What does open source AI actually mean? Finally there is a definition

In the busy world of artificial intelligence, boring things like definitions often get overlooked. The term “artificial intelligence” itself is used so broadly that it can refer to everything from linear regression models to killer robots.

But when it comes to regulating new technologies, clear and precise definitions are important. Without them, you’ll end up with those stupid “if only it weren’t so serious” debates that state legislators across the country are having, such as whether the language they wrote to ban deceptive deepfakes would also apply to spell checking.

So while it may not be the most sexy project, it’s notable that after years of research and global debate, the Open Source Initiative has finally agreed on a definition of “open source AI” that the nonprofit hopes can become basis of international regulation.

To achieve open source status under new definitionThe AI ​​system, including its component code, weights, and training data, should be made available for free in a way that allows anyone to use it for any purpose without permission, learn how it works, modify it, and share it with others. other.

This is quite a change from how some tech companies have used the label amid the generative AI arms race.

In particular, Meta advertises The Llama family of models is open source in that they are free to use (as long as developers comply with the company’s license terms) and some of the code is publicly available. Last year Metal also helped create a lobbying coalition called the AI ​​Alliance that advocates for policies that benefit its particular brand of open source technology.

Llama models I don’t meet the requirements as open source, since their licenses still place restrictions on how they can be used for some commercial purposes, such as improving other large language models. They also strictly prohibit uses that may violate various laws or cause harm. Meta also has not fully disclosed training data for its Llama models.

The new definition of the Open Source Initiative is the result of years of discussions between stakeholders from more than 35 countries, the organization said.

“Reaching today’s version 1.0 (the open source definition of AI) has been a difficult journey filled with new challenges for the OSI community,” Executive Director of the Open Source Initiative. This is stated in a statement by Stefano Maffulli.. “Despite this delicate process, filled with differing opinions and uncharted technical boundaries, as well as occasional heated discussions, the results are consistent with the expectations set out at the beginning of this two-year process. This is a starting point for further community engagement efforts to improve the definition over time as we work with the broader open source community to develop the knowledge to read and use OSAID v.1.0.”