Open-source artificial intelligence is an AI system that is freely available to use, study, modify, and share.[1] These attributes extend to each of the system's components, including datasets, code, and model parameters, promoting a collaborative and transparent approach to AI development.[1]Free and open-source software (FOSS) licenses, such as the Apache License, MIT License, and GNU General Public License, outline the terms under which open-source artificial intelligence can be accessed, modified, and redistributed.[2]
The open-source model provides wider access to AI technology, allowing more individuals and organizations to participate in AI research and development.[3][4] In contrast, closed-source artificial intelligence is proprietary, restricting access to the source code and internal components.[3] Companies often develop closed products in an attempt to keep a competitive advantage in the marketplace.[5] However, some experts suggest that open-source AI tools may have a development advantage over closed-source products and have the potential to overtake them in the marketplace.[5][4]
Popular open-source artificial intelligence project categories include large language models, machine translation tools, and chatbots.[6] For software developers to produce open-source artificial intelligence (AI) resources, they must trust the various other open-source software components they use in its development.[7][8] Open-source AI software has been speculated to have potentially increased risk compared to closed-source AI as bad actors may remove safety protocols of public models as they wish.[4] Similarly, closed-source AI has also been speculated to have an increased risk compared to open-source AI due to issues of dependence, privacy, opaque algorithms, corporate control and limited availability while potentially slowing beneficial innovation.[9][10][11]
There also is a debate about the openness of AI systems as openness is differentiated[12] – an article in Nature suggests that some systems presented as open, such as Meta's Llama 3, "offer little more than an API or the ability to download a model subject to distinctly non-open use restrictions". Such software has been criticized as "openwashing"[13] systems that are better understood as closed.[10] There are some works and frameworks that assess the openness of AI systems[14][12] as well as a new definition by the Open Source Initiative about what constitutes open source AI.[15][16][17] Some large language models are released as open-weight, which means that their trained parameters are publicly available, even if the training code and data aren't.[18][19]
The history of open-source artificial intelligence is intertwined with both the development of AI technologies and the growth of the open-source software movement.[20] Open-source AI has evolved significantly over the past few decades, with contributions from various academic institutions, research labs, tech companies, and independent developers.[21] This section explores the major milestones in the development of open-source AI, from its early days to its current state.
1990s: Early development of AI and open-source software
The concept of AI dates back to the mid-20th century, when computer scientists like Alan Turing and John McCarthy laid the groundwork for modern AI theories and algorithms.[22] An early form of AI, the natural language processing "doctor" ELIZA, was re-implemented and shared in 1977 by Jeff Shrager as a BASIC program, and soon translated to many other languages. Early AI research focused on developing symbolic reasoning systems and rule-based expert systems.[23]
During this period, the idea of open-source software was beginning to take shape, with pioneers like Richard Stallman advocating for free software as a means to promote collaboration and innovation in programming.[24] The Free Software Foundation, founded in 1985 by Stallman, was one of the first major organizations to promote the idea of software that could be freely used, modified, and distributed. The ideas from this movement eventually influenced the development of open-source AI, as more developers began to see the potential benefits of open collaboration in software creation, including AI models and algorithms.[25][26]
In the 1990s, open-source software began to gain more traction,[27] the rise of machine learning and statistical methods also led to the development of more practical AI tools. In 1993, the CMU Artificial Intelligence Repository was initiated, with a variety of openly shared software.[28]
2000s: Emergence of open-source AI
In the early 2000s open-source AI began to take off, with the release of more user-friendly foundational libraries and frameworks that were available for anyone to use and contribute to.[29]
In 2007, Scikit-learn was released.[32] It became one of the most widely used libraries for general-purpose machine learning due to its ease of use and robust functionality, providing implementations of common algorithms like regression, classification, and clustering.[33][34]Theano was also released in the same year.[35]
Rise of open-source AI frameworks (2010s)
Open-source deep learning framework as Torch was released in 2002 and made open-source with Torch7 in 2011, and was later augmented by PyTorch, and TensorFlow.[36][37] These frameworks allowed researchers and developers to build and train neural networks for tasks like image recognition, natural language processing (NLP), and autonomous driving.[38][39]
In 2014, GloVe, a competitor to Word2vec, was released source code under an Apache 2.0 license, documented the datasets they trained on, and released the model weights under a Public Domain Dedication and License.[43]
Open-source generative AI (2020s–Present)
With the announcement of GPT-2, OpenAI originally planned to keep the source code of their models private citing concerns about malicious applications.[44] After OpenAI faced public backlash, however, it released the source code for GPT-2 to GitHub three months after its release.[44] OpenAI has not publicly released the source code or pretrained weights for the GPT-3 or GPT-4 models, though their functionalities can be integrated by developers through the OpenAI API.[45][46]
The rise of large language models (LLMs) and generative AI, such as OpenAI's GPT-3 (2020), further propelled the demand for open-source AI frameworks.[47][48] These models have been used in a variety of applications, including chatbots, content creation, and code generation, demonstrating the broad capabilities of AI systems.[49] At the time of GPT-3's release GPT-2 was still the most powerful open source language model in the world, spurring EleutherAI to train and release GPT-Neo[50] and GPT-J[50][51] in 2021.
