-
AI models will inevitably make mistakes in their outputs
-
The ability to track down the source data behind AI errors is key
-
Complex multi-cloud environments make model traceability tricky, NetApp told us
Like it or not, artificial intelligence (AI) and multi-cloud are here. While these technologies are thorny enough to deal with on their own, their intersection could leave CIOs and CTOs reaching for Extra-Strength Advil.
It’s true that it’s still early days for AI. But some of the top minds working on the technology — including Red Hat’s CTO — have predicted the number of AI models will explode in the coming years. That’s because enterprises are expected to begin customizing the handful of today’s models by training them with their proprietary data.
However, the problem looming on the horizon is all about AI model traceability.
As Jeff Baxter, NetApp’s former Field CTO and current VP of product marketing, told Silverlinings, these AI models will inevitably make mistakes in their outputs. And being able to trace the source of those mistakes is critical to not only correcting them, but also maintaining trust with users.
“Being able to provide traceability to AI models — and traceability on what the data set looked like when [the models] were trained to go back and figure out what happened with [an] error — is one of those unanswered questions that the industry is still grappling with,” Baxter said.
This question fits alongside one about how to manage AI models over time to avoid what neXt Curve analyst Leonard Lee called “AI dementia” at our Cloud Executive Summit in December.
AI dementia is the idea that the data models were trained on becomes outdated with time and thus, the models themselves become less accurate if they are not maintained.
This sounds sensible enough but you may be asking, "What’s all this got to do with multi-cloud?"
Well, Baxter explained that the cloud adds complexity to AI models since each cloud comes with its own standards and methods of cataloging the data stored in them.
“If this was two decades ago and all my company’s data was sitting in one data pool in one data center, then yeah, it’s going to be easier” to trace the source of an AI error, Baxter said. “Hybrid multi-cloud has opened up a world of opportunities for businesses…but anyone who tells you that it hasn’t increased complexity is wrong.”
Baxter said NetApp’s answer to the problem are new unified data storage solutions which smooth over the differences between cloud storage systems with a third-party control plane that can be used in both hyperscale and on-premises environments.
The architecture covers several different kinds of storage, including file, block and object. Check out our reporting on the intersection of AI and object storage here. Maybe storage is interesting, after all? Stay tuned on that front.