SANTA MONICA, CA, JANUARY 8, 2025: Smoke from nearby wildfires fills the sky like Kane stunts … [+]
AI is in equilibrium. The use of open source technologies is influencing the development of artificial intelligence to an almost equal (not greater) extent than proprietary closed source technologies that do not enjoy community distribution and user access in the same way as their counterparts. their open.
This balancing act needs leveling, or at least a formalization of weights and measures, so that the data science engineering teams and software application developers now building AI models know where they stand. . The standardization that the industry hopes will tip the scales comes from the Open Source Alliance as it now unveils the draft Open Weight Definition (OWD) as a milestone for the mature AI industry.
OSA says this definition will “bridge the gap” between closed and open source AI models by allowing users to download and deploy advanced AI technologies independently without charge. This open promise is said to apply to all users, no matter who they are—and no matter what their fields of endeavor—and exists as an invitation without the need to ask for permission. This evolution reflects the growing need for clarity in language as, obviously, AI becomes more and more prevalent in enterprise IT groups in the public and private sector arenas.
A Pragmatic Approach
By not requiring access to the components required for the expected reproducibility of open source, the Open Weight Definition protects two of the four core free software freedoms: the ability to use and share, but not necessarily study or modify, a model . This is hoped to lower the barrier to entry and provide a level of flexibility and access for vendors who are not yet able to meet the open source definition that covers all four.
“In pursuit of better global collaboration across open source topographies, the definition of open weights is in line with the sharing of AI,” said Amanda Brock, CEO at OpenUK. “It’s critical that we define levels of openness across the separate but critical components of AI, whether that’s data, algorithmic weights, or modeling. We have seen this approach favored by Stamford University and Radboud and it certainly seems to be more practical and workable than a small group creating a definition that is not fit for purpose.”
Brock reminds us that the Open Sourve Initiative is “at the beginning of the journey” with the definition of the Open Source AI Definition, or OSAID. To her mind, the move OSA has made shows that the approach of trying to “define open source AI” is wrong.
“Instead, we need to take this compartmentalized approach to the challenge and look at the underlying technology, including the training data and what it means to be open.” Open source doesn’t define the law, and it shouldn’t. It’s about enabling anyone to use the ‘resource’ of technology including data for any purpose,” advised Brock. “But this is subject to the law. And if the laws—whether privacy or otherwise, or contractual relationships—mean that an element of data cannot be opened, that is frankly irrelevant to any attempt to define openness.”
She emphasizes her position on this point and insists that we should not define openness in any way by trying to second-guess legislation from either one or several countries. For Brock, the definition of “fully open” can mean that there are various legal or other choices that create qualifiers, especially for data, which means there is also a level of partial openness.
Complicated exchanges
Although not guaranteed by the definition, users can still enjoy more limited opportunities to study and modify a model, for example by observing its results for given inputs, or by tuning new data accordingly.
They may face challenges fully addressing the ethical issues of fairness and bias inherent in the data, and will not be able to fully add or remove data already processed during training, or retrain or re-architect the model entirely. For many applications this is an acceptable compromise. Similarly, proprietary software is still widely used today, despite the large and growing open source industry that supports modern society.
“Today, open weight models are indispensable tools for open innovation, allowing anyone to download and deploy cutting-edge AI models independently,” said Sam Johnston, convener of the Open Source Alliance. “We chose to base the draft Open Weight Definition on the tried and tested Open Source Definition because it is aimed at vendors who aspire to label their products as open source but are not yet willing or able to provide the data – and for AI, the data is the source”.
The open weight definition clarifies any trade-offs by emphasizing clear labeling and responsible use, ensuring users understand the limitations of these models and the freedoms they offer. This distinction is argued to be critical, as the terms “open source” and “open weight” have—in the past—been used interchangeably, despite their significant differences. The term “open source” is used to describe models closed under proprietary licenses that share limited or no data.
Ask the audience
By introducing this definition, distributed models without essential replication elements, including training data, can be conveniently described, bringing clarity to what is expected from everyone in the AI ecosystem.
The draft Open Weight Definition is now open for public consultation.
Although the movement here may appear—at least to the outside observer—to represent something of a techie enclave in itself, the efforts being made here are steeped in altruistic openness and a systemic meritocracy based on utility, functionality, and freedom of use. . Yes, proprietary AI tools, engines, models, and entire organizations will continue to exist, but the efforts here are designed to help all entities live with each other on the same planet, or at least the same hardware and user interface.