What is a probabilistic algorithm?
A probabilistic algorithm is a procedure that uses random choices or probability estimates as part of its steps. Two runs on the same input can return different results, or the result comes with a likelihood attached instead of a guarantee. Algorithms of the first kind are also called randomized algorithms. The counterpart is a deterministic algorithm, which always returns the same output for the same input.
Much of modern AI works this way. Monte Carlo simulation estimates a range of likely project finish dates by running thousands of random scenarios. A large language model assigns a probability to each possible next token and samples from that distribution, so the same prompt can return different wording. Setting the temperature to zero removes most of that variation, although small differences can persist because of how GPUs batch and round calculations.
Probabilistic algorithm in practice
A law firm uses a language model to flag discovery documents that may be privileged. When the firm runs the same set twice, most labels match, but a few borderline emails change category. The firm sends any document with an unstable or low-confidence label to an associate for review. Because rerunning the model will not reproduce an earlier answer, the firm acts on the label stored in its review log.
Probabilistic methods are often the only practical way to solve a problem, and their error rates can be measured and managed. Testing such a method means running the same cases many times and measuring how often the output is acceptable.