In February 2022 EleutherAI released GPT-NeoX-20B, taking back the title of most powerful open source language model in the world from Meta whose FairSeq Dense 13B model had surpassed GPT-J at the end of 2021.[52] 2022 also saw the rise of larger and more powerful models under various non-open source licenses including Meta's OPT[53] and Galactica,[54][55] the BigScience Research Workshop's BLOOM,[56][57] and Tsinghua University's GLM.
During early negotiations in 2021 and 2022 around AI legislation in Europe, proposals were made to avoid over-regulating open-source AI.[58] Noting that some organizations were mis-applying the "open-source" label to their work, in 2022, the Open Source Initiative, which originally came up with the widely accepted standard for open-source software in 1998, started working with experts on a definition of "open-source" that would fit the needs of AI software and models. The most controversial aspect relates to data access, since some models are trained on sensitive data which can't be released. In 2024, they finalized the Open Source AI Definition 1.0 (OSAID 1.0), with endorsements from over 20 organizations.[59][60] It requires full release of the software for processing the data, training the model and making inferences from the model. For the data, it only requires "sufficiently detailed information about the data used to train the system so that a skilled person can build a substantially equivalent system".[60]
In 2023 Llama 1 and 2, MosaicML's MPT,[61][62] and Mistral AI's Mistral and Mixtral models were released.
In 2024, Meta released a collection of large AI models, including Llama 3.1 405B, comparable to the most advanced closed-source models.[63] The company claimed its approach to AI would be open-source, differing from other major tech companies.[63] The Open Source Initiative and others stated that Llama is not open-source despite Meta describing it as open-source, due to Llama's software license prohibiting it from being used for some purposes.[64][65][66]
DeepSeek released their V3 LLM in December 2024, and their R1 reasoning model on January 20, 2025, both as open-weights models under the MIT license.[67][68]
Since the release of OpenAI's proprietary ChatGPT model in late 2022, there have been only a few fully open (weights, data, code, etc.) large language models released. Among these are the OLMo series of models[69][70] released by the Allen Institute for AI.
Ethics
In parallel with the development of AI models, there has been growing interest in ensuring ethical standards in AI development.[71][72] This includes addressing concerns such as bias, privacy, and the potential for misuse of AI systems.[71][72] As a result, frameworks for responsible AI development and the creation of guidelines for documenting ethical considerations, such as the Model Card concept introduced by Google,[73] have gained popularity, though studies show the continued need for their adoption to avoid unintended negative outcomes.[74][75]
Frameworks
The LF AI & Data Foundation, a project under the Linux Foundation, has significantly influenced the open-source AI landscape by fostering collaboration and innovation, and supporting open-source projects.[76] By providing a neutral platform, LF AI & Data unites developers, researchers, and organizations to build cutting-edge AI and data solutions, addressing critical technical challenges and promoting ethical AI development.[77]
As of October 2024, the foundation comprised 77 member companies from North America, Europe, and Asia, and hosted 67 open-source software (OSS) projects contributed by a diverse array of organizations, including silicon valley giants such as Nvidia, Amazon, Intel, and Microsoft.[78] Other large conglomerates like Alibaba, TikTok, AT&T, and IBM have also contributed.[78] Research organizations such as NYU, University of Michigan AI labs, Columbia University, Penn State are also associate members of the LF AI & Data Foundation.[78]
In 2024, while the OSAID was in development, the Linux Foundation was also developing a rubric of components of an AI system, and published a draft Model Openness Framework (MOF).[79] The MOF is a system for evaluating and classifying the completeness and openness of machine learning models. It included three classes of openness, from more open to less open: Class I: Open Science Model; Class II: Open Tooling Model, and Class III: Open Model.[79][80] The Linux Foundation participated in the OSAID development, and OSAID adopted the same rubric of components.[81]
In September 2022, the PyTorch Foundation was established to oversee the widely used PyTorch deep learning framework, which was donated by Meta.[82] The foundation's mission is to drive the adoption of AI tools by fostering and sustaining an ecosystem of open-source, vendor-neutral projects integrated with PyTorch, and to democratize access to state-of-the-art tools, libraries, and other components, making these innovations accessible to everyone.[83]
