Three leading researchers in human-computer interaction have told the journal Nature about their concerns over the failure of current copyright licences to protect their work from misuse by AI.
In a Correspondence published today in Nature, Katie Seaborn, Daniel Gardner and Katta Spiel point out that while "generative artificial-intelligence models, notably large language models, are disrupting intellectual-property rights", existing copyright licences are proving a very clumsy way to defend them.
In particular, they raise the problem that the copyright licences intended to stop AI from misusing their work also have the effect of stopping humans from legitimately re-using it — something they feel is vital to human expression.
This needs to change, the authors say. "Copyright licence options need to be amended now. We argue that a new convention for copyright is urgently needed, and one that explicitly accounts for nuanced control over AI while allowing human (re)use."
When an organisation is publishing a researcher's paper, it offers them a choice of copyright licences. "Many of us choose the Creative Commons BY-ND 'no derivatives' licence, which forbids distribution of any further re-use, re-fashioning or extension of the work," explains Dr Katie Seaborn, an Associate Professor here in the Department.
"We choose it as the only option that can combat AI use of our work. However, this option not only bans AI from using our work, but also forbids humans (and potentially even ourselves) from re-using, remixing or extending our work in future."
"So, when trying to protect our work from AI misuse, we're forced to choose licenses that prevent legitimate re/use for creative human expression," agrees Dr Daniel Gardner, a lecturer and researcher in human-computer interaction at Robert Gordon University, Aberdeen.
Demand for unrestrained AI-driven data collection
The problem has come about, he adds, because of the rapid development of AI and subsequent demand for unrestrained AI-driven data collection. This is fundamentally challenging the impetus for copyright licences like the dominant CC BY with its ethos of "helping build and sustain a commons of shared knowledge and culture".
The issue is widespread, the researchers say. "AI mongers exploit free and easy access to the ever-expanding trove of human output that resides on the Internet – and subsequently may not honour copyright or ask for permission," says Dr Katta Spiel, Assistant Professor in Human Computer Interaction at the Technical University of Vienna.
"A new convention for copyright is urgently needed, and one that explicitly accounts for nuanced control over AI while allowing human (re)use." Katie Seaborn, Daniel Gardner and Katta Spiel
"A new convention for copyright is urgently needed, and one that explicitly accounts for nuanced control over AI while allowing human (re)use."
Katie Seaborn, Daniel Gardner and Katta Spiel
As the authors put it in their letter to Nature, "Unchecked mining of text and images by AI firms is thereby shaping and constraining human autonomy through copyright licence models, forcing creators — including artists, authors and designers — to reassess their relationship with copyright or surrender it.
"The failure to predict or safeguard against the emergence of an industry built on automated content recycling was naïve," they add. "Now it is crucial to ensure that the human ethos of enabling and sustaining cultural output and shared knowledge survives in the age of AI."
The authors' concerns over the way that AI is forcing a re-examination of copyright issues are also shaped by another issue: the copyright licences offered to researchers by the ACM, the world's largest educational and scientific computing society, when it publishes their research.
The ACM recently introduced a new set of AI tools to its digital library. These produce AI-generated summaries and 'podcasts' of articles by AI-generated podcasters without the researcher's involvement.
The ACM says that the objective is "to provide AI functionality that helps authors promote their research and makes accessing the Digital Library more productive and accessible." It adds that these tools "are designed to complement articles, not replace them".
AI misrepresenting researchers' output
But the three researchers noticed that by producing unauthorised AI-generated summaries of their work, the ACM is infringing the terms of the 'no derivatives' (CC by-ND) copyright licence that both parties agreed to. Moreover, they discovered examples where the summaries and podcasts contained superficial, inaccurate, and hallucinated content — thereby adverse to research communication, as well as misrepresenting authors' own words and output.
Concerned by this, Dr Seaborn (pictured left) requested that the ACM withdraw the AI-generated summary and podcast for one paper, 'Operationalizing Perceptions of Agent Gender: Foundations and Guidelines'. Within a month, the ACM honoured the request for removal of AI summaries and podcasts on all Dr Seaborn's papers. A spokesperson was asked for a comment but did not wish to provide a statement.
Such issues have fed Dr Seaborn's perspective that "we cannot allow free reign by AI systems without first protecting human creative expression.
"First, AI must be accounted for as an entity separate to human production," Dr Seaborn says, "as AI cannot hold copyright nor make decisions. Second, humans must be centred as conscious actors with creative liberty."
"AI is a copyright shredder," Dr Gardner concludes, "but it doesn't need to be. We need to create a new standard for the age of AI."
