded to works created by humans – but the government has asked for views as to whether this protection should remain unchanged, reduced further in respect of scope and/or duration, or scrapped altogether.

TDM exceptions

As explained in the consultation, "text and data mining is the use of automated computational techniques to analyse large amounts of information to identify patterns, trends and other useful information", which often means copying the material (some of which will be copyright-protected) in order to analyse it. TDM can be used to develop and "train" AI and can also enable research, e.g. the analysis of medical and scientific data, business intelligence and data analytics. TDM automates and speeds up what would usually be done by eye such as reading a document, making notes and assessing relationships, patterns and trends. As such, it is key to allow access to content that will advance and improve the quality of research.    

Copyright works being used to "train" AI continues to be a hotly debated topic. The issue stems from the fact that, while "copies" made inside a human brain do not infringe copyright (i.e. you would not be faced with a lawsuit if you ever find yourself unable to get that Rick Astley song out of your head), copies made within an AI “brain” may be considered infringing.

For example, if an AI was to store a copy of "Never Gonna Give You Up" within its memory, this could be a problem even if it never produces an infringing output.

The risk of infringing copyright or other rights in training AI may therefore hinder the development of the technology altogether, particularly if it results in a stronger reliance on a smaller data pool of out-of-copyright works that, in turn, could further exacerbate data bias problems in AI software.

With these issues in mind, the consultation has asked for views on the following policy options:

  1. do nothing;
  2. improve the licensing environment for TDM;
  3. expand the scope of the TDM exception (which currently applies only to non-commercial use) to cover commercial research and databases;
  4. adopt TDM exception for any use, with rights holder opt out provision (as has been introduced by the new Copyright Directive 2019/790, which the UK it is not implementing due to Brexit); or
  5. adopt TDM exception for any use with no opt out provision.

AI patent inventors

Another key issue in the AI and IP conversation is AI inventorship, particularly in light of the UK Court of Appeal decision in the DABUS case, (on which we will be publishing a blog shortly and see another related blog here). In brief, the latest Court of Appeal ruling from September 2021 held that a patent cannot be granted where an AI system (in this case a machine called DABUS) is named as the inventor because, under UK law, a “person” must be named as the inventor. This is also in line with patent applications naming DABUS as inventor being rejected before the USPTO and the EPO. However, the decision goes against earlier rulings in South Africa and Australia that an AI system such as DABUS can be named as the inventor and South Africa is the first country to have actually granted a patent with an AI inventor (albeit without any formal examination so it could be challenged and the decision in Australia has in fact been appealed). The UK story may not be over yet, however, as there is speculation that Thaler (the team behind the DABUS patent applications) has recently applied to the UK Supreme Court for permission to appeal, (likely bolstered by the dissenting voice of Lord Justice Birss in the Court of Appeal ruling).  

The options here are to do nothing, or to expand the definition of "inventor" to include either humans responsible for AI systems and/or the AI system itself. One further suggestion is to develop a new type of protection, similar to a patent but with more limited exclusive rights, that would catch those AI-devised inventions where a human inventor cannot be clearly identified.


The consultation is open for responses until 7 January 2022. It will be very interesting to see how the different interested parties respond and how the Government reacts to ensure the UK becomes the 'global AI superpower' that it envisages.

Areas of Expertise

Intellectual Property