An open conversation about disclosing the use of AI: How do we do it and what are the consequences?

Anthropic’s announcement in August that they would add an invisible digital watermark to any content and metadata generated by their AI sparked quite a bit of debate, including in the MERL sector. Would this solve any of the fake “AI slop” content issues? Would this mean that those who use AI would be judged or thought to be producing lower-quality content? Won’t people find ways around it? Or just use other GenAI tools? 

At MTI, we’ve been having great internal conversations about AI disclosure. Our current position is that those working in the MERL and social sectors have a responsibility to disclose and explain their use of AI. Especially in the field of knowledge work, provenance and transparency about methodology, research processes, and data analysis are crucial to maintaining trust and integrity. We believe that transparency about how findings are achieved, how writing and analysis are done, and even how ideas are generated (and by whom!) is indispensable, and that disclosure should include how, where, and why AI was used and additionally how AI outputs were reviewed. 

Internally, we have noted that when a team member uses AI extensively and without sufficient review or disclosure, it often leads to frustration or uncertainty among other team members who are responsible for reading or reviewing this work. 

Others have raised objections about a requirement for disclosure of AI use. They worry that non-native English speakers or those who rely heavily on AI for research and analysis may be discriminated against if they are required to disclose their use of AI, because AI-generated work is sometimes considered to be less trustworthy or of lower quality. Others consider that “everyone is using AI,” so disclosure is a waste of time – we don’t disclose when we use Grammarly or Google, so why is it needed for Generative AI? Yet other concerns are that people might just haphazardly disclose or use AI to write their disclosure, and it will be devoid of meaning.

Join us on October 05, at 12 pm ET for a discussion on the various sides of the AI disclosure debate. While we will touch on the watermark idea, the conversation will be focused on e the place and the value of disclosing AI use for those who choose to use it. We hope to explore questions such as:

• What are some of the reasons for accepting or rejecting disclosure? What are the nuances or variations on positions about disclosure? Will AI disclosure make people reluctant or ashamed of using AI? Or might AI disclosure function as a tool for transparency and trust-building, and a way to explore how professionals in the sector are using it, and where AI is really working, and where it isn’t?
• What are some positive, unintended benefits of more radical AI disclosure in our sector?
• What are some manageable ways to handle disclosure, both in day-to-day AI use and in more formal MERL work? How might it help us to achieve meaningful transparency, accountability, and trust both internally in our organizations and in the wider professional field of MERL? 

We will be joined by speakers:

• Tiago Peixoto is a political scientist with over 20 years of hands-on experience in public sector performance, digital government, participatory democracy, and civic tech. He has written about how, in his perspective, AI disclosure norms may carry a price for users, reproducing unbalanced power dynamics regarding what is seen as trusted, high-quality work.
• Georgia Iacovou is a writer and creative storyteller who has produced research and written work for The Wellcome Trust, The Royal Society, Careful Industries, AWO, Defend Digital Me, Turn2Us, and others. She is also an award-nominated podcast producer with nearly a decade of experience running content calendars and engaging in strategic comms for mission-driven orgs. Clients include The Maybe, Mozilla, The AI Now Institute, and Real ML. She explores the politics of technology in her newsletter “Horrific/Terrific”. 
• Kelly Church has 15 years of experience at the intersection of technology transformation and governance, and organizational change. As a Core Collaborator at MTI, she recently led a participatory process to develop our AI Use and Governance Policy, which, among other things, frames the disclosure of AI use in our work as a necessary practice.