Today, we are excited to unveil a new suite of AI offerings for IBM Z that are designed to help clients improve business outcomes by speeding the implementation of enterprise AI on IBM Z across a wide variety of use cases and industries. We are bringing artificial intelligence (AI) to emerging use cases that our clients (like Swiss insurance provider La Mobilière) have begun exploring, such as enhancing the accuracy of insurance policy recommendations, increasing the accuracy and timeliness of anti-money laundering controls, and further reducing risk exposure for financial services companies.

The new offerings are designed to help accelerate the adoption of popular AI frameworks and tooling. We are also announcing an enhanced Machine Learning for z/OS and advanced intelligence and operational improvements with the latest IBM z/OS operating system.

Our rich AI offerings—coupled with low-latency inferencing that leverages the IBM Telum on-chip AI accelerator—are designed to help clients integrate AI workloads into their most mission-critical business applications running on IBM Z and meet their throughput SLAs.

New capabilities

• AI Toolkit for IBM Z and LinuxONE: Expected to be generally available in Q4, the toolkit is anticipated to support popular industry-standard open-source frameworks—such as IBM Z Accelerated for TensorFlow, IBM Z Accelerated for TensorFlow Serving, IBM Z Accelerated for Snap ML and others—so that businesses can start implementing trustworthy AI.
• Python AI Toolkit for IBM z/OS: This enhanced toolkit is a library of open-source Python software to support AI and machine learning workloads that adheres to IBM Security and Privacy by Design practices. Clients can now leverage zIIP-eligible packages with Python AI Toolkit for IBM z/OS and IBM Open Enterprise SDK for Python 3.11 to embed AI into their applications.

To help data scientists, developers and IT teams implement AI together, these toolkits for Z are designed to support clients as they aim to connect mainframe data and applications to open-source frameworks and packages. These tools include frameworks and libraries that are optimized and supported for IBM Z, and they are built to allow developers to start implementing trustworthy AI capabilities on z/OS. Supported by the same underpinnings for IBM watsonx.ai that empower data scientists and developers to build, run and manage machine learning models, these toolkits will evolve over time to become an integral part of the IBM watsonx platform.

• Machine Learning for IBM z/OS – Enterprise and Core Editions: The enhanced Machine Learning for IBM z/OS is IBM’s flagship full-lifecycle platform designed to help organizations build, deploy, manage and operationalize machine learning and deep learning models on z/OS. Built for developers and data scientists, this platform is a watsonx.ai extension for z/OS and is designed to support faster development, deployment and monitoring of machine learning models. In our view, organizations must be clear about how their AI models are trained, what data is used in that training and what goes into an AI model’s recommendations. Today, clients can implement trustworthy AI capabilities on IBM Z through IBM Cloud Pak for Data capabilities, helping to ensure models and workflows are transparent and explainable. In the coming months, IBM plans to roll out these capabilities natively for workloads on IBM z/OS.
• Cloud Pak for Data on IBM Z: This is an enhanced, powerful Auto AI tool within Cloud Pak for Data 4.7 for automating the process of building machine learning models. It allows users to upload their data, choose the problem type, specify constraints and run a series of automated experiments that generate a range of high-performing pipelines quickly and easily.
• AI-infused IBM z/OS 3.1: Generally available on September 29, 2023, IBM z/OS 3.1 represents a new era in operating system intelligence. Using the new AI System Services for IBM z/OS, the system is designed to learn and predict how to optimize IT processes, simplify management, improve performance and reduce special skills requirements.

How La Mobilière is unleashing the future of insurance with AI

We’ve committed to optimizing the IBM Z platform for AI from the silicon up, starting with the IBM Telum processor. IBM z16 delivers an on-chip integrated accelerator for low-latency inferencing, backed by IBM Z resiliency and security. Focusing on reducing latency during inferencing allows clients to shorten the time to receive insights and recommendations from AI models.

“According to our research, we estimate that about 70% of global transactions, on a value basis, run on IBM Z,” said Neil Katkov, PhD, Director at Celent. “Applying fine-tuning and inferencing to mission-critical data at this scale is a powerful innovation for use cases across business and IT operations.”

Analyzing transactions at scale could mean that insurance companies can more quickly make insurance offer recommendations. For example, La Mobilière has depended on IBM Z to process high volumes of transactions and provide a secured environment for their most sensitive data. Leveraging SQL Data Insights on IBM z16 allowed them to uncover patterns in their mission-critical data to improve their insurance offer predictions and reduce cost.

“In the insurance sector, customers look for the most personalized service possible to get the assurance they need to feel protected. As a leading provider in Switzerland, our goal is to use the latest technologies available in order to deliver on this promise to our customers,” said La Mobilière’s IT architect Thomas Baumann. “We worked together with IBM to apply the AI capabilities on their trusted IBM z16 systems to process insurance offer recommendations faster and more accurately. By unlocking hidden data patterns with NLP-based AI functions in near-real time while ensuring privacy and security, we saw 94% accuracy in prediction results. These very promising results have motivated us to integrate this technology into our underwriting processes in the near future.”

IBM Z supports the entire AI lifecycle

There’s a lot of innovation happening with generative AI, including the recently announced IBM watsonx Code Assistant for Z—a new generative-AI-assisted product for mainframe application modernization that will help enable faster translation of COBOL to Java on IBM Z and enhance developer productivity on the platform.

However, for many businesses, the first step to deriving value from AI today means focusing on the entire AI lifecycle, which also includes the fine-tuning, inferencing and deployment of machine learning and deep learning models. For companies to make the most of their AI investments, we believe that they need to tap into their mission-critical data. IBM z16 is designed to score business transactions at scale delivering the capacity to process up to 300B deep learning inference requests per day with 1ms of latency.[1] This scale opens significant opportunities for AI use cases on the mainframe, such as fraud detection, anti-money laundering, clearing and settlement, healthcare and application modernization.

To learn more about the new suite of AI offerings and get started on your AI journey, visit the AI on IBM Z website, and read the recently published analyst paper from Karl Freund.

IBM’s plans, directions, and intentions may change or be withdrawn at any time at IBM’s discretion without notice. Information about potential future products and improvements is provided to give a general idea of IBM’s goals and objectives and should not be used in making a purchase decision. IBM is not obligated to provide any material, code, or functionality based on this information.

[1] Disclaimer: Performance result is extrapolated from IBM internal tests running local inference operations in a z16 LPAR with 48 IFLs and 128 GB memory on Ubuntu 20.04 (SMT mode) using a synthetic credit card fraud detection model exploiting the Integrated Accelerator for AI. The benchmark was running with 8 parallel threads each pinned to the first core of a different chip. The lscpucommand was used to identify the core-chip topology. A batch size of 128 inference operations was used. Results may vary.