The ongoing discussion within the Biden administration revolves around the debate on whether the most advanced artificial intelligence systems should follow an “open source” or closed model.
In an effort to gather public input, the White House announced its interest in exploring the advantages and drawbacks of making a key AI system’s components publicly accessible for modification and use by anyone. This initiative is part of a broader executive order signed by President Joe Biden in October to oversee the rapidly evolving technology landscape.
Various tech companies hold differing stances on the level of openness for their AI models. Some highlight the risks associated with widely available AI components, while others emphasize the importance of open science for researchers and startups. Notably, Facebook’s parent company Meta Platforms and IBM have been vocal proponents of an open approach.
President Biden’s order specifically referenced “dual-use foundation models with widely available weights” as open models requiring further examination. Weights, which are numerical values influencing an AI model’s performance, raise concerns when publicly shared online due to potential innovation benefits as well as security risks from the removal of protective measures within the model.
Commerce Secretary Gina Raimondo has been tasked to consult with experts and present recommendations on managing the benefits and risks associated with open AI models by July. Additionally, the Commerce Department’s National Telecommunications and Information Administration is inviting public comments over a 30-day period to contribute ideas for an upcoming report to the president.
Alan Davidson, an assistant Commerce Department secretary, highlighted the nuanced nature of the issue, indicating that solutions can be found to promote both innovation and safety. Meta Platforms intends to share its insights on building AI technologies openly with the Biden administration to ensure widespread benefits from AI advancements.
Google, known for its more closed approach, recently introduced Gemma, a group of open models stemming from the technology behind its Gemini chatbot app. These open models are positioned as a lighter version of the closed Gemini. Google emphasized safety in releasing open models like Gemma, calling for a collaborative evaluation of risks and benefits within the AI community.
Despite the accessibility of open AI components, the utilization of such models may still require substantial resources primarily controlled by a few major companies, according to Cornell University researcher David Gray Widder. The decision for companies to adopt open or closed strategies is multifaceted, with some aiming to leverage external contributions for profit, while others prioritize safeguarding their AI systems to maintain their competitive edge.
