Impact of open-source software on AI Policies
Disclaimer: First published on medium.com by Sanjay Basu on 12/29/2021
From research to ethics, open-source code is playing a central role in the developing use of AI, yet there is a consistent absence of open-source developers from policy discussions. This piece is to highlight how open-source software impacts AI ecosystem. This essay is based on a report published by Brookings Institution.
Open-source software (OSS) affects nearly every issue in AI policy, but it is often absent from discussions around AI policy. OSS helps to make AI tools more widely available and easier to use, and enables the development and analysis of big data.
The most advanced tools for machine learning are often free and publicly available. The relationship between OSS and AI policy is less acknowledged and a European Parliament report, that I read sometime back, does not address the relationship at all.
OSS affects nearly every issue in AI policy as it allows for faster adoption of AI in science and industry and accelerates proliferation of ethical AI practices. OSS enables AI adoption by reducing the level of mathematical and technical knowledge necessary to use AI. OSS can be collaborative, competitive, and result in high quality code. This is especially important because many data scientists do not have the mathematical training necessary to implement highly complex machine learning algorithms. Well-written open-source AI code significantly expands the capacity of the average data scientist and enables economic growth.
Open-source AI tools can help data scientists detect and mitigate AI bias. Tools like IBM’s AI Fairness 360, Microsoft’s Fairlearn, and the University of Chicago’s Aequitas can help data scientists better understand the inner workings of their models. Government should fund OSS for ethical AI as a different lever to improve AI’s role in society. Funding for OSS for ethical AI is vital to its development and widespread adoption.
Even more than technology companies, scientific researchers benefit from open-source AI. For instance, the Open-Source AI project Keras is used to identify subcomponents of mRNA molecules and build neural interfaces to better help visually impaired people see.
In 2007, several researchers argued that the lack of openly available algorithmic implementations is a major obstacle to scientific progress. Open-source software is still very important in the sciences and policymakers should continue to encourage it. OSS software can be used for reproducible research, because the same code can be used by many different researchers. This makes it easier to evaluate research results, even when small changes are made to the algorithm. OSS has significant ramifications for competition policy, too. Open-source machine learning tools will likely enable more AI adoption, but will not stop the growing influence of the largest technology companies.
While the open sourcing of Google and Facebook’s deep learning tools, Tensorflow and PyTorch, may increase transparency and make it more accessible to the public, this open sourcing may further entrench Google and Facebook in their already fortified positions.
Google and Facebook are gaining influence over the AI market by dominating open-source deep learning tools through Tensorflow and PyTorch, while other open-source AI tools are losing significance.
The rapidly emerging field of AI presents a challenge for international bodies such as ISO, IEEE, CEN-CENELEC, NIST, and many others. In other industries, standards bodies have tried to disseminate best practices, but the machine learning world already has a diverse ecosystem of OSS. To encourage consistency and interoperability, standards may have to make a significant investment in this industry. Standards bodies may be reluctant to cede influence to Google and Facebook in the development of deep learning methods, but the large tech companies are engaged and exerting influence through the standards bodies.