IBM opens the doors of watsonx.governance to organizations that want to use AI responsibly. This platform is designed to automate AI model management and policies.

IBM today announced that watsonx.governance will be available for use in early December. Governance is the third component IBM is adding to the Watsonx platform, which it launched earlier this year. It is certainly not the most important thing: under the pressure of public opinion and impending legislation, organizations are being asked to assume their responsibilities in developing AI models.

trilogy

Watsonx.governance is part of watsonx, an integrated platform designed to cover the entire lifecycle of AI models. The platform launches at watsonx.ai, where developers can start building and training models. They can deploy their own neural networks or choose from IBM’s catalog. Watsonx.data is the component that takes care of storing the large amount of data required for training LLMs. Watsonx.governance has now been added.

Responsible AI

Organizations that work with AI cannot and should not just do everything with it. Models are very sensitive to variations in the data, which can distort the output of systems. Christina Montgomery, Chief Privacy & Trust Officer at IBM, explains the importance of this third component in a press conference. “Guidelines should not be an afterthought when developing AI models. Organizations need to do it now.”

WatsonX.Governance not only facilitates compliance with internal policies, industry standards, and current and future regulations, but also helps customers maintain accountability at the use case level. The key is to understand how models are trained, what data they use, how they arrive at their recommendations, and whether they are regularly checked for harmful biases.

IBM offers automated capabilities that cover all three key pillars of AI governance: lifecycle management, risk management and compliance. The platform includes:

• Monitoring new LLM measurements: Monitor and alert both the input and output of LLMs when preset quality measurement thresholds are exceeded.
• Insight into LLM development: Automatically collect information about the model building process while explaining decisions to reduce hallucinations and other emerging risks.
• AI lifecycle visibility for LLMs: Automatically document model data at all stages of the lifecycle, monitor text model drift, and track health metrics such as data size, latency, and throughput to identify bottlenecks and compute-intensive workloads.
• Validation tools for LLMs: Enable engineers to integrate language model (LLM) output into question-and-answer scenarios with provided context and reference data. This allows them to assess whether the LLM is appropriately influenced by the reference data, ensuring that the output generated is relevant to the specific context.

IBM watson.governance will be available in early December. IBM could expand its capabilities even further in the first quarter of 2024.