AIQ
AIQ
Everything a teacher needs to deliver this lesson — pick your grade's script below once you've read the background.
This lesson runs across all four AIQ age modes, so the depth of what's covered changes a lot from Explorer to Architect — but every version builds toward the same core idea: getting around is one of the places AI has moved fastest from research lab to daily life, and it does its "driving" and "deciding" through sensors and data, not through anything resembling human intuition. By the end of the lesson, a student should be able to:
You don't need an engineering background to teach this well — you need one mental model, and the lesson gives it to you directly: a self-driving car doesn't have eyes, so it has to build a picture of the world out of sensors and turn that picture into decisions, over and over, many times a second. Everything else is detail on top of that.
Start with how the car perceives its surroundings. Most self-driving systems combine several sensor types because each one is good at different things and bad at different things, and combining them (often called "sensor fusion") covers for each one's weak spots. Cameras are cheap and see color and detail — useful for reading a lane line, a traffic light color, or a road sign — but they struggle in fog, glare, or darkness, and they don't directly measure distance. Lidar (short for "light detection and ranging") fires millions of laser pulses per second in every direction and times how long each one takes to bounce back, building a precise 3D map of everything around the car regardless of lighting — but it's expensive and can struggle in heavy rain or snow. Radar uses radio waves instead of light, which cuts through fog and rain well and measures how fast other objects are moving, but at much lower resolution than a camera or lidar. A car built with all three is trying to make sure that whatever fools one sensor doesn't fool the others at the same time.
Once the car has a picture of the world, something has to decide what to do with it — this is where "AI" is really doing its job. Modern self-driving software is generally described in four stages: perception (what's out there? a pedestrian, a cyclist, a parked car, an open lane), prediction (where is each of those things about to go?), planning (given all of that, what should this car do next?), and control (actually turning the wheel, applying the brake, adjusting the throttle to carry out the plan). That loop runs continuously — the lesson's own line is "thousands of decisions every second," and that's a fair description of how often a modern system re-evaluates the driving situation, not a one-time reaction to a single event like a deer jumping into the road.
The same underlying idea — sense the world, then decide — shows up outside individual cars too, and the lesson deliberately includes those examples so kids don't think AI-in-transportation means only self-driving cars. Adaptive traffic lights use sensors and historical data to change their timing based on actual traffic flow rather than a fixed clock. Route-planning apps like Google Maps use huge amounts of aggregated location data to predict congestion and suggest faster paths. Sidewalk delivery robots (the lesson names Starship, a real company operating on several U.S. and U.K. college campuses) use cameras and mapping to navigate around pedestrians and obstacles at walking pace, which is a meaningfully easier problem than driving on a road at highway speed.
For the oldest students, it's worth knowing the phrase "the long tail problem," which the lesson introduces directly at Hacker/Architect level. It refers to the fact that the vast majority of driving is fairly routine and easy to train a system for, but safety is determined by the rare situations — a ball rolling into the street followed by a child, an unusual construction detour, a driver signaling with a hand gesture — that are individually uncommon but collectively unavoidable over millions of miles of driving. This is the same reason the lesson also introduces V2X communication (vehicles talking to traffic infrastructure and each other) and, at the most advanced level, occupancy networks (a technique for predicting which small 3D regions of space around the car are physically occupied by something, even before the system knows what that something is) — both are attempts to close some of that long-tail gap by giving the car more information than its own sensors alone can provide.
This is a light-prep lesson. You need: