Arr, me hearties! Methinks Joe Biden's scheme to 'mark the waters' o' AI-made content be a near bout impossible, aye!
2023-09-12
Aye, me hearties! We be yearnin' to harness the power o' AI to tackle our society's grandest troubles. But mark me words, me mateys, 'tis a treacherous journey to etch a 'watermark' upon AI content, a task as tough as a peg-legged pirate walkin' a tightrope!
President Biden recently announced that key AI technology firms, including Amazon, Google, Meta, Microsoft, and OpenAI, have agreed to four commitments after a meeting with executives. These commitments range from enhancing system security and product testing to using AI to solve societal challenges in areas like healthcare. However, watermarking AI-produced content, one of the commitments, may be challenging and raises questions about what constitutes 'AI generated' content and whether technology providers should label such content.Watermarking techniques vary by medium and often rely on a shared secret between the developer and those who detect the watermark. If this secret becomes widely known, AI users can easily remove the watermark. Incorporating watermarking into open-source products is particularly difficult because users can disable or remove the functionality. State actors may also bypass watermarking, limiting its effectiveness in identifying misinformation campaigns.
Another proposed approach is signing content sourced from cameras as legitimate, but this method is also easily overcome. Furthermore, AI content detectors face challenges in identifying content generated using AI tools that don't watermark or content where the watermark has been removed. These detectors can also mistakenly flag legitimate content.
While it is important to identify and combat AI-generated fakes, government regulation in this area raises First Amendment questions and may be ineffective due to technological advancements. Maintaining source credibility and relying on content producers to vet what they publish can be a non-technological solution. Implementing watermarking may prove to be technologically difficult, and the President's goal of using AI to solve societal challenges may be more achievable.
Overall, the battle between developing watermarking techniques and finding ways to bypass them is likely to continue, and regulations must be agile and unharmful to be effective in this evolving field.