Supervised Machine Learning
Supervised machine learning, the technical term for what is popularly called 'AI', is a method where a computer refines an algorithm by comparing its outputs to known correct answers and adjusting the model until the errors are minimized. The key insight: instead of programming the algorithm, you let the data shape it.
In traditional programming, a developer writes the rules: if X, then Y. But in many problems, recognizing faces, translating languages, predicting which flight seats will sell, the rules are too complex to write by hand. Supervised machine learning inverts the process. You provide labeled input-output pairs ('this image is a cat,' 'this image is a dog'), build a working mathematical structure, and let the computer adjust that structure through millions of iterations until it reliably produces the right output.
Three conditions are required. First, clear metrics: you need an objective, measurable output (profit per flight, face match percentage) rather than something subjective. Second, clean data: the training examples must be accurate and representative, garbage in, garbage out. Third, a working algorithm structure: computers can optimize an existing structure but cannot create one from scratch. That still requires human ingenuity.
These constraints explain why self-driving cars remain unsolved despite billions in investment. The 'edge case', a human who deliberately wants to crash into you, cannot be programmed around. Any system where a human can override the outcome with intent is, at some level, uncontrollable. 'AI is a scam,' the article argues. 'What exists is supervised machine learning. That's it.'
Related terms
Frequently asked questions
What is supervised machine learning? ＋
Supervised machine learning is the technical process behind what people call AI. A computer is given labeled input-output pairs, builds a mathematical model, and iteratively adjusts that model until it reliably predicts the correct output. Facial recognition, translation software, and recommendation engines all work this way.
Why can't AI solve self-driving cars? ＋
Because of the 'edge case' problem: you cannot build an algorithm that accounts for a human who deliberately wants to cause an accident. Every other scenario can be modeled, but intentional human interference cannot. This is why self-driving features exist but fully autonomous driving does not, and may never exist.