The PyTorch Foundation also separates business and technical governance, with the PyTorch project maintaining its technical governance structure, while the foundation handles funding, hosting expenses, events, and management of assets such as the project's website, GitHub repository, and social media accounts, ensuring open community governance.[83] Upon its inception, the foundation formed a governing board comprising representatives from its initial members: AMD, Amazon Web Services, Google Cloud, Hugging Face, IBM, Intel, Meta, Microsoft, and NVIDIA.[83]
Open-source AI has assisted in developing and adopting of Large Language Models (LLMs). While proprietary models like OpenAI's GPT series have redefined what is possible in applications such as interactive dialogue systems and automated content creation, fully open-source models have also made significant strides. Google's BERT is an open-source model widely used for tasks like entity recognition and language translation, establishing itself as a versatile tool in NLP.[84] These open-source LLMs have democratized access to advanced language technologies and reduce reliance on proprietary systems.[85]
Hugging Face's MarianMT is an example of Machine Translation, providing support for a wide range of language pairs.[86] Another notable is model OpenNMT.[87] Alongside these open-source models, open-source datasets such as the WMT (Workshop on Machine Translation) datasets including, Europarl Corpus, and OPUS contribute to the sector[88][89] and assist developers to train and fine-tune models for specific languages.[88]
Libraries including OpenCV support real-time computer vision applications, such as image recognition, motion tracking, and facial detection.[90][91] Originally developed by Intel, OpenCV has become one of the most popular libraries for computer vision.[91][90] Other open-source computer vision models including YOLO (You Only Look Once) and Detectron2 also offer similar features.[92][93]
Unlike the previous generations of Computer Vision models, which process image data through convolutional layers, newer generations of computer vision models use Vision Transformers[94] which break down an image into smaller patches to identify which areas of the image are most relevant[94] which generally produces more accurate results.[95]
Open-source artificial intelligence has made a notable impact in robotics by providing a flexible, scalable development environment for both academia and industry.[96] The Robot Operating System (ROS)[97] is an example of a framework used by developers to work across different hardware platforms and robotic architectures.[96]Gazebo is another an open-source robotic simulation software used to test robotic systems in a virtual environment before real-world deployment.[98]
In the healthcare industry, open-source AI has been used in diagnostics, patient care, and personalized treatment options.[99] Open-source libraries have been used for medical imaging for tasks such as tumor detection, improving the speed and accuracy of diagnostic processes.[100][99] Additionally, OpenChem, an open-source library specifically geared toward chemistry and biology applications, enables the development of predictive models for drug discovery, helping researchers identify potential compounds for treatment.[101]
Meta's Llama models, which have been described as open-source by Meta, were adopted by U.S. defense contractors like Lockheed Martin and Oracle after unauthorized adaptations by Chinese researchers affiliated with the People's Liberation Army (PLA) came to light.[102][103] The Open Source Initiative and others have contested Meta's use of the term open-source to describe Llama, due to Llama's license containing an acceptable use policy that prohibits use cases including non-U.S. military use.[66] Chinese researchers used an earlier version of Llama to develop tools like ChatBIT, optimized for military intelligence and decision-making, prompting Meta to expand its partnerships with U.S. contractors to ensure the technology could be used strategically for national security.[103] These applications now include logistics, maintenance, and cybersecurity enhancements.[103]
Benefits
Democratizing access
Open-source AI democratizes access to cutting-edge tools, lowering entry barriers for individuals and smaller organizations that may lack resources.[104] By making these technologies freely available, open-source AI allows developers to innovate and create AI solutions that might have been otherwise inaccessible due to financial constraints, enabling independent developers and researchers, smaller organizations, and startups to utilize advanced AI models without the financial burden of proprietary software licenses.[104] This affordability encourages innovation in niche or specialized applications, as developers can modify existing models to meet unique needs.[104][105]
Collaboration and faster advancements
By sharing code, data, and research findings, open-source AI enables collective problem-solving and innovation.[105] Large-scale collaborations, such as those seen in the development of frameworks like TensorFlow and PyTorch, have accelerated advancements in machine learning (ML) and deep learning.[106]
The open-source nature of these platforms also facilitates rapid iteration and improvement, as contributors from across the globe can propose modifications and enhancements to existing tools.[106][25] Beyond enhancements directly within ML and deep learning, this collaboration can lead to faster advancements in the products of AI, as shared knowledge and expertise are pooled together.[25][105]
