IT managers ever had to revert to a previous version of your software? That's like asking a plumber if they've ever had to tighten a pipe or an architect if they've ever had to Walter a blueprint. It's a key feature needed to successfully run an Enterprise software deployment and until now it's been nearly impossible to do with an AI system. Casper Labs and IBM are building a solution that allows all organizations to do just that. Let's see what this looks like in practice.

Meet Alice. Alice is a customerservice executive for Acme software Co. Acme has developed a generative AI system that powers chatbots for insurance businesses. Meanwhile, Bob is an insurance agent for ABC Insurance. Bob and ABC o rely on Acme's chatbot to more efficiently reach prospective customers that are coming to them to acquire a timely and accurate life insurance quote. Automating the chatbot process has saved Bob's business valuable amounts of time and money as they serve a growing customer base. But one day Bob notices that his chatbot is spewing out nonsensical responses to customers seeking a quote. If this problem isn't quickly fixed it will cost Bob customers and risk harming ABC's reputation and bottom line.

As soon as he notices the problem, Bob reaches out to Alice Alice securely logs into the AI chatbots dashboard and notices that Acme's AI system recently ingested a new version of the large language model or llm on which it is trained she's able to quickly pull up a highly serialized session identification log she sees that the problems arosewithin the past hour the same time that ABC's chatbot was fed the new version of the llm Alice suspects that a glitch in the new llm version is causing the system to malfunction. With a single click, Alice is able to revert to the previous bug-free iteration of the chatbot. She then checks her results against the results that Bob received both before and after the issue emerged she confirms with Bob that the chatbot is working as intended again and just like that the AI performance issue was fixed in a matter of seconds.

Chatbots are just a single example of how important Version Control and auditability are when it comes to AI governance. The same principle applies to automated investing healthare management self-driving cars and more realizing AI's true potential means establishing a better way to manage and monitor AI systems especially when results don't go as planned.