“If you manage to learn a little bit of Python, have a good idea of how models are actually built and you really know your data,” you’re well on your way to developing AI for your IT environment, doctoral AI researcher Thomas Decorte said in November on Charlie Guarino’s monthly virtual IBM i meeting, iChime. 

IBM i users are likely to at least have that third factor covered. “They know their data perfectly,” Decorte said. “It's why we tend to refer to end users as power users, because they actually know how the data looks, how it's structured, how it's connected. Then, you can move towards a very nice use case.”

The initial challenge that organizations tend to face in developing AI, Decorte said, is having quality data, in great enough quantities. “Sometimes there's just not enough data to actually be able to deduce something from,” he explained. “And in other use cases, the data quality is not really up to par, or some extra data needs to be collected.”

That’s less likely to be the case on IBM i, according to Chicola. “​​It's a great AI machine because you've got good data, real good data,” he said. 

Companies that have an IBM i have usually had it for decades, Decorte noted. “They have gathered so much data because they don't tend to throw that much data away,” he said.

This, Decorte continued, “is why I think the platform is actually very well suited to build AI models on, because there's just so much data that's continuously kept over time within a very good database, the DB2.”

For Chicola, it’s easy to see how organizations on IBM i accumulate so much data, defying the idea that IBM i is just for smaller operations. “Our largest customer probably has 15,000 users on an IBM i, one machine,” he said. 

While sufficient data is a necessity, “the biggest positive of doing AI on IBM i is that the platform itself, number one, is stable; number two, is secure,” Chicola said, calling the system a “workhorse” that runs everything from hospitals, to banks, to casinos, to trucking companies. 

Whatever the AI mission, Chicola is adamant that IBM i is up to the task. “There is no shortage of horsepower on an IBM i,” he said. “…I mean they could run a small community bank, right? We’ve got plenty of those. And it could run a monster company, some of the biggest companies in the world.” 

Decorte noted, however, that IBM i isn’t the best machine for every task; sometimes, the oomph of a GPU is what’s called for. “In some use cases, the algorithm that's used doesn't benefit from the parallel computation, so then it can be just run on an IBM i,” he said. But in cases where the system is at full capacity, he continued, “then you might opt to offload [the AI task] to a GPU somewhere, usually a GPU in the cloud.”

Then again, “you can technically train anything on the IBM i. It might just take a little bit longer than you would on a GPU,” he said.