Open Weight Definition Adds Balance To Open Source AI Integrity

AI is in balance. The use of open source technologies is impacting the development of artificial intelligence in somewhere near equal (of not greater) measure than closed source proprietary technologies that do not enjoy community distribution and user access in the same way as their open counterparts.

This balancing act needs leveling off, or at least it needs a formalization of weights and measures so that data science engineering teams and software application developers now creating AI models know where they stand. The standardization that the industry hopes will level the scales comes from the Open Source Alliance as it now unveils the draft Open Weight Definition (OWD) as a milestone for the maturing AI industry.

The OSA says that 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 openness promise is said to apply to all users regardless of who they are - and regardless of their fields of endeavor - and exists as an invitation without the need to ask for permission. This evolution reflects the increasing need for clarity in language as, obviously, AI becomes increasingly pervasive in enterprise IT stacks across public and private sector arenas.

A Pragmatic Approach

By not demanding access to the components required for the reproducibility expected of open source, the Open Weight Definition protects two of the four essential freedoms of free software: the ability to use and share, but not necessarily to study or modify a model. This is hoped to lower the barrier to entry and provide a level of flexibility and accessibility to vendors not yet able to satisfy the open source definition covering all four.

“In the pursuit of better global collaboration across open source topographies, defining open weights is in line with the disaggregation of AI,” said Amanda Brock, CEO at OpenUK. “It is critical that we define levels of openness across disaggregated but critical components of AI, whether that be data, algorithmic weight or model. We've seen this approach favoured by Stamford and Radboud University and it certainly seems to be more practical and workable than a small group creating a definition that isn't fit for purpose.”

Brock reminds us that the Open Sourve Initiative is "at the start of the journey" with the definition of the Open Source AI Definition or OSAID. To her mind, the move that the OSA has made shows that the approach of trying to “define open source AI" is the wrong one.

“Rather we should follow this disaggregated 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 law and it should not. It's about what enables anyone to use the technology's ‘source’ including data for any purpose,” advised Brock. “But this is subject to law. And if laws - whether privacy or otherwise, or contractual relationship - mean an element of the data cannot be opened up, that frankly is irrelevant to any attempt to define openness.”

She emphasizes her stance on this point and insists that we should not be defining openness in any form, trying to second guess legislation whether from any one or multiple countries. For Brock, the definition of “fully open” may mean that there are different legal or other choices creating qualifiers, particularly for data, meaning that there is also a level of partial openness.

Tricky Trade-Offs

While not guaranteed by the definition, users can still enjoy more limited opportunities to study and modify a model, for example by observing its outputs for given inputs, or by fine-tuning on new data respectively.

They may face challenges fully addressing ethical issues of fairness and bias inherent in the data and would be unable to effectively add or fully remove data already processed during training, or to re-train or re-architect the model entirely. For many applications this is an acceptable compromise. Similarly, proprietary software is still used extensively today despite the large and growing open source industry underpinning 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, convenor of the Open Source Alliance. "We opted to base the draft Open Weight Definition on the tried and tested Open Source Definition because it addresses vendors who aspire to label their products as open source but are not yet willing or able to deliver the data — and for AI, the data is the source."

The Open Weight Definition clarifies any trade-offs by placing emphasis on clear labeling and responsible use, ensuring that users understand the limitations of these models and the freedoms they provide. 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" has been used to describe closed models under proprietary licenses that share limited data, or none.

Ask The Audience

By introducing this definition, models distributed without reproduction essentials including training data can be described appropriately, bringing what is hoped by all to be clarity to the AI ecosystem.

The draft Open Weight Definition is now open for public consultation.

Although the movement here might appear - to the external observer at least - to represent something of an enclave of techies in and of itself, the efforts being made here are steeped in altruistic openness and a system meritocracy founded in usefulness, 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 device and user interface.