AI-generated content will be watermarked to prevent deepfakes and misinformation.

2023-07-24

Including seven companies such as OpenAI, Microsoft, Google, Meta, Amazon, Anthropic, and Inflection have pledged to develop technology to provide clear watermarks for AI-generated content. The Biden administration hopes that this will help to securely avoid questioning the authenticity of AI-generated text, videos, audios, and images when sharing them. The working principle of the watermark is currently unclear, but it may be embedded in the content, allowing users to trace the source of the AI tools used for its generation. Deepfake technology has become a concern for internet users and policymakers, and tech companies are also striving to address the controversial use of AI tools. Earlier this year, the image generator Midjourney was used to create fake images of Donald Trump's arrest, which quickly went viral. Although these images were obviously manipulated to many people, Midjourney decided to take action to ban users who created these images. Perhaps if watermark technology was available at that time, Eliot Higgins, the founder of Bellingcat, would not have faced such serious consequences for claiming that he was not trying to cleverly create illusions but was just playing around with Midjourney. However, there are even more serious abuses of AI tools, and watermark technology may help protect some internet users from harm and disputes. Earlier this year, there were reports that AI voice generation software was used to defraud people of thousands of dollars. Just last month, the FBI warned that more and more malicious actors are using AI-generated deepfakes in extortion schemes. The White House states that watermark technology will enable "creative work with the help of AI to flourish while reducing the risks of fraud and deception." OpenAI stated in its blog that it has agreed to "develop robust mechanisms, including source and/or watermark systems for audio or visual content" and "systems or APIs to determine if a piece of content was created using their tools." This applies to most AI-generated content, but there will be some exceptions, such as the default voice of AI assistants not being watermarked. OpenAI stated, "Content that can be easily distinguished from reality or is designed to be generated by our AI systems, such as the default voice of AI assistants, is not covered by this commitment." Google stated that in addition to watermark technology, they will also "integrate metadata" and "other innovative technologies" to "promote trustworthy information." With growing concerns about the misuse of AI, President Joe Biden will meet with tech companies today. This will help President Biden and Congress gain important insights for issuing executive orders and bipartisan legislation to regain control over rapidly advancing AI technology. In a blog post, Microsoft praised the Biden administration's actions, "laying the foundation for ensuring that the promise of AI always exceeds the risks," and "bringing the tech industry together to develop concrete measures to make AI safer and more beneficial to the public." In addition to developing watermarks for AI-generated content, tech companies have made other commitments in a series of voluntary pledges announced by the White House last week. Among them, tech companies have agreed to conduct internal and external testing of AI systems before deployment. They also expressed willingness to invest more in cybersecurity and share information across the industry to help mitigate AI risks. These risks include biases or discrimination brought by AI and obstacles to the development of advanced weapons, as mentioned in OpenAI's blog. Microsoft's blog emphasized its other commitments to the White House, including supporting the establishment of a national registry for high-risk AI systems. OpenAI stated that these commitments by tech companies are "an important step towards meaningful and effective AI governance, both in the United States and globally." The company, which develops ChatGPT, GPT-4, and DALL-E 2, also pledged to "invest in research in areas that can help shape regulatory policies, such as technologies to assess the potential dangers of AI models." Meta's Global Affairs President Nick Clegg agrees with OpenAI's viewpoint, calling the commitments by tech companies "an important first step in ensuring that AI builds responsible protective barriers." Google describes these commitments as a "milestone that unites the entire industry to ensure that AI benefits everyone." Source: https://arstechnica.com/ai/2023/07/openai-google-will-watermark-ai-generated-content-to-hinder-deepfakes-misinfo/

RECENT AI NEWS

RECENT AI TOOLS