The Biden administration is delving into a controversial discussion surrounding whether the most powerful artificial intelligence systems should be “open source” or closed.
Public Comment on Key AI Components
On Wednesday, the White House announced its intention to gather public feedback on the advantages and drawbacks of making an AI system’s fundamental 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 aimed at regulating the rapidly advancing technology.
Differing Perspectives in Tech Companies
Tech firms are split on the level of openness they should adopt for their AI models. While some highlight the risks associated with widely available AI-model components, others underscore the significance of open science for researchers and startups. Notably, Facebook parent Meta Platforms and IBM have been prominent advocates for an open approach.
Examining Open Models with Widely Available Weights
Biden’s order referred to open models as “dual-use foundation models with widely available weights” and stressed the necessity for further examination. Weights, which are numerical values shaping an AI model’s performance, pose both innovation opportunities and security risks when made public online, according to the order.
Seeking Recommendations on Managing Benefits and Risks
Commerce Secretary Gina Raimondo has been tasked with consulting experts and presenting recommendations to the president by July on how to navigate the potential benefits and risks associated with open AI models. The National Telecommunications and Information Administration, under the Commerce Department, has initiated a 30-day comment period to gather suggestions for an upcoming report.
Finding a Middle Ground
Assistant Commerce Department Secretary Alan Davidson highlighted that experts recognize the complexity of the issue, emphasizing that it is not a black-and-white matter but rather presents various degrees of openness. Davidson expressed optimism in identifying solutions that foster both innovation and safety in the realm of AI technology.
Meta’s Approach to AI Technology Sharing
Meta, formerly Facebook, plans to collaborate with the Biden administration in sharing insights gained from a decade of building AI technologies in an open manner. Nick Clegg, the company’s president of global affairs, expressed the company’s commitment to ensuring the benefits of AI technology are universally accessible.
Google’s Shift Towards Open Models
While Google has traditionally favored a more closed approach to AI development, the tech giant announced the release of a new set of open models named Gemma. Derived from the technology behind its Gemini chatbot app and paid service, Gemma is described as a more lightweight version of the closed Gemini model.
In a recent technical paper, Google emphasized the importance of safety in releasing open models like Gemma due to their irreversible nature. The company called for a collaborative approach within the AI community to evaluate both risks and benefits critically.
The Accessibility Challenge
Although making AI system components publicly available may appear transparent, the reality is that utilizing open models still requires significant resources primarily held by a few major corporations. Cornell University researcher David Gray Widder highlighted this issue, pointing out that accessibility and scrutiny are not straightforward with open models.
Widder also noted that the decision between an open or closed approach to AI development is multifaceted. Companies advocating for open-source models may seek to harness external contributions for profit, while those citing safety as a reason for closed systems may aim to solidify their market position.