Equitable development
The openness of the development process encourages diverse contributions, making it possible for underrepresented groups to shape the future of AI. This inclusivity not only fosters a more equitable development environment but also helps to address biases that might otherwise be overlooked by larger, profit-driven corporations.[107] With contributions from a broad spectrum of perspectives, open-source AI has the potential to create more fair, accountable, and impactful technologies that better serve global communities.[107]
Transparency and obscurity
A video about the importance of transparency of AI in medicine
One key benefit of open-source AI is the increased transparency it offers compared to closed-source alternatives.[108] With open-source models, the underlying algorithms and code are accessible for inspection, which promotes accountability and helps developers understand how a model reaches its conclusions.[14] Additionally, open-weight models, such as Llama and Stable Diffusion, allow developers to directly access model parameters, potentially facilitating the reduced bias and increased fairness in their applications.[14] This transparency can help create systems with human-readable outputs, or "explainable AI", which is a growingly key concern, especially in high-stakes applications such as healthcare, criminal justice, and finance, where the consequences of decisions made by AI systems can be significant (though may also pose certain risks, as mentioned in the Concerns section).[109]
Privacy and independence
A Nature editorial suggests medical care could become dependent on AI models that could be taken down at any time, are difficult to evaluate, and may threaten patient privacy.[9] Its authors propose that health-care institutions, academic researchers, clinicians, patients and technology companies worldwide should collaborate to build open-source models for health care of which the underlying code and base models are easily accessible and can be fine-tuned freely with own data sets.[9]
Concerns
Quality and security
Current open-source models underperform closed-source models on most tasks, but open-source models are improving faster to close the gap.[110]
Open-source development of models has been deemed to have theoretical risks. Once a model is public, it cannot be rolled back or updated if serious security issues are detected.[4] For example, Open-source AI may allow bioterrorism groups like Aum Shinrikyo to remove fine-tuning and other safeguards of AI models to get AI to help develop more devastating terrorist schemes.[111] The main barrier to developing real-world terrorist schemes lies in stringent restrictions on necessary materials and equipment.[4] Furthermore, the rapid pace of AI advancement makes it less appealing to use older models, which are more vulnerable to attacks but also less capable.[4]
In July 2024, the United States released a presidential report saying it did not find sufficient evidence to restrict revealing model weights.[112]
Equity, social, and ethical implications
There have been numerous cases of artificial intelligence leading to unintentionally biased products. Some notable examples include AI software predicting higher risk of future crime and recidivism for African-Americans when compared to white individuals, voice recognition models performing worse for non-native speakers, and facial-recognition models performing worse for women and darker-skinned individuals.[113][107][114]
Researchers have also criticized open-source artificial intelligence for existing security and ethical concerns. An analysis of over 100,000 open-source models on Hugging Face and GitHub using code vulnerability scanners like Bandit, FlawFinder, and Semgrep found that over 30% of models have high-severity vulnerabilities.[115] Furthermore, closed models typically have fewer safety risks than open-sourced models.[4] The freedom to augment open-source models has led to developers releasing models without ethical guidelines, such as GPT4-Chan.[4]
Data quality
There are numerous systemic problems that may contribute to inequitable and biased AI outcomes, stemming from causes such as biased data, flaws in model creation, and failing to recognize or plan for the possibility of these outcomes.[75] As highlighted in research, poor data quality—such as the underrepresentation of specific demographic groups in datasets—and biases introduced during data curation lead to skewed model outputs.[114]
A study of open-source AI projects revealed a failure to scrutinize for data quality, with less than 28% of projects including data quality concerns in their documentation.[75] This study also showed a broader concern that developers do not place enough emphasis on the ethical implications of their models, and even when developers do take ethical implications into consideration, these considerations overemphasize certain metrics (behavior of models) and overlook others (data quality and risk-mitigation steps).[75]
Transparency and "black boxes"
