Meta and IBM have launched a new group called the AI Alliance to advocate for an “open-science” approach to AI development, positioning themselves against rivals such as Google, Microsoft, and ChatGPT-maker OpenAI.
These two diverging camps – the open and the closed – disagree on whether AI should be developed in a way that makes the underlying technology widely accessible. The debate centers around safety, but also who profits from AI advancements.
Proponents of the open approach favor a non-proprietary and open method, as stated by Darío Gil, a senior vice-president at IBM. The AI Alliance – led by IBM and Meta, and including Dell, Sony, AMD, Intel, several universities, and AI startups – aims to assert that the future of AI will be built on open scientific exchange and open innovation, including open-source technologies.
Meta’s chief AI scientist, Yann LeCun, criticized companies like OpenAI, Google, and Anthropic on social media for “massive corporate lobbying” to shape rules in favor of their powerful AI models. These companies, along with Microsoft, have formed their own industry group, the Frontier Model Forum.
LeCun expressed concerns on X (formerly Twitter) about scientists’ doomsday warnings on AI, fearing they could fuel a ban on open-source research and development.
LeCun argues for open-source platforms to reflect the entirety of human knowledge and culture, especially as AI systems become repositories of such information.
IBM, an early supporter of the open-source Linux OS in the 1990s, sees the dispute as part of a longer-standing competition predating the AI boom.
Chris Padilla, head of IBM’s global government affairs, views opposition to open-source innovation as a classic tactic, likening it to Microsoft’s long-standing model against open-source programs competing with Windows or Office.
“Open-source” refers to a decades-old practice of building software where the code is freely accessible for examination, modification, and building upon.
Open-source AI, more than just code, is debated among computer scientists based on the public availability of technology components and usage restrictions. Some refer to open science as a broader philosophy.
Despite its name, OpenAI’s development of closed AI systems, like ChatGPT and Dall-E, adds to the confusion around open-source AI.
Ilya Sutskever, OpenAI’s chief scientist, acknowledges short-term commercial incentives against open-source but warns of long-term risks involving powerful AI systems too dangerous for public access.
Sutskever illustrated open-source dangers with an AI system capable of starting its own biological laboratory.
Even current AI models pose risks, like escalating disinformation campaigns to disrupt elections, warns David Evan Harris of the University of California, Berkeley.
“Open-source is great in many technology dimensions,” but AI is different, Harris says.
Citing the movie Oppenheimer, Harris underscores the need for caution in sharing scientific discoveries that could fall into the wrong hands.
The Center for Humane Technology, a critic of Meta’s social media practices, highlights the risks of open-source or leaked AI models.
The group’s Camille Carlton deems deploying these models to the public without safeguards as “completely irresponsible.”
An increasingly public debate emerges over the pros and cons of an open-source approach to AI development.
Joe Biden’s sweeping executive order on AI briefly touched on the “open-source” debate.
Biden’s order described open models as “dual-use foundation models with widely available weights,” needing further study. Weights are AI model parameters influencing performance.
Publicly posted weights offer innovation benefits but also security risks, like removing model safeguards, the order stated. Commerce Secretary Gina Raimondo was tasked with developing recommendations by July on managing these risks and benefits.
The European Union, finalizing world-leading AI regulation, is debating provisions, including exemptions for certain “free and open-source AI components” from commercial model rules.