As part of our Thought Leaders series, we explore artificial intelligence and copyright.
Forward: features discuss and celebrate the best of innovation and exploration from the scientific and entrepreneurial worlds.
“There once was a bot named ChatGPT
Whose creativity was quite the set
With words so bright
And ideas just right
In writing, it's hard to forget”
This limerick was written by ChatGPT. It took under a second to compose, and isn't terrible, even if the opening couplet doesn't rhyme. ChatGPT is now being used by university students to write essays, by marketing agencies to script ads, and teachers to build lesson plans. Launched in November 2022 by Open AI, more than 100 million users are now generating text with the bot.
The formidable creativity of ChatGPT and image creation engines such as Dall-E and Midjourney raise a legal question: do these engines deserve copyright protection?
Recent AI developments
AI has been in the mainstream media recently since the creation of ChatGPT made by OpenAI. ChatGPT was released in November 2022 and by January 2023 had been used by 100 million users. The release of this new chatbot has made AI accessible for a huge number of people around the world and aims to reduce the workhours needed to perform a vast array of tasks, as well as attempting to push the boundaries of computer-generated language.
What makes ChatGPT so different, and more impressive, than previous uses of AI is the amount of extra training that it was provided with. This training involved the AI being asked to give multiple answers to the same question with a human AI trainer then ranking the answers worst to best.
Director of foundational AI research at the Alan Turing Institute in London, Professor Michael Wooldridge said that it “would take 1000 human lifetimes to read the amount of text the system [has been] trained on.” Access to this amount of knowledge is mind boggling and allows the system to in theory answer most questions and even form creative writing pieces and compose music.
Current rights of AI
There is currently some pre-existing copyright protection for computer-generated works without a human author. Within the UK this protection lasts for 50 years in comparison to the usual life of author plus 70 years of a standard copyright protection. However, the “author”, in the case of computer-generated works, is referred to as “the person by whom the arrangements necessary for the creation of the work are undertaken.”
This protection is often discussed and protested due to the traditional UK test of skill, labour and judgment for copyright to be granted. So, the question is does AI have skill? Or judgement?
Further, the European test for copyright depends on inherently human attributes such as “personality”, “free and creative choices” and the “stamp of the author’s personal touch” therefore should it be possible in any case for computer generated works to hold copyright?
This links heavily to a previously considered and discussed case of DABUS which relates to whether AI can be classed as an inventor. As of December 2021, the EPO has rejected two appeals on whether DABUS could be designated as inventor for a patent application. However, in September 2022 it was announced that the UK Supreme Court would hear the DABUS appeal on 2 March 2023 and we await their ruling in this case.
AI and copyright
Some argue that granting copyright to AI-generated content could incentivize the development of more sophisticated AI programs, which could benefit society as a whole. Others, however, argue that AI lacks the capacity for creativity and originality necessary for copyright protection, and that doing so would risk undermining the fundamental principles of copyright law.
It is important to note that the legal landscape around AI's copyright rights is still very much in flux, and it remains to be seen how this issue will be resolved in the years to come. Ultimately, the question of whether AI should be granted copyright protection will likely depend on a variety of factors, including the level of creativity and originality displayed by AI-generated content, as well as the policy preferences of lawmakers and the broader public.
Artists and authors
One aspect which is currently causing a lot of controversy around advancing AI is the use of artists work in the training of image creation. Artists’ work has been used, without their permission, within image creation apps and websites. By using the website https://haveibeentrained.com/ it has been possible for artists to see which pieces of their work has been used as training. This involves the AI using the style and substance of the artwork to replicate something similar when instructed by a user in a certain way. These instructions don’t even need to reference the artist by name for the AI to create something of a similar style. The evidence of signature marks of these computer-generated images shows how much influence this previous art has had on the AI created images.
In theory AI cannot infringe copyright intentionally as it does not have any conscious thought, meaning the act of infringing lies in the hands of the users or the creators of the AI themselves. Although it is very much possible for the AI to create copyright infringing work and raises questions surrounding the artistic originality and authenticity of AI generated images.
Almost more importantly is the question of whether AI art reinforces the existing social biases and inequalities within society. The lack of diversity already in the industry means that trained algorithms also hold this bias therefore encouraging it to appear in AI-generated art too.
A study by researchers at Fujitsu found many examples of the biases within computer generated art. They found that the over simplification of styles which occurs when “researchers conveniently define style in a manner that suits their model’s performance” results in many, often culturally significant, art styles being restricted to one defining characteristic resulting in the generated imaged failing to connect with the style at all. Other examples were seen where certain parameters resulted in the face colour of the piece “Black Matriarch” was tinted shades of blue or red whereas when the same parameters were used on fairer faces this change did not occur. The study concluded that “there is a noticeable difference in the way faces are converted across styles based on the colour of the face, indicating potential racial biases.”
While AI systems can generate novel and unexpected output based on their training data, this does not mean that they can truly produce an original output which has not been pre-defined by their creators so this could continue to give artists the upper hand.
Conclusion
It is interesting to see how similar ChatGPT’s work can be to our own work. For that reason, part of this article was in fact written by the AI itself. Can you tell which part was, and would you have noticed?
AI has the opportunity to really change the world but only in a safe and morally correct way with the correct legislation in place. It will be interesting to see how policies develop over time to protect human and computer alike and what new intellectual property protection will result from this.
This Forward Thought Leader was originally written by Ellie Knowles.
