Last week, the UK government published its response to the second consultation on AI and IP. The consultation asked for input on three topics: inventions made by AI, copyright works created by AI and a specific Text and Data Mining exception for copyright. The consultation included six roundtables with participants including individuals, large and small businesses, and a range of organisations, from sectors including the creative industries, tech industry, pharmaceuticals, the third sector, academia, and legal and IP professions. More than 60 people participated in the roundtables and shared their experiences and thoughts. There were also 88 written submissions received. The UKIPO also talked to AI service/equipment providers, artists, entrepreneurs, inventors, academics and other IP offices.
Of the three topics, the response maintains the status quo as far as inventions made by AI and works created by AI are concerned. On the former, as reported previously, the UK require inventors to be natural persons. The result of the consultation means that there is no change in that respect, and there will also not be a special dedicated right for inventions made by AI. The response notes that many respondents felt that inventions by AI should be patentable once AI reaches a stage where this is a possibility and that respondents from the tech sector acknowledged that AI is not yet advanced enough to invent without human intervention. In summary, most respondents favoured no change to patent law, which would be premature. However, it was noted that the UK should stay ahead of technological developments and monitor this area.
As those following my newsletter will know, I agree with this assessment and the conclusion. Unlike some respondents, I think it will be difficult to "stay ahead" of technology here. This is because any law change should be based on evidence of the commercial needs and economic impact on innovation and concerned industries (foremost the pharma industry), which seems not possible ahead of autonomous machine invention becoming a reality. Wait and see seem to be the right call here. It is also noteworthy that the UKIPO stressed that inventions made with assistance from AI are patentable like any other invention in the UK. A helpful reminder to avoid confusion about the issues of inventions made by AI influencing filing decisions for inventions made using AI.
AI copyright works
On the question of copyright works created by AI, these are currently protected by copyright, which protects computer-generated works (CGW) for a reduced term compared to human-created works. Most respondents agreed that this form of copyright is not widely used and that there is no evidence one way or another whether to change the law in this area, so no change was the preferred option. The government has decided to make no changes to existing protection for computer-generated works. As the use of AI to generate creative content is still in its early stages, the future impacts of this provision are uncertain. It is unclear whether removing it would promote or discourage innovation and the use of AI for the public good. The government decided to keep the status quo here and monitor developments.
An interesting point arises from the definition of CGW in the relevant legislation (specifically sections 9 and 178). A CGW is a work generated by a computer in circumstances such that there is no human author. In those cases, the person by whom the arrangements necessary for creating the work are undertaken owns the copyright. Powerful AI systems, such as OpenAI’s Dall-E and Google's Imagen that create all manners of images with all the apparent visible attributes of artistic works based on textual prompts, raise some interesting questions already in this context. While these systems create an image in response to any prompt, creating content using these systems requires a lot of effort to design the prompt that will generate an image meeting the author’s requirements. This LinkedIn post recounting the creation of a magazine cover using Dall-E perfectly illustrates this point. In those circumstances, is the author of the prompt an author of the image? If so, there is no CGW, and regular copyright seems to apply. And If the answer is "no", such that there is no human author and the image, who is the person who made the arrangements necessary? The prompt's author, the system's owner, or both? I am hoping to delve deeper into this question in a future edition.
Text and Data Mining (TDM)
And what about the copyright in the images with which systems like Dall-E and Imagen are trained? The remaining point covered in the consultation addresses this. The remaining topic of the consultation, TDM, should help here. There was a division of opinion in the consultation between rights holders favouring the status quo and users of text and data mining favouring a broader exception not limited to non-commercial use as provided by current UK copyright law.
The current legislation includes an exception for non-commercial text and data mining. Based on the consultation, the government has decided to broaden the exception by removing the requirement for a non-commercial purpose of the TDM. This will make the UK's regime on TDM one of the most liberal ones in the world, for example, compared to the EU, where the exception is limited to non-commercial use, and the US, where it is necessary to demonstrate "fair use". The government intends to facilitate an environment that encourages innovation in the AI space and make the UK an attractive place for AI development. The exception only applies to "lawful" access to safeguard copyright holders' interests. Copyright holders can still ask for payment to access works by putting them behind a paywall.
The broadened exception will make crawling the web for AI training data much more straightforward in that it removes the need to check the copyright status of the crawled works, independent of the commercial or non-commercial purpose of the data collection. This will remove a layer of admin that should make AI development in the UK more efficient. It is interesting to note that little empirical data was put forward in the consultation, so it is difficult to judge what effect the change will have. Depending on how the new exception is implemented, there may be some open questions since the current provision has not been tested in court. For example, what is meant by “computational analysis of anything recorded in the work"? Is training a neural network using portions of text "computational analysis of anything recorded in the work"?. What about Dall-E, Imagen and similar systems – is training such systems from images from the web "computational analysis of anything recorded in the work"? More broadly, is training a neural network or other machine learning algorithm "computational analysis"? I would hope so for these provisions to have the intended effect. Time will tell.
It is excellent that the UK wants to move to lighter regulation in this area. I will be following the developments closely and report here.