What AI Cannot Do

AI is impressive. It is also genuinely, fundamentally limited in ways that matter for how you use it. Understanding these limitations is not pessimism — it is the difference between using AI well and using it badly.

AI does not understand — it predicts

When an AI gives you an answer, it is not reasoning through the problem. It is predicting what a good response would look like based on patterns in its training data. This means it can produce fluent, confident, completely wrong answers — a phenomenon called hallucination.

AI has a knowledge cutoff

Most AI models are trained on data up to a certain date and know nothing about events after that. Asking an AI about recent news, recent research, or anything that happened after its training cutoff will likely produce outdated or fabricated information.

AI cannot verify facts

AI does not check sources. It cannot tell you whether the information it has given you is accurate. When it cites a study, that study may not exist. When it quotes a law, that law may have changed or been misquoted. Always verify anything consequential.

AI has no common sense

AI fails at simple reasoning that humans find effortless. Ask it to count how many letters are in a word, describe what happens if you drop a ball in a swimming pool, or solve a simple logical puzzle — and it may confidently get it wrong. It has learned language, not the world behind language.

AI is not neutral

AI reflects the biases of its training data and the choices of the people who built it. It may perform better for some demographics than others, represent some perspectives more fairly than others, and have blind spots shaped by who created it and where.

The most dangerous AI user is someone who trusts it without understanding its limits. The most effective AI user is someone who knows exactly when to trust it — and when to check.