Another key concern with many AI systems with respect to issues such as safety and bias is their lack of transparency.[114][116] Many open-source AI models operate as "black boxes", where their decision-making process is not easily understood, even by their creators.[114][117] This lack of interpretability can hinder accountability, making it difficult to identify why a model made a particular decision or to ensure it operates fairly across diverse groups.[114]
Furthermore, when AI models are closed-source (proprietary), this can facilitate biased systems slipping through the cracks, as was the case for numerous widely adopted facial recognition systems.[114] These hidden biases can persist when those proprietary systems fail to publicize anything about the decision process which could help reveal those biases, such as confidence intervals for decisions made by AI.[114] Especially for systems like those used in healthcare, being able to see and understand systems' reasoning or getting "an [accurate] explanation" of how an answer was obtained is "crucial for ensuring trust and transparency".[118]
Frameworks for improvement
Efforts to counteract these challenges have resulted in the creation of structured documentation frameworks that guide the ethical development and deployment of AI:
Model Cards: Introduced in a Google research paper, these documents provide transparency about an AI model's intended use, limitations, and performance metrics across different demographics.[119][73] They serve as a standardized tool to highlight ethical considerations and facilitate informed usage.[119][73] Though still relatively new, Google believes this framework will play a crucial role in helping increase AI transparency.[73]
Measurement Modeling: This method combines qualitative and quantitative methods through a social sciences lens, providing a framework that helps developers check if an AI system is accurately measuring what it claims to measure. The framework focuses on two key concepts, examining test-retest reliability ("construct reliability") and whether a model measures what it aims to model ("construct validity"). Through these concepts, this model can help developers break down abstract ideas which can't be directly measured (like socioeconomic status) into specific, measurable components while checking for errors or mismatches that could lead to bias. By making these assumptions clear, this framework helps create AI systems that are more fair and reliable.[113]
Datasheets for Datasets: This framework emphasizes documenting the motivation, composition, collection process, and recommended use cases of datasets.[120] By detailing the dataset's lifecycle, datasheets enable users to assess its appropriateness and limitations.[120]
Opening up ChatGPT: tracking openness of instruction-tuned LLMs: A community-driven public resource that evaluates openness of text generation models .[121]
Model Openness Framework: This emerging approach includes principles for transparent AI development, focusing on the accessibility of both models and datasets to enable auditing and accountability.[122]
European Open Source AI Index: This index collects information on model openness, licensing, and EU regulation of generative AI systems and providers. It is a non-profit public resource hosted at Radboud University Nijmegen, the Netherlands.[123]
As AI use grows, increasing AI transparency and reducing model biases has become increasingly emphasized as a concern.[109] These frameworks can help empower developers and stakeholders to identify and mitigate bias, fostering fairness and inclusivity in AI systems.[113][109]
^Thummadi, Babu Veeresh (2021). "Artificial Intelligence (AI) Capabilities, Trust and Open Source Software Team Performance". In Denis Dennehy; Anastasia Griva; Nancy Pouloudi; Yogesh K. Dwivedi; Ilias Pappas; Matti Mäntymäki (eds.). Responsible AI and Analytics for an Ethical and Inclusive Digitized Society. 20th International Federation of Information Processing WG 6.11 Conference on e-Business, e-Services and e-Society, Galway, Ireland, September 1–3, 2021. Lecture Notes in Computer Science. Vol. 12896. Springer. pp. 629–640. doi:10.1007/978-3-030-85447-8_52. ISBN978-3-030-85446-1.
^ abcToma, Augustin; Senkaiahliyan, Senthujan; Lawler, Patrick R.; Rubin, Barry; Wang, Bo (December 2023). "Generative AI could revolutionize health care — but not if control is ceded to big tech". Nature. 624 (7990): 36–38. Bibcode:2023Natur.624...36T. doi:10.1038/d41586-023-03803-y. PMID38036861.
^ abLiesenfeld, Andreas; Lopez, Alianda; Dingemanse, Mark (19 July 2023). "Opening up ChatGPT: Tracking openness, transparency, and accountability in instruction-tuned text generators". Proceedings of the 5th International Conference on Conversational User Interfaces. Association for Computing Machinery. pp. 1–6. arXiv:2307.05532. doi:10.1145/3571884.3604316. ISBN979-8-4007-0014-9.
^Liesenfeld, Andreas; Dingemanse, Mark (5 June 2024). "Rethinking open source generative AI: Open washing and the EU AI Act". The 2024 ACM Conference on Fairness, Accountability, and Transparency. Association for Computing Machinery. pp. 1774–1787. doi:10.1145/3630106.3659005. ISBN979-8-4007-0450-5.
