Probability & Uncertainty

AI systems deal in probabilities, not certainties - even when they don't show it. When a model identifies a photo as a cat with 97% confidence, that means it's also 3% uncertain, and sometimes that 3% matters. Language models are probabilistic by design: they don't produce one "right" answer but sample from a distribution of possible outputs, which is why you can ask the same question twice and get different responses. Many people find this unsettling because we're used to computers being deterministic - you type 2+2 and always get 4. AI doesn't work that way. It's more like asking a knowledgeable friend for advice: you'll get a useful answer most of the time, but it won't be exactly the same every time, and occasionally it'll be wrong. Understanding this probabilistic nature is essential for using AI well. It means building in checks rather than blind trust, running important queries multiple times, and recognising that confidence in presentation doesn't equal confidence in accuracy. The most dangerous AI outputs are the wrong ones that sound absolutely certain.