IBM’s open source strategy champions AI trust and transparency
What do railroad tracks, ink-jet printers, and nutrition labels have to do with open source technologies and AI? Each helps IBM® define its commitment to open source, a 25-year heritage that is at the core of developing AI with trust and transparency. More recently, IBM has ramped up initiatives in open sourcing technologies that support fair and unbiased AI.
Open source is in the DNA of IBM. Long before our 2019 acquisition of Red Hat, a global leader in open source technology, IBM helped establish The Linux Foundation, The Apache Software Foundation and Eclipse Foundation. We contributed software projects for the open source community to host. We championed open governance and standards, and we advocated for public collaboration and transparency. This decades-long history is actually one of our best-kept secrets, and it helps unlock the “black box of AI” in a modern era when trust is a top concern for many companies considering an AI investment.
According to Mike Hind, Distinguished Research Staff Member at IBM Research, two key objectives lie at the heart of the open source culture at IBM:
- Accelerate science: Scientific advancement builds on results achieved by others. When these results are provided in an open source system, other researchers can focus on advancing science and avoid the burden of re-implementing the software described in research papers.
- Democratize the industry: Providing software building blocks allows people to develop more sophisticated projects on top of already-established standards, reducing redundancies.
“Think of it like a railroad,” says Hind. “If the tracks are already there, you can focus on building better trains. You’re not starting from scratch.”
Although a company doesn’t earn revenue directly from open source technology, it can benefit from creating an ecosystem that uses open source tools. Similar to how an ink-jet printer can create a new market for ink cartridges, this ecosystem can open new business opportunities such as consulting services and related offerings.
IBM’s open source strategy for trusted AI
So how does IBM embrace open source — and what is the connection to trust and transparency in AI? Mike Hind details a combination of software and accessible educational materials that form the cornerstone of IBM’s trusted AI open source strategy:
- Github repositories containing source codes that provide software building blocks for the open source community
- Three projects on trusted AI that were contributed to and hosted at the LF AI Foundation. “These state-of-the-art open source toolkits provide true multi-vendor open governance,” says Jim Spohrer, PhD, Director of IBM Research Cognitive Opentech Group.
- The AI Fairness 360 Toolkit helps data scientists easily detect and mitigate bias in machine learning models and datasets.
- The Adversarial Robustness 360 Toolbox allows researchers and developers to secure AI systems by detecting and simulating adversarial attacks on deep neural networks.
- The AI Explainability 360 Toolkit provides diverse explainability algorithms for machine learning models.
- AI FactSheets that provide transparency into AI, much like nutrition labels that summarize useful information on packaged foods. The AI FactSheets website describes an AI transparency approach using consumer-friendly key facts about how an AI model or service was developed, tested, deployed, monitored and modified over time. The site contains examples of models in an open source model catalog, a methodology for creating useful FactSheets and insight about how FactSheets can be used to support AI governance.
- A public Slack channel, more than 800 members strong, for anyone to ask questions
Creating enterprise-ready, explainable AI with the help of the open source community
Nurturing our open source community provides IBM with a pragmatic approach to implementing our research in the real world. “People think when there’s a research result published in a paper, that it’s done. It’s not done,” says Hind. “When we productize research, incorporating customer feedback from our open source community is crucial. Part of the transition from research to product is to determine which research gems are appropriate for an enterprise environment.” When it comes to completing the end-to-end product lifecycle, including capabilities for explainable AI and model monitoring, evaluating different use cases in an open system has been vital in turning theory into reality.
Who benefits from a robust open source community?
Mike Hind identifies an expansive ecosystem. By using open source:
- Data scientists can try out ideas in real software without reading reports and implementing the ideas themselves.
- Researchers can share results quickly and make true apples-to-apples comparisons on a common platform.
- Executives and business leaders can develop a framework of thinking that helps them make key decisions as informed buyers and evaluators.
- Policy makers and social scientists gain more education so they can provide key feedback on important social issues such as trust and transparency in AI.
“It’s a big mistake to have just data scientists and researchers define trust,” explains Hind. “Trust is a human characteristic. You need psychologists, philosophers and other stakeholders to inform what that really means.” Open source toolkits that help all participants — from both the arts and sciences — speak the same language can foster the deep collaboration needed to protect trust in AI.
When businesses can explain and trust AI, they can increase the number and accuracy of models in production — resulting in measurable economic value. Open source is a key part of this effort. People can share ideas, advance science and gain visibility into a secure and fair AI lifecycle, all of which are rooted in core IBM values to make technology that positively impacts the world. Cultivating AI trust and transparency through open source technologies is excellent for the bottom line. But even more important, it’s simply the right thing to do.
Integrate a wide array of open source technologies on our unified data and AI platform, IBM Cloud Pak® for Data.
See product benefits, use cases, customer stories and testimonials at our website.
Learn about the future of AI-powered analytics in our newsletter featuring complimentary research from Gartner.
Dive deeper into how IBM supports trust and transparency with explainable AI in this two-sheeter.