Probabilistic AI

Definition

AI whose outputs or behaviour are based on probabilities rather than entirely predetermined rules, meaning results can vary across apparently similar tasks.

Put simply

The AI is generating a likely response rather than retrieving one fixed, pre-programmed answer.

What boards and leaders should understand

This is one of the most important differences between generative AI and much traditional software.

The consequence is that assurance increasingly asks:

How often does it work? Under what conditions? What kinds of failures occur?

rather than simply:

Did the test pass?

Probabilistic behaviour also helps explain why different models can produce different answers from the same information and why a system may need repeated evaluation across a range of scenarios.

Questions to consider

  • Where are we still applying deterministic software assumptions to systems whose performance is inherently variable?