AI Auditability: Why Explainability Matters in Regulated Industries

If you work in regulated industry, if you work in something where you have to be able to prove to your auditors, um, how you came to a decision. When you deploy your favorite AI model into, you know, the cloud, and then you set up your data pipelines to be able to derive insight from that model, can you explain how it came to that conclusion? And that is key for so much reg tech type implementations. Like, it's great to use AI, it's great to adopt the latest and greatest frameworks and be able to obviously like, you know, speed stuff up. But the key point to all of this is the auditability from a regulation perspective.

And, you know, it's just something to think about as you go through the excitement of trying to optimize your pipelines or do something different, is just consider if an auditor came down tomorrow, now you've switched the thing on, do you have a way of being able to prove how it came to that decision? Now, of course, there are different ways to be able to do it. The problem is with a lot of LLMs is they are very much a black box, especially the ones where you don't host them yourselves and you're just sending stuff to a third party. It is, you know, LLM bingo as to like how it came to a decision, which version you were using. Can you go back and still run the stuff on the older versions to prove how you came to that decision?

Would it come to the same decision twice because they're not deterministic if you put the same inputs in? Now, it's not to say that you shouldn't use AI, but you've got to make sure that you have an audit proof workflow so that when someone comes and asks how, you can show them. So put some thought into that. If you would like to have a discussion, happy to have that discussion. But I think it's something that's pertinent in, you know, the world that we now live in is the explainability of all this, especially in financial compliance, in health tech, all that type of stuff, is how do you get from data input to the result that's coming out of your LLM.

And yeah, there you go. Let me know your thoughts about that in the comments, uh, or send me an email, tom@conceptcloud. com. I'd love to, uh, hear the audience's thoughts on that. And maybe we'll follow up, uh, in another episode about some of the ways that you can, of course, you know, comply with the audit, auditability, um, requirements, um, when building out your AI pipeline.

Okay. That's it. That's Friday. Thank you very much. I'm off to go and find some paracetamol for my shoulder.

I will speak to you all on Monday. Have a great weekend. Why hire when you can partner? Concept Cloud's leading engineers build your startup's prototype without the overhead. Launch faster.

Conceptcloud. com.