Open-Source Artificial Intelligence: At Last, a Clear Definition is Established
In the dynamic realm of artificial intelligence (AI), essential aspects like clarifications tend to get overlooked. The term "artificial intelligence" itself is so vastly applied that it encompasses everything from basic linear regression models to advanced robotic systems capable of lethal force.
However, when it comes to setting guidelines for emerging technologies, precise and concise explanations are crucial. Without them, you end up with amusing-yet-serious predicaments, such as the ones state lawmakers across the nation grapple with, like whether their laws banning deceptive deepfakes extend to spell check.
Thus, although it may not be the most thrilling undertaking, it's significant that, after numerous years of investigation and worldwide debate, the Open Source Initiative (OSI) has finally established a definition for "open source AI" that the organization hopes will serve as a foundation for global regulation.
For an AI system to be tagged as open source under the brand-new definition, its code, weights, and training data must be openly accessible, allowing anyone to utilize it for any purpose, scrutinize its inner workings, modify it, and distribute it to others without seeking consent.
This is a significant departure from the way some tech corporations have utilized the label during the competitive generative AI race.
Notably, Meta's Llama family of models is promoted as open-source because they're free to use (provided developers adhere to the company's license terms) and some of the code is publicly available. Last year, Meta also contributed to forming a lobbying coalition called the AI Alliance to support policies beneficial to its specific type of open-source technologies.
However, Llama models do not meet the criteria to be labeled as open source. This is due to the restrictions their licenses impose on their use in certain commercial applications, like enhancing other sophisticated language models. Furthermore, they explicitly prohibit uses that conflict with various laws or potentially cause harm. Meta has yet to disclose the training data utilized for its Llama models in full.
The OSI's new definition was the result of extensive deliberations between stakeholders from over 35 nations, the organization stated.
"Reaching today’s [open source AI definition] version 1.0 was a challenging process for the OSI community," Open Source Initiative Executive Director Stefano Maffulli said in a statement. "Despite the delicate nature of this process, filled with contrasting viewpoints and uncharted technical territories—and the occasional heated argument—the outcomes align with the goals laid out at the start of this two-year expedition. This is merely a starting point for an ongoing effort to engage with the communities to enhance the definition over time as we progress alongside the broader Open Source community to develop knowledge on how to read and apply OSAID v.1.0."
In the future of tech, the clear definition of open source artificial intelligence (AI) is essential for global regulation. Meticulous guidelines ensure that open source AI systems, such as Meta's Llama family models, adhere to universal standards, allowing for full access to code, weights, and training data.
