Some Thoughts On Big Blue’s GenAI Strategy For IBM i

August 12, 2024 Timothy Prickett Morgan

In a world that has gone half mad with generative AI, it is refreshing to see the people who control the IBM i platform being skeptical, hopeful, and practical about how the technology might be used to help the companies who choose Power Systems running IBM i as the platform for their mission critical applications.

IBM Rochester has always been practical and often innovative when it comes to adopting hardware and software technologies, so the strategy that IBM i chief technology officer Steve Will laid out in a recent IBM i & AI – Strategy & Update as part of the COMMON Guided Tours 2024 series was, well, exactly what we expected.

Integrating with AI is not a new thing for the IBM i folks, of course. Almost as soon as IBM had created the Watson QA system that played the Jeopardy! game show and begun commercializing it, there were hooks linking it to IBM i applications and databases. The commercial success of the Watson stack of AI tools is debatable, and what is clear is that the first wave of this QA variant of AI was not anything like we are seeing from generative AI running on large language models today.

But aside from this, IBM has worked with customers to do more traditional, statistical versions of AI that are not based on LLMs and that are used to validate credit scores and do risk assessments of the financial states of companies and people, that track and predict food safety, that guide the design of products and predict customers preferences for manufacturers, and that employ chatbots for operations and customer interfacing.

As Will explained on the Guided Tour, rather than try to boil the ocean with GenAI, IBM Rochester has “three clear use cases” for generative AI that it is focused on right now:

We have already talked about the third area – developer experience – where IBM is asking for donations of raw RPG code and code/explanation pairs to create a code assistant for RPG. We revealed the work on the code assistant product back in May and talked about the way IBM was going to use donated code for to create a Granite 8B model that understands RPG code in July.

“Our strategy has been to invest in things that will allow the solutions running on IBM i to take advantage of IBM i and Power to do whatever you need to do now and next,” Will explained. “And it has been about giving you choices on how you move forward with that advancement of your solution, and making it as simple as possible by integrating the technology. So when it comes to AI, what that means is, we know that our customers want AI that works with their solutions and their data – people aren’t going to buy IBM i just to run a new AI workloads. That’s not what we’re about. What we are about is making sure that the solutions that you have can be extended to take advantage of AI, whether it is in how it uses database or how it interacts with AI workloads.”

With Db2 data analytics, generative AI is going to be used to do trend analysis and anomaly detection, things that have been done procedurally and programmatically with algorithms up to this time but now can have generative AI models doing the analysis and detection. These GenAI models, as Will pointed out, can be trained to notice oddities in data without having to be programmed – in essence, the LLM codes a neural network based on past data to notice patterns and see new ones as they emerge without having to have an algorithm, or set of algorithms, encode that pattern searching. What customers really want is to be able to see if there is anything weird in their transactional data as it is happening.

This is the top thing that application developers asked for when they were prompted for ideas about how GenAI might be useful to the IBM i platform, by the way. This was higher in the list than a coding assistant.

The next priority that users had when IBM asked them about how GenAI might help the platform was for automation in the management of IBM i platforms, and particularly for those who are managing fleets of machines either for themselves or on behalf of customers who rent IBM i capacity from them. As you are well aware, the AS/400 and its progeny are legendary when it comes to self-tuning and self-management, but there is always work to be done here to make the box better, and AI will do its part in enhancing how the IBM i platform takes care of itself so you don’t have to.

You will notice, of course, that IBM is not making a coding assistant to port RPG code to Java, as its mainframe brethren are doing to help port COBOL code to Java on mainframes. The System z team might have taken a similar tack and helped COBOL shops better extend and modernize their COBOL applications and therefore help preserve the need for System z mainframes. But we also assume that a tool porting from COBOL to Java is what customers asked IBM to do.

IBM is not creating GenAI modules that replace workers on assembly lines or in warehouses, it is not building systems that replace customer support reps or sales reps. It is not doing any of that.

We are not saying that IBM i shops are not interested in such things – perhaps they are. But IBM and its IBM i users will wait for third-party software developers to come up with such things, we think. Some IBM i shops might build such GenAI systems from scratch as well to gain some kind of competitive and cost advantage in their markets. Such work would require huge investments in GPU-accelerated hardware and expertise in GenAI models, of course, and that might mean IBM i shops cannot get the budget together for such efforts. Excepting the use cases that Will outlined above, GenAI might be a game played mostly by the hyperscalers, cloud builders, and Nvidia.

This Issue Sponsored By

Table of Contents

Content archive

Recent Posts

Subscribe

Pages

Search