^ abcWhite, Matt; Haddad, Ibrahim; Osborne, Cailean; Xiao-Yang Yanglet Liu; Abdelmonsef, Ahmed; Varghese, Sachin; Arnaud Le Hors (2024). "The Model Openness Framework: Promoting Completeness and Openness for Reproducibility, Transparency, and Usability in Artificial Intelligence". arXiv:2403.13784 [cs.LG].
^Bergstra, J.; O. Breuleux; F. Bastien; P. Lamblin; R. Pascanu; G. Desjardins; J. Turian; D. Warde-Farley; Y. Bengio (30 June 2010). "Theano: A CPU and GPU Math Expression Compiler"(PDF). Proceedings of the Python for Scientific Computing Conference (SciPy) 2010.
^Singh, Kanwar Bharat; Arat, Mustafa Ali (2019). "Deep Learning in the Automotive Industry: Recent Advances and Application Examples". arXiv:1906.08834 [cs.LG].
^Goldman, Sharon (2022-11-18). [venturebeat.com/ai/what-meta-learned-from-galactica-the-doomed-model-launched-two-weeks-before-chatgpt/ "What Meta learned from Galactica, the doomed model launched two weeks before ChatGPT"]. VentureBeat. Retrieved 2025-07-21. {{cite web}}: Check |url= value (help)
^Devlin, Jacob; Chang, Ming-Wei; Lee, Kenton; Toutanova, Kristina (2019-05-24). "BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding". arXiv:1810.04805 [cs.CL].
^ abDosovitskiy, Alexey; Beyer, Lucas; Kolesnikov, Alexander; Weissenborn, Dirk; Zhai, Xiaohua; Unterthiner, Thomas; Dehghani, Mostafa; Minderer, Matthias; Heigold, Georg (2021-06-03). "An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale". arXiv:2010.11929 [cs.CV].
^Khan, Salman; Naseer, Muzammal; Hayat, Munawar; Zamir, Syed Waqas; Khan, Fahad Shahbaz; Shah, Mubarak (2022-01-31). "Transformers in Vision: A Survey". ACM Computing Surveys. 54 (10s): 1–41. arXiv:2101.01169. doi:10.1145/3505244. ISSN0360-0300.
^ abMacenski, Steve; Foote, Tully; Gerkey, Brian; Lalancette, Chris; Woodall, William (2022-05-25). "Robot Operating System 2: Design, Architecture, and Uses In The Wild". Science Robotics. 7 (66): eabm6074. arXiv:2211.07752. doi:10.1126/scirobotics.abm6074. ISSN2470-9476. PMID35544605.{{cite journal}}: CS1 maint: article number as page number (link)
^M, Quigley (2009). "ROS : an open-source Robot Operating System". Proc. Open-Source Software Workshop of the Int'l. Conf. On Robotics and Automation (ICRA), 2009. Archived from the original on 2025-01-21. Retrieved 2024-11-16.
^Kathikar, Adhishree; Nair, Aishwarya; Lazarine, Ben (2023). "Assessing the Vulnerabilities of the Open-Source Artificial Intelligence (AI) Landscape: A Large-Scale Analysis of the Hugging Face Platform". 2023 IEEE International Conference on Intelligence and Security Informatics (ISI). pp. 1–6. doi:10.1109/ISI58743.2023.10297271. ISBN979-8-3503-3773-0.
^Gohel, Prashant; Singh, Priyanka; Mohanty, Manoranjan (12 July 2021). "Explainable AI: current status and future directions". arXiv:2107.07045 [cs.LG].
^ abMitchell, Margaret; Wu, Simone; Zaldivar, Andrew; Barnes, Parker; Vasserman, Lucy; Hutchinson, Ben; Spitzer, Elena; Raji, Inioluwa Deborah; Gebru, Timnit (2018-10-05). Model Cards for Model Reporting. pp. 220–229. arXiv:1810.03993. doi:10.1145/3287560.3287596. ISBN978-1-4503-6125-5.
^ abGebru, Timnit; Morgenstern, Jamie; Vecchione, Briana; Vaughan, Jennifer Wortman; Wallach, Hanna; Daumé III, Hal; Crawford, Kate (2021-12-01). "Datasheets for Datasets". arXiv:1803.09010 [cs.DB].
^White, Matt; Haddad, Ibrahim; Osborne, Cailean; Liu, Xiao-Yang Yanglet; Abdelmonsef, Ahmed; Varghese, Sachin; Hors, Arnaud Le (2024-10-18). "The Model Openness Framework: Promoting Completeness and Openness for Reproducibility, Transparency, and Usability in Artificial Intelligence". arXiv:2403.13784 [cs.LG].