PSA: Saying “1 in 5” Is Better Than “20%” When Informing about risks publicly
This idea may be obvious to many. But given how important it is now to warn about the dangers of AI and that lots of people knowledgeable about AI risks can live in a different environment than decision-makers, I’ll take the liberty of writing it here just in case: Most people often find it challenging to interpret probability when it’s expressed in percentages.
Expressing probabilities as “1 in 5” rather than “20%” helps people understand risks more intuitively and avoids the cognitive barriers associated with percentages. Based on my personal experience (running workshops on uncertainty communication and Tetlock-style forecasting in different settings) it can be good to express risk in this way even to some people who work with probability on an everyday basis.
Cognitive reasons for expressing “1 in 5” rather than “20 %”
Percentages require mental calculations that can feel overwhelming for many. For example, understanding “20%” requires translating it into a fraction or practical terms, which adds an extra cognitive step. Many studies have shown that people grasp probabilities more intuitively when expressed as natural frequencies (e.g., “1 in 10” instead of “10%”). Frequencies are easier to visualize and relate to real-world scenarios.
The Association of Percentages with Math: A Barrier to Practical Reasoning
While percentages are taught in school, they are rarely practiced in a way that connects them to real-life decision-making. Therefore lots of people associate percentages with math and calculating equations rather than practical reasoning. As a result people may know how to calculate percentages but they still fail to apply probability meaningfully in real life or an unfamiliar context.
For instance (anecdotal evidence), during a workshop with bank employees from the risk management department, we introduced an interactive calibration game where participants bet play-money on the accuracy of different knowledge-based questions. Despite their professional experience in financial risk assessment, half of the participants lost all their play-money within the first few rounds—demonstrating that probability skills do not always transfer seamlessly across domains.
Of course, the probability given as natural frequencies has its disadvantages (greater inaccuracy of the information conveyed). But in general, I find it useful to consider this type of expression when communicating uncertainties and risks outside the EA and rationalist settings.
You say “Many studies have shown that people grasp probabilities more intuitively when expressed as natural frequencies (e.g., “1 in 10” instead of “10%”).” but don’t cite any.
They very well might exist, but I’m not sure and can’t quickly inspect this claim, so won’t leave reading this as convinced as I might have.