In an era where AI-generated misinformation is increasingly infiltrating our digital spaces, Microsoft has unveiled a strategic plan aimed at establishing authenticity online. The proposal, shared with MIT Technology Review, comes from Microsoft’s AI safety research team, which has been investigating the effectiveness of current methods for documenting digital manipulation. This initiative is particularly relevant in light of recent incidents, such as the dissemination of altered images and videos that mislead the public, including those targeting the ongoing conflict in Ukraine.
To illustrate Microsoft’s vision, consider the analogy of verifying a Rembrandt painting’s authenticity. Just as art experts might provide detailed provenance records, apply invisible watermarks, and use digital scans to create a mathematical signature of the artwork, Microsoft suggests implementing similar practices to verify online content. The team rigorously evaluated various combinations of these techniques to determine which methods could withstand common manipulation tactics, such as metadata removal and content alteration. The goal is to identify robust combinations that platforms can confidently use to provide users with clearer insights about the nature of the content they encounter.
Microsoft’s Chief Scientific Officer, Eric Horvitz, noted that the company’s initiative is partly a response to emerging legislation like California’s AI Transparency Act, which takes effect this August. While the company acknowledges the inherent limitations of these verification tools—they do not assess the truthfulness of content, only whether it has been manipulated—Horvitz emphasizes the importance of providing users with transparency about content origins. Experts in the field, such as UC Berkeley’s Hany Farid, believe that adopting Microsoft’s proposed standards could significantly reduce the prevalence of misleading content, although it may not eliminate the problem entirely. As major tech platforms navigate the balance between misinformation and user engagement, the effectiveness of Microsoft’s recommendations will ultimately depend on their willingness to prioritize transparency over profitability.
Source: Microsoft has a new plan to prove what’s real and what’s AI online via MIT Technology Review