Washington – The Digital Trust & Safety Partnership (DTSP) released a first-of-its-kind report outlining best practices for the use of artificial intelligence (AI) and automation in trust and safety operations. Developed by DTSP’s AI and Automation Working Group, with assistance from BSR, the report provides valuable guidance on how to leverage AI to address content- and conduct-related abuse on digital products and services.
“AI and automation offer significant potential to keep users safe by detecting, enforcing and reviewing content and conduct to address online abuse at scale,” said DTSP Executive Director David Sullivan. “Automated tools have been a key part of trust and safety from the very beginning, but recent advances in AI, when deployed responsibly, can increase the capacity and accuracy of efforts to counter abusive content and behavior online.”
The report identifies ways that AI can be used to implement the DTSP Best Practices Framework, across product development, governance, enforcement, improvement, and transparency, including potential opportunities for generative AI to further the practices. After exploring limitations and challenges for AI in trust and safety, the report goes on to identify additional best practices and areas where greater future collaboration is needed.
The report is a culmination of extensive research, including interviews with participating companies and generative AI experts. Key points include:
- AI and automation enrich and support the role of trust and safety professionals.
- Companies offering very different digital products and services are aligned on both the challenges and opportunities for integrating AI and generative AI into their trust and safety operations.
- When deployed responsibly, AI and automation can enhance all aspects of trust and safety, from product development, to governance, enforcement, improvement, and transparency. AI can also improve the wellbeing of trust and safety teams and reduce their exposure to traumatic content.
- Generative AI has the potential to significantly increase the capacity and accuracy of content moderation; however, further experimentation and risk assessment and mitigation is needed.
- Best practices that acknowledge the limitations of AI while encouraging collaboration across the trust and safety field provide a path forward on the internet’s toughest challenges.
The best practices adhere to the DTSP Best Practices Framework and were informed by input from the DTSP AI and Automation Working Group. Members of the working group include Bitly, Bumble Inc., Discord, Google, LinkedIn, Match Group, Meta, Microsoft, Pinterest, Reddit, TikTok, and Zoom. The findings reflect the diverse set of products and services offered by these participants.