As IT leaders, the constant evolution of workload trends is a familiar challenge and the pressure to operationalize AI workloads, deliver real-time insights, and maintain performance requirements is ever-present. AI workloads are demanding more from our systems, and enterprises are feeling the strain. To meet these demands, a robust, scalable environment is essential—one that can turn ambitious AI use cases into reality and provide real-time AI insights with trust and transparency.

We're excited to announce the latest release of Machine Learning for IBM z/OS v3.2, which introduces features focused on Trustworthy AI. These new capabilities include AI model explainability, drift detection, and overall model lifecycle governance - designed to help and assist organizations in complying with the growing body of AI regulations along with operationalizing transactional AI use cases that require high throughput and low latency.

AI Model Explainability

In the realm of AI, transparency is crucial. AI model explainability ensures that your AI models are interpretable and explainable. This feature allows you to understand how your AI models make decisions, providing insights into model behavior and ensuring that decisions are made for the right reasons. This is particularly important for regulated industries like finance, insurance, and healthcare, where decision-making processes must be transparent and justifiable.

Figure: Example of MLz's AI model explainability capability showing top features influencing the model towards a predicted outcome

Drift Detection

AI models are not static; they can degrade over time as new data alters the baseline they were trained on. Drift Detection helps you identify when your models are becoming less effective due to changes in data patterns. By proactively monitoring for drift, you can maintain the accuracy and reliability of your AI systems, ensuring they continue to deliver high-quality insights.

Figure: Example of MLz's AI model drift detection capability showing top features that are drifting 

Model Lifecycle Management

Managing the lifecycle of AI models—from development to deployment and ongoing maintenance—is complex. Our full model lifecycle governance capabilities provide a structured approach to this process. With tools for model versioning and monitoring, you can ensure that your AI models are consistently meeting your organizational standards and regulatory requirements.

With the announcement of Machine Learning for IBM z/OS v3.2, we have also released a Statement of Direction for IBM Synthetic Data Sets which are intended to assist our clients with training AI models in scenarios where real data may be too sensitive or difficult to access and manage in a responsible and compliant way. This is intended to help our clients accelerate their adoption of AI on IBM Z.

Why IBM Z for AI?

IBM Z is uniquely positioned to handle the demands of AI workloads in transactional environments. With the industry’s first on-chip AI accelerator, here’s why we are uniquely positioned to help our clients:

• High Throughput and Low Latency: IBM Z systems are designed to process massive volumes of transactions with minimal delay, making them ideal for real-time AI applications.
• Scalability: IBM Z’s architecture allows for seamless scaling, accommodating the growing demands of AI without compromising performance.
• Security and Reliability: Built with enterprise-grade security and reliability, IBM Z ensures that your AI workloads are both protected and available when you need them.

Real-World Impact -

Consider a financial institution that processes millions of transactions per day. With our enhanced AI features on IBM Z, they can deploy explainable AI models to detect fraudulent activity in real time, adapt quickly to evolving patterns of fraud through drift detection, and maintain rigorous governance over their AI models to comply with regulatory standards. The result is a more resilient, efficient, and trustworthy AI system that delivers immediate value.

Get Started Today:

Machine Learning for IBM z/OS will be generally available from IBM and certified business partners on June 14, 2024.

To learn more about Machine Learning for IBM z/OS, please visit our product page and connect with our team aionz@us.ibm.com.

Read the full announcement here.

Abid Alam

Sr. Product Manager

AI on IBM Z & LinuxONE