From the course: Foundations of AI and Machine Learning for Java Developers

Determinism vs. probability - Java Tutorial

From the course: Foundations of AI and Machine Learning for Java Developers

Start my 1-month free trial Buy for my team

Determinism vs. probability

“

- [Instructor] Before we put our generative AI systems in production, let's compare traditional IT systems, that is deterministic IT, to this new type of gen AI-based IT, which is probabilistic. The vast majority of systems that all of us are familiar with are deterministic. That means you get the same output if you give the same input to a traditional IT system. Same input, same output. Certainly there are several advantages or benefits of deterministic IT. When you get the same output for the same input, it makes it very easy to test and monitor. You know what the results should be, so if something doesn't fall within a certain threshold of your output expectations, that's probably an error. That means it's more straightforward to monitor and test. Certainly auditing deterministic IT is more straightforward since the output is more predictable, and over the past few decades, there have been many traditional IT best practices to follow. Now, AI systems are probabilistic. For the same input, you might get the same output, or you might not. Recall that these are what are called stochastic systems. They're probabilistic. So, you might get a range of outputs for the same input. If you work on investment models for a financial services company or meteorological models, you'll appreciate the difficulty of managing stochastic systems. This is a concern when putting probabilistic systems into production because the same input can generate various outputs, which could be quite difficult to manage. It's very, very hard to test and monitor, 'cause you may get a result that appears to be correct, but not, or worse, you may get a result that's potentially risky to your company, so it's certainly hard to monitor using traditional infrastructure. Auditing is certainly more complicated, and dealing with probabilistic infrastructure is a brand new issue for software developers, and certainly for IT management. Probabilistic systems are unfamiliar to most companies and enterprises, so in some cases, if you need deterministic systems, just use traditional IT.

Contents