Chapter 03 · What AI does

What AI is actually good at

AI is not a general-purpose intelligence. It is a collection of powerful but narrow tools, each trained for a specific family of tasks. Understanding what it can and cannot do is the foundation of using it wisely.

The tasks AI handles well share a common structure: they involve finding patterns in large amounts of data, where the right answer can be inferred from enough examples. If you can define a task clearly enough to label thousands of examples of it, you can probably train a system to do it.

Click each capability to explore

Each card shows the capability, its strength, and a real-world example. Click to expand.

👁️
Seeing & recognizing
Identifying objects, faces, and patterns in images and video.
Capability
✍️
Reading & writing language
Understanding and generating written language at human-level fluency.
Capability
🎯
Predicting outcomes
Forecasting what will happen based on patterns in historical data.
Capability
🎨
Generating new content
Producing original images, code, music, and text from a description.
Capability
🔊
Listening & speaking
Converting between spoken and written language with high accuracy.
Capability
♟️
Optimizing decisions
Finding the best solution among millions of options in complex systems.
Capability

What's conspicuously absent

The capabilities above are real and impressive. But the gaps are equally important to understand.

Task
AI performance
Recognizing a cat in a photo
Exceeds human accuracy
Translating Spanish to English
Near-human quality
Detecting fraud in transactions
Millisecond accuracy
⚠️
Understanding a joke's meaning
Inconsistent
Common sense about physical objects
Frequently fails
Knowing when it doesn't know
Poorly calibrated
Genuine reasoning from first principles
Not reliable

"AI excels at pattern recognition at scale. It struggles with anything that requires genuine understanding, common sense, or reliable self-awareness."