Inspiring the next generation ... of machines
With Artificial Intelligence (AI) becoming increasingly integrated into our daily lives, it is important for us to know how to use it, and what our rights are when we do so. This article focuses on how the recent developments in AI, like ChatGPT, can be squared with copyright legislation.
Who owns an AI output?
When we use ChatGPT to produce an output, OpenAI assigns all its proprietary rights in that output to the user, which means that, provided the input is yours, the output can be as well. That entitles the user to profit from the product. For example, if I input this entire article into ChatGPT and prompted it to check for grammar errors, the whole article would then become output which OpenAI then assigns to me.
Companies are already taking advantage of this generous assignment by OpenAI and profiting from it, for example Duolingo have introduced two new features in Duolingo Max which use GPT-4. ‘Explain My Answer’ and ‘Roleplay’ are chat functions in Duolingo Max which explain answers and provide clarification if a user has answered a question incorrectly, or engages in a conversation with the user to help them practice their language of choice.
What about my inputted content?
OpenAI’s Ts & Cs state that they “do not use Content that you provide to or receive from our API (API Content) to develop or improve our Services”. API Models are the more complex models available from OpenAI, which perform more complex tasks than ChatGPT, such as Whisper, which converts audio into text, and Moderation, which detects harmful content. However, OpenAI can still use what it calls Non-API Content, that is both input and output, to train their models. Non-API services include ChatGPT and DALL-E (which “can generate and edit images given a natural language prompt”). Therefore, your input can be used to train OpenAI models, even if the entire content was created by you.
Simply inputting a prompt into ChatGPT is unlikely to produce an output over which the user can assert any proprietary rights, assuming there are any proprietary rights to be had over what is produced. Under English law, a work is only copyrightable if it is original. The meaning of original is simply “not copied” with some “labour, skill and judgement” having been expended (University of London Press Ltd v London Tutorial Press Ltd). To know whether a work produced by AI is just copied from other works or original, we need to know what the input was.
How do you know what was used as input?
The recent and final Beatles song, "Now and Then", which was released earlier this month is an elegant example of AI being used to produce a copyrightable work. For this groundbreaking song, AI was trained to recognise John Lennon’s voice, meaning recordings of John Lennon speaking and singing were used as the input. The AI was then able to isolate his voice from a poor quality demo tape recorded in 1977 into a new recording to be used with instrumentals from Sir Paul McCartney and Sir Ringo Starr.
In contrast, AI like ChatGPT is trained on a huge volume of data. It is therefore exceptionally difficult to trace what data was used to produce a specific output. It is possible to ask ChatGPT to cite its sources, however the sources it cites may not actually exist, as it may just produce a jumble of letters that look like a URL, for example, and there is no way to confirm that any real sources cited are the only ones which have been used.
Going one step further, what do we mean by used? ChatGPT has been trained on reams of data so that it can write like a human being. Requesting a source from ChatGPT is like asking me to cite every teacher I have ever had, as well as everything I have ever read, which taught me to write how I write today.
What about Machine Learning? Who owns its output?
As machines are getting more and more advanced, their output is exhibiting the ability to mimic intelligent human behaviour. When machines can learn without having to be programmed or prompted, it is possible that the machine alone might produce something which is “original” for the purposes of Intellectual Property Law.
The Copyright, Designs and Patents Act 1988 does envisage copyrightable works which are computer-generated, but the owner in law is “the person by whom the arrangements necessary for the creation of the work are undertaken”. If machines get to a point where they can learn and then produce copyrightable work by themself, it might become difficult for us to point to a person who has made the arrangements necessary for the creation of the work, apart from the creator of the machine in the first place, such as OpenAI. This would make OpenAI, and other AI companies like it, owners of infinite copyrightable works.
It is notable that the government decided not to make any changes to the law on computer-generated works or AI-devised inventions. In its June 2022 response to consultation, it concluded that there was “no evidence at present that protection for CGWs [Computer Generated Works] is harmful and the use of AI is still in its early stages” (Artificial Intelligence and Intellectual property: copyright and patents: Government response to consultation). This was five months before the launch of ChatGPT and one month after the Writers Guild of America strike, which highlighted the concerns of writers in Hollywood over the growing use of AI in the profession.
Conclusion
Children are often asked what they want to be when they grow up, but for ChatGPT the opportunities are endless. ChatGPT is only celebrating its first birthday, but it has already achieved so much. Indeed, it is predicted that OpenAI’s baby could add up to $4.4 trillion every year to the global economy. How many parents can say that?