The struggle between open source and proprietary software<\/a> is well understood. But the tensions permeating software circles for decades have shuffled<\/a> into the burgeoning artificial intelligence space, with controversy in hot pursuit.<\/p>\n The New York Times recently published a gushing appraisal<\/a> of Meta CEO Mark Zuckerberg, noting how his \u201copen source AI\u201d embrace had made him popular once more in Silicon Valley. The problem, though, is that Meta\u2019s Llama-branded<\/a> large language models aren\u2019t really open source<\/a>.<\/p>\n Or are they?<\/p>\n By most estimations, they aren\u2019t. But it highlights how the notion of \u201copen source AI\u201d is only going to stir more debate in the years to come. This is something that the Open Source Initiative (OSI<\/a>) is trying to address, led by executive director Stefano Maffulli<\/a> (pictured above), who has been working on the problem for over two years through a global effort spanning conferences, workshops, panels, webinars, reports and more.<\/p>\n The OSI has been a steward of the Open Source Definition<\/a> (OSD) for more than a quarter of a century, setting out how the term \u201copen source\u201d can, or should, be applied to software. A license that meets this definition can legitimately be deemed \u201copen source,\u201d though it recognizes a spectrum of licenses<\/a> ranging from extremely permissive to not quite so permissive.<\/p>\n But transposing legacy licensing and naming conventions from software onto AI is problematic. Joseph Jacks<\/a>, open source evangelist and founder of VC firm OSS Capital<\/a>, goes as far as to say that there is \u201cno such thing as open-source AI<\/a>,\u201d noting that \u201copen source was invented explicitly for software source code.\u201d<\/p>\n In contrast, \u201cneural network<\/a> weights\u201d (NNWs) \u2014 a term used in the world of artificial intelligence to describe the parameters or coefficients through which the network learns during the training process \u2014 aren\u2019t in any meaningful way comparable to software.<\/p>\n \u201cNeural net weights are not software source code; they are unreadable by humans, nor are they debuggable,\u201d Jacks notes. \u201cFurthermore, the fundamental rights of open source also don\u2019t translate over to NNWs in any congruent manner.\u201d<\/p>\n This led Jacks and OSS Capital colleague Heather Meeker<\/a> to come up with their own definition of sorts<\/a>, around the concept of \u201copen weights.\u201d<\/p>\n So before we\u2019ve even arrived at a meaningful definition of \u201copen source AI,\u201d we can already see some of the inherent tensions in trying to get there. How can we agree on a definition if we can\u2019t agree that the \u201cthing\u201d we\u2019re defining exists?<\/p>\n Maffulli, for what it\u2019s worth, agrees.<\/p>\n \u201cThe point is correct,\u201d he told TechCrunch. \u201cOne of the initial debates we had was whether to call it open source AI at all, but everyone was already using the term.\u201d<\/p>\n This mirrors some of the challenges in the broader AI sphere, where debates abound on whether the thing that we\u2019re calling \u201cAI\u201d today really is AI<\/a> or just powerful systems taught to spot patterns among vast swathes of data. But naysayers are mostly resigned to the fact that the \u201cAI\u201d nomenclature is here, and there\u2019s no point fighting it. <\/p>\n Founded in 1998, the OSI is a not-for-profit public benefit corporation that works on a myriad of open source-related activities around advocacy, education and its core raison d\u2019\u00eatre: the Open Source Definition. Today, the organization relies on sponsorships for funding, with such esteemed members as Amazon, Google, Microsoft, Cisco, Intel, Salesforce and Meta.<\/p>\n Meta\u2019s involvement with the OSI is particularly notable right now as it pertains to the notion of \u201copen source AI.\u201d Despite Meta hanging its AI hat on the open-source peg<\/a>, the company has notable restrictions in place regarding how its Llama models can be used: Sure, they can be used gratis for research and commercial use cases, but app developers with more than 700 million monthly users must request a special license from Meta, which it will grant purely at its own discretion.<\/p>\n Put simply, Meta\u2019s Big Tech brethren can whistle if they want in.<\/p>\n Meta\u2019s language around its LLMs is somewhat malleable. While the company did call its Llama 2 model open source<\/a>, with the arrival of Llama 3<\/a> in April, it retreated somewhat from the terminology, using phrases<\/a> such as \u201copenly available\u201d and \u201copenly accessible\u201d instead. But in some places, it still refers to<\/a> the model as \u201copen source.\u201d<\/p>\n \u201cEveryone else that is involved in the conversation is perfectly agreeing that Llama itself cannot be considered open source,\u201d Maffulli said. \u201cPeople I\u2019ve spoken with who work at Meta, they know that it\u2019s a little bit of a stretch.\u201d<\/p>\n On top of that, some might argue that there\u2019s a conflict of interest here: a company that has shown a desire to piggyback off the open source branding also provides finances to the stewards of \u201cthe definition\u201d?<\/p>\n This is one of the reasons why the OSI is trying to diversify its funding, recently securing a grant from the Sloan Foundation<\/a>, which is helping to fund its multi-stakeholder global push to reach the Open Source AI Definition. TechCrunch can reveal this grant amounts to around $250,000, and Maffulli is hopeful that this can alter the optics around its reliance on corporate funding.<\/p>\n \u201cThat\u2019s one of the things that the Sloan grant makes even more clear: We could say goodbye to Meta\u2019s money anytime,\u201d Maffulli said. \u201cWe could do that even before this Sloan Grant, because I know that we\u2019re going to be getting donations from others. And Meta knows that very well. They\u2019re not interfering with any of this [process], neither is Microsoft, or GitHub or Amazon or Google \u2014 they absolutely know that they cannot interfere, because the structure of the organization doesn\u2019t allow that.\u201d<\/p>\n The current Open Source AI Definition draft sits at version 0.0.8<\/a>, constituting three core parts: the \u201cpreamble,\u201d which lays out the document\u2019s remit; the Open Source AI Definition itself; and a checklist that runs through the components required for an open source-compliant AI system.<\/p>\n As per the current draft, an Open Source AI system should grant freedoms to use the system for any purpose without seeking permission; to allow others to study how the system works and inspect its components; and to modify and share the system for any purpose. <\/p>\n But one of the biggest challenges has been around data \u2014 that is, can an AI system be classified as \u201copen source\u201d if the company hasn\u2019t made the training dataset available for others to poke at? According to Maffulli, it\u2019s more important to know where the data came from, and how a developer labeled, de-duplicated and filtered the data. And also, having access to the code that was used to assemble the dataset from its various sources.<\/p>\n \u201cIt\u2019s much better to know that information than to have the plain dataset without the rest of it,\u201d Maffulli said.<\/p>\n While having access to the full dataset would be nice (the OSI makes this an \u201coptional\u201d component), Maffulli says that it\u2019s not possible or practical in many cases. This might be because there is confidential or copyrighted information contained within the dataset that the developer doesn\u2019t have permission to redistribute. Moreover, there are techniques to train machine learning models whereby the data itself isn\u2019t actually shared with the system, using techniques such as federated learning, differential privacy and homomorphic encryption.<\/p>\n And this perfectly highlights the fundamental differences between \u201copen source software\u201d and \u201copen source AI\u201d: The intentions might be similar, but they are not like-for-like comparable, and this disparity is what the OSI is trying to capture in its definition.<\/p>\n In software, source code and binary code are two views of the same artifact: They reflect the same program in different forms. But training datasets and the subsequent trained models are distinct things: You can take that same dataset, and you won\u2019t necessarily be able to re-create the same model consistently.<\/p>\n \u201cThere is a variety of statistical and random logic that happens during the training that means it cannot make it replicable in the same way as software,\u201d Maffulli added.<\/p>\n So an open source AI system should be easy to replicate, with clear instructions. And this is where the checklist facet of the Open Source AI Definition comes into play, which is based on a recently published academic paper<\/a> called \u201cThe Model Openness Framework: Promoting Completeness and Openness for Reproducibility, Transparency, and Usability in Artificial Intelligence.\u201d<\/p>\n This paper proposes the Model Openness Framework (MOF), a classification system that rates machine learning models \u201cbased on their completeness and openness.\u201d The MOF demands that specific components of the AI model development be \u201cincluded and released under appropriate open licenses,\u201d including training methodologies and details around the model parameters. <\/p>\n The OSI is calling the official launch of the definition the \u201cstable version,\u201d much like a company will do with an application that has undergone extensive testing and debugging ahead of prime time. The OSI is purposefully not calling it the \u201cfinal release\u201d because parts of it will likely evolve.<\/p>\n \u201cWe can\u2019t really expect this definition to last for 26 years like the Open Source Definition,\u201d Maffulli said. \u201cI don\u2019t expect the top part of the definition \u2014 such as \u2018what is an AI system?\u2019 \u2014 to change much. But the parts that we refer to in the checklist, those lists of components depend on technology? Tomorrow, who knows what the technology will look like.\u201d<\/p>\n The stable Open Source AI Definition is expected to be rubber stamped by the Board at the All Things Open conference<\/a> at the tail end of October, with the OSI embarking on a global roadshow in the intervening months spanning five continents, seeking more \u201cdiverse input\u201d on how \u201copen source AI\u201d will be defined moving forward. But any final changes are likely to be little more than \u201csmall tweaks\u201d here and there.<\/p>\n \u201cThis is the final stretch,\u201d Maffulli said. \u201cWe have reached a feature complete version of the definition; we have all the elements that we need. Now we have a checklist, so we\u2019re checking that there are no surprises in there; there are no systems that should be included or excluded.\u201d<\/p>\n<\/div>\nAI ain\u2019t software code<\/h2>\n
![\"Llama](\"https:\/\/techcrunch.com\/wp-content\/uploads\/2024\/06\/GettyImages-959993436-e1718640411389.jpg?w=680\")
Working definition of open source AI<\/h2>\n
![\"Concept](\"https:\/\/techcrunch.com\/wp-content\/uploads\/2024\/06\/GettyImages-1383744480-e1718635120202.jpg?w=680\")
Stable condition<\/h2>\n
![\"Stefano](\"https:\/\/techcrunch.com\/wp-content\/uploads\/2024\/06\/Stef-OSAID-e1718873379686.jpg?w=680\")