Monitor the model with Watson OpenScale

IBM Watson OpenScale is an enterprise-grade environment for AI applications that provides your enterprise visibility into how your AI is built, is used, and delivers return on investment. Its open platform enables businesses to operate and automate AI at scale with transparent, explainable outcomes that are free from harmful bias and drift.

Watson OpenScale can automatically generate a de-biased model endpoint to mitigate your fairness issues and provide an explainability view to help you understand how your model makes its predictions. In addition, Watson OpenScale uses drift detection. Drift detection tells you when runtime data is inconsistent with your training data or, if there is an increase in drift, that the data that is likely to lead to lower accuracy. With the Watson OpenScale service, you can scale adoption of trusted AI across enterprise applications on hosted on-premises environments or in a private cloud environment.

Watson OpenScale can drive fairer outcomes by detecting and helping mitigate model biases to highlight fairness issues. The platform provides plain text explanations of the data ranges that have been impacted by bias in the model as well as visualizations that help data scientists and business users understand the impact on business outcomes. As biases are detected, Watson OpenScale automatically creates a de-biased companion model that runs beside a deployed model, thereby previewing the expected fairer outcomes to users without replacing the original.

Watson OpenScale helps enterprises bring transparency and auditability to AI-infused applications by generating explanations for individual transactions being scored, including the attributes used to make the prediction and weightage of each attribute.

Watson OpenScale capabilities

• Production monitoring for compliance and safeguards: Detect and mitigate model biases, audit and explain model decisions, and validate models for acceptance

• Ensure model resilience across changing business situations: Detect drift in data and anomalies in model behavior

• Align model performance with business outcomes: Correlate model metrics and business KPIs to measure business impact, actionable metrics and alerts.

See Watson OpenScale on Cloud Pak for Data for information on how to install, administer, develop on, and use Watson OpenScale.