Biden is embroiled in a debate about whether the most powerful AI systems should be "open source" or closed.
The White House said Wednesday that it was seeking public comment on the risks and benefits of making key components of AI systems publicly available for anyone to use and modify. The investigation is part of a broader executive order signed by Joe Biden** in October to govern rapidly evolving technology.
Tech companies are divided on how open AI models are, with some highlighting the dangers of widely used AI model components, while others emphasize the importance of open science for researchers and startups. Facebook's parent company, Meta Platforms and IBM, are the strongest enablers of the open approach.
Biden's order describes open models, which have the technical name "dual-use base model with widely available weights" and says they require further study. Weights are numerical values that affect the performance of an AI model.
When these weights are publicly released on the internet, Biden's order said, "could bring huge benefits to innovation, but also huge security risks, such as the removal of safeguards within the model." He asked Commerce Secretary Gina Raimondo to speak with experts by July and offer advice on how to manage potential benefits and risks.
Now, the Department of Commerce's National Telecommunications and Information Administration has said it will also open a 30-day comment period to gather ideas that will be included in the report submitted to **.
The encouraging news is that experts are clear that this is not a binary issue. There is a gradient in the degree of openness. Alan D**Idson, assistant secretary of commerce and NTIA administrator, said. Davidson told reporters Tuesday that it is possible to find solutions that promote innovation and safety.
According to a written statement from Nick Clegg, Meta plans to share with Biden "what we've learned over the past decade building AI technology in an open way so that everyone can continue to share in the benefits of AI." President of Corporate Global Affairs.
Google has largely leaned towards a more closed approach, but on Wednesday released a new set of open models called Gemma, which are derived from the same technology used to create its recently released Gemini chatbot app and paid service. Google describes the open model as a more "lightweight" version of the larger, more powerful Gemini, which is still closed.
Google said in a technical statement on Wednesday that it prioritizes security due to the "irreversibility" of releasing open models such as GEMMA and urged "the broader AI community to move beyond the simplistic 'open vs. closed' debate and avoid exaggeration." or minimizing potential harm, as we believe it's critical to take a nuanced, collaborative approach to risk and benefit. ”
Simply releasing components of an AI system to the world doesn't necessarily make it easily accessible or censored by outsiders, according to Cornell University researcher D**ID, because using an open model still requires "resources to be concentrated in the hands of a few large companies" Gray Wader.
Vader said the motivations for companies to take a more open or closed approach are also complex. Widder says that those who support open source lobbyists may want to profit from outside contributions, while those who believe that security concerns force them to tightly protect AI systems may also want to cement their position as early movers.