IBM Watson updates focus on data privacy and explainability
Updates include new federated learning capabilities within IBM Watson Studio and more insights into how predictions are made in IBM Planning Analytics with Watson.
IBM today previewed new updates to IBM Watson aimed at addressing explainability in AI models and data privacy concerns.
The updates include a new data privacy management module in IBM OpenPages with Watson, more explainability for planning forecasts in IBM Planning Analytics with Watson, and new federated learning and time series capabilities with IBM Watson Studio.
Addressing data privacy
The updates are "timely, if arguably overdue, particularly for businesses outside of the U.S.," said Alan Pelz-Sharpe, founder and principal analyst at Deep Analysis.
"Governments worldwide are starting to bring in more regulations and guidelines for the ethical and lawful use of AI and data and clearly there is mounting public concerns," he continued.
IBM OpenPages with Watson is a suite of risk management applications. The new data privacy management module integrates Watson Knowledge Catalog, a data catalog for AI governance, to show users how private data is used, in real time, in applications and AI models throughout an organizations.
Explainability in planning forecasts is essential.
Alan Pelz-SharpeFounder and principal analyst, Deep Analysis
The capability, according to IBM, is meant to help users automate the reporting of personally identifiable information to improve accuracy and reduce audit times.
Meanwhile, IBM revealed a new statistical details page within IBM Planning Analytics with Watson aimed at providing users with more insights into how their forecasting predictions are generated.
"Explainability in planning forecasts is essential," Pelz-Sharpe said. "If you make business decisions based on planning forecasts driven by black box AI and they turn out to be wrong, who is responsible, and why were they wrong?"
"These are questions that need to be answered but previously could not be," he said.
IBM said it plans to make the new feature generally available later in the second quarter.
Typically, training data for machine learning models is pooled together before a model is trained on it. Federated learning keeps the data sources separate, instead pooling parameters after individual models are trained. This method of machine learning training is particularly useful in situations in which data can't be pooled for privacy or regulatory concerns, such as in healthcare.
IBM released the new federated learning capabilities within IBM Watson Studio as a tech preview.
The tech giant also highlighted time series capabilities, in beta, in IBM Watson Studio. The new features, according to IBM, are meant to enable businesses to develop models that predict future values of a time series based on past values. This could be useful, for example, when trying to predict sales volume for the upcoming month, or travel time for a shipment.
Pelz-Sharpe noted that while IBM's AI updates are welcome, other tech giants, including Amazon, Microsoft and Google, will likely make similar moves over the course of the year.
