The U.S. Supreme Court recently struck down the “Chevron deference,” a 40-year-old ruling that required courts to defer to federal agencies’ interpretations of congressional laws. This ruling allowed agencies to make their own rules when Congress left aspects of its statutes ambiguous. However, with the elimination of Chevron deference, courts will now have to exercise their own legal judgment. This decision could have wide-reaching effects, making it harder for Congress to pass nationwide AI regulations.
Prior to this ruling, Congress was already struggling to pass a basic AI policy framework. State regulators on both sides of the aisle felt the need to step in due to the lack of progress. With the elimination of Chevron deference, any regulation that Congress writes will need to be highly specific to survive legal challenges. This poses a significant challenge given the rapid and unpredictable nature of the AI industry.
During oral arguments, Justice Elena Kagan questioned whether courts or agencies should fill the gaps left by Congress in an artificial intelligence bill. With the elimination of Chevron deference, courts will now be responsible for filling those gaps. This could lead federal lawmakers to believe that passing AI bills is futile, resulting in a lack of regulation for the industry.
In other news, Google’s Environmental Report fails to address the energy usage of its AI, which is known to be power-hungry. Figma, a design tool company, has temporarily disabled its “Make Design” AI feature due to concerns that it was copying designs from Apple’s Weather app. Meta, formerly known as Facebook, has changed its “Made with AI” label to “AI info” after photographers complained about mistakenly applied labels. The company is distributing robot animals to combat loneliness among the elderly in New York state. Apple is also expanding its use of AI, planning to bring AI features to its Vision Pro mixed-reality headsets.
A research paper from Northeastern University explores how text-generating models like OpenAI’s GPT-4o process and generate human-sounding text. The researchers found that these models develop an implicit vocabulary that maps groups of tokens to semantically meaningful “units.” This discovery could be a significant step towards understanding how lexical representations form in AI models.
Meta has trained models to create 3D assets from text descriptions, which can be used in projects like apps and video games. These models, called AssetGen and TextureGen, generate high-quality 3D shapes with textures in less than a minute. Meta plans to incorporate tools like these into its metaverse game development efforts.
Apple may secure an observer seat on OpenAI’s board as part of their partnership. This move would give Apple significant power, considering its plans to integrate OpenAI’s chatbot platform, ChatGPT, into its devices. In an interesting twist, Apple is reportedly not paying OpenAI for the integration, arguing that the PR exposure is more valuable. Microsoft, a major investor in OpenAI, could end up subsidizing Apple’s ChatGPT integration without receiving much in return.
Overall, the elimination of Chevron deference by the U.S. Supreme Court makes it significantly harder to regulate AI in the United States. Congress will need to write highly specific regulations to overcome legal challenges. Meanwhile, various companies are making advancements in AI technology and grappling with issues such as energy usage, design infringement, and loneliness among the elderly. The research community is also working to understand how AI models process and generate text, as well as developing new techniques for creating 3D assets from textual prompts.
