BASE – When AