Why Understanding Probability Is the Key to Making Sense of Artificial Intelligence
Ask an AI for the winning lottery numbers and it will give you an answer. But it won’t help you win.
Artificial intelligence doesn’t know what will happen next. It calculates what’s likely to happen next, based on patterns in vast amounts of data. And when it comes to something like lottery numbers—where randomness is the whole point—there are no patterns to learn.
Understand that, and AI stops feeling mysterious. It starts to make sense.
The Lottery Problem
Lotteries are designed to resist prediction. In games like Powerball lottery or Mega Millions, every number has the same odds every time. There’s no memory, trend, or “due” number waiting its turn. That’s randomness. And it’s exactly where AI stops being helpful.
You can ask an AI model for “likely” lottery numbers, but that question doesn’t quite work. There is no “likely.” Every combination carries the same probability.
As David Spiegelhalter writes in The Art of Statistics, probability “does not tell us what will happen, only how likely different outcomes are.” In a lottery, those outcomes are evenly matched, which is another way of saying they can’t be predicted.
What AI Is Doing
When you ask AI a question, it isn’t pulling a fact from a shelf. It’s building a response—one word at a time—based on what it has learned is most likely to come next. For example:
- The capital of France is … Paris
- Peanut butter and … jelly
These aren’t retrieved answers in the traditional sense. They’re high-probability continuations.
As Christopher Manning explains in Stanford’s CS224N course, language models are trained to “predict the next word” from patterns in massive data sets.
That simple goal—repeated billions of times—creates something that feels like understanding.
Underneath, it’s still probability.
Where AI Works and Where It Doesn’t
The lottery exposes the boundary. AI works well when patterns exist, such as in language, images, and human behavior. However, lotteries don’t have patterns or structure. So, the system has nothing to latch onto—no signal, just noise. “Indeed, it is the ability of lottery outcomes to be as random as possible and free from predictability, influence, or manipulation that guarantees that the ‘house’ wins,” said ASA Executive Director Ron Wassertsein.
That’s why AI can draft an essay, summarize a report, but can’t tell you next week’s winning numbers.
Confidence Isn’t Certainty
AI comes across as confident. The sentences are smooth. The answers arrive fully formed.
But confidence is part of the output—not proof of accuracy.
Because the model is designed to produce the most probably-expressed response, it will do so fluently—even when the underlying information is shaky.
Research from Pew Research Center on public awareness of artificial intelligence shows many people interpret confident AI responses as authoritative—even when they’re wrong. That gap—between how reliable something is expressed and how reliable it is—is where misunderstanding creeps in.
The Mindset Shift
Once you see AI as a probability machine, the questions change. You stop asking, “Is this right?” And you start asking:
- “How likely is this?”
- “What is this based on?”
- “Where could this go wrong?”
That’s statistical thinking, which is quickly becoming a basic form of literacy.
The Final Word
AI can do remarkable things. But it can’t tell you the winning lottery numbers—and that limitation is more revealing than it seems.
It’s a reminder that behind the fluency and speed, AI is still working in the language of probability: weighing patterns, making predictions, choosing what comes next. Not certainty, just likelihood.
Valerie Nirala
ASA Publications Strategist
A Short Guide to Probability
Probability is a way of talking about uncertainty. Not right or wrong. Not true or false. Just how likely is this?
If you flip a fair coin, the probability of heads is 50%. That doesn’t mean heads will come up next—it means that, over time, heads shows up about half the time. The same idea shows up in places such as weather forecasts, medical risks, and election predictions.
As Nate Silver writes in The Signal and the Noise, probability helps us “quantify uncertainty” so we can make better decisions—not eliminate doubt.
AI works the same way. It doesn’t deal in certainties. It deals in likelihoods.
Leave a Reply