The Board of Directors of the Science Fiction and Fantasy Writers Association (SFWA) has been monitoring the rapidly evolving technological and legal landscape as it pertains to machine-generated works. We continue to hold conversations inside our organization, as evidenced by our recent call for posts that brought in a wide variety of important perspectives. Additionally, the Board has simultaneously reached outward to other organizations made up of publishing creators, with a focus on fiction writers.
The Board recognizes there is much to learn, and is cautious to offer advice or warnings in this area for two reasons: one, the rate of adaptation and technological evolution is accelerating at unprecedented speeds; and two, foundational legal principles pertaining to copyright and fair use have yet to be definitively established.
Despite this, SFWA feels that there are relevant time-honored principles that can help guide our community through understanding and reacting to the evolving technological and legal landscape. This statement should not be construed as legal advice, not least because our community of creators spans many countries, but rather as guidelines to help creators around the globe.
This will not be SFWA’s last word on the subject; in addition to the initiatives mentioned below, expect further announcements over the next few months and calls for volunteers with subject matter expertise to staff the newly founded Emerging Technology Committee.
In the meantime, these principles must continue to hold true:
1. Creators must be compensated for the use of their work.
While the degree and scope of data collection involving published work is still unclear, all indications are that the popular machine learning products have incorporated substantial amounts of creative work without regard to the rights of the authors and artists whose work makes those technologies possible. To whatever extent these machines appear to generate works of art, it is because they incorporate the work of creative people. To date we have seen no real recognition of this debt, even as these software tools are increasingly used by opportunists to flood our markets with trash.
SFWA counts among its members many highly creative and forward-looking authors with diverse opinions about how best to compensate writers for the use of their work. Core to these arguments, however, is the fundamental understanding that if we, as a society, are to have culture, we must have writers and those writers must be paid.
For over 50 years, SFWA has pursued this principle, from our fight on J.R.R. Tolkien’s behalf to our insistence that Disney Must Pay. SFWA therefore calls on legislators, regulators, and technologists to ensure that creative workers are adequately compensated when their work is used to train these systems, or else to ensure that their work is not used in the first place.
2. Creators’ contributions to a work must be credited.
One of the difficulties for our community in grappling with these technologies is the lack of transparency we’ve encountered from the companies developing them. Any complex technology has barriers to understanding, but hiding the inputs into a complex system only compounds its problems. SFWA again calls upon legislators, regulators, and technologists to make sure that training sets are clearly disclosed, so that creators may satisfy themselves that their rights have been protected.
That lack of transparency has implications in our community as well for those writers who seek technological assistance in producing work that is genuinely their own. SFWA’s policy has long been that all contributors to written work must be credited, except by explicit contractual agreement. When creators join SFWA as members, they put forward work that they have created, with the expectation that all co-authors have been credited. Likewise, markets that publish fiction typically require all co-authors to be disclosed. We do not see how work generated using a Large Language Model (LLM) trained on the work of thousands of authors can satisfy such a requirement.
We have seen demonstrations in which generative AI tools appear to return text copied from previously published works. We are aware of no current safeguards to absolutely prevent that, short of custom-training one’s own AI tools. With these dangers in mind, we urge authors as a matter of self-protection not to put themselves into a position where they do not know who contributed to a work with their name on it.
3. Creators’ privacy must be protected, especially for unpublished work.
SFWA strongly believes that a thriving community is necessary to writing as a profession. We must have places to privately discuss the industry and the craft, and to exchange our work for criticism or consideration for honors. This is especially true for new writers, who are frequently preyed upon by unscrupulous actors who rely on their lack of experience. To preserve these benefits, online communities, including SFWA’s forums and Discord server, often have strict policies against sharing what other authors post with third parties and thus violating their privacy.
SFWA has become aware of several incidents and trends that threaten these online communities. We wish to be clear that any distribution of material copied from our own online communities to any third party, regardless of purpose, is a violation of that privacy. Once material is entered into another website or API, regardless of what representations are made by those third parties, control over the material has been lost. It may be used for training machine learning and other uncompensated uses and becomes vulnerable to intentional or inadvertent disclosure that could damage an author’s reputation or future ability to sell unpublished work.
Likewise, we urge publishers and agents to exercise great caution with unpublished materials submitted to them for consideration. We are aware of several tools that purport to identify AI-written text. Aside from reported issues of accuracy in their claims, we again stress the loss of control over material submitted to third parties, and the responsibility to live up to the great trust shown by writers when submitting unpublished work.
In all cases involving the handling of a writer’s words and work, whether in online communities or during the submissions process, SFWA strongly believes that the risks inherent in involving third party tools should only be taken with the fully informed consent of the author. If any such tools are shown to have significant false positives, train based on submitted fiction, or experience a data breach, authors must be informed if their work and data has been affected and have some means of seeking redress.
4. Writing and publishing genre fiction is a business with important norms.
Several fiction markets have recently made clear that they will not consider works generated wholly or in part from AI tools. SFWA reminds writers to check submission guidelines every time they submit work to stay current on what that market is considering. Markets may bar work that is AI-written, -developed, or -assisted; unfortunately, there are no widely accepted definitions for these terms. In the absence of such consensus, SFWA urges transparency from all parties: for markets to articulate their terms as clearly as possible and to be forgiving of good-faith corner cases and mistakes, and for writers to be candid about technologies used.
SFWA urges all actors in the speculative fiction marketplace to seek clarity, transparency, and accuracy in navigating the spread and evolution of machine-generated art, and to take the points we’ve laid out into careful consideration. We may occupy different roles in this community, occasionally adversarially, but the spread of machine learning has only underscored just how pivotal this business of speculative fiction has become.
In the interest of clarity and transparency, SFWA is developing a Request for Comment to better define terms and propose common language in this area. We expect the first draft to be available for feedback in the near future, pending the full staffing and consultation of our Emerging Technology Committee.
In light of the concerns discussed in this statement and elsewhere, SFWA’s Board of Directors has instructed its staff and volunteers to avoid seeking out and intentionally using generative AI/ML tools for any internal or publicly published material for SFWA, except for the purpose of discussing these technologies as part of our educational mission. That instruction is not as clear a line as we would prefer, but the rapid spread of these technologies, the lack of transparency into their development and use, and the lack of terminology consensus increases the difficulty of navigating the best path forward. We pledge to our peers in the creative world that we are doing our best not to profit by the misappropriation of their work, and to forgive honest missteps as they return that courtesy to us.
It is not clear yet how the technological and legal landscape will change in the coming months and years. Legislation and court decisions in multiple countries may radically change the situation. Community consensus and standards will evolve as technology develops and we learn more about how it is brought to market. Please take this statement not as our final word on the subject but as part of a vital conversation. Speculative fiction and the world shape each other. The more society needs fiction to understand the world, the more critical it becomes to support the writers who create it.