AI Learns to Race: A Hands-On Intro to Deep Reinforcement Learning

Ever wondered how AI actually learns? There's no better way to understand it than watching an AI figure out how to race MarioKart.

Reinforcement learning can feel abstract and intimidating - rewards, policies, neural networks, training loops. But when you see an AI go from crashing into walls to successfully navigating Rainbow Road, those concepts suddenly click.

In this demo-led session, I'll walk you through building a deep reinforcement learning model that learns to play MarioKart Wii. You'll see the complete pipeline: connecting Python to a game emulator, defining reward structures that encourage good driving behaviour, and training a neural network that improves through trial and error—lots of error.

But this isn't just about gaming. The patterns here apply directly to real-world problems data professionals face: optimising processes, automating decision-making, and building systems that improve over time. Understanding reinforcement learning fundamentals opens doors to applications from automated resource allocation to intelligent query optimisation.

I'll also demonstrate how AI-assisted development tools like Claude Code can dramatically lower the barrier to entry for ML experimentation. You don't need a PhD to start exploring these techniques—just curiosity and the right tools.

Walk away with a clear mental model of how reinforcement learning works, practical ideas for your own experiments, and hopefully some entertainment watching an AI learn that banana peels are bad.

Finally understand how AI "learns" by watching one fail spectacularly at MarioKart before getting good Take away a reusable pattern for connecting ML models to real-world systems and feedback loops Leave inspired and equipped to start your own AI experiments—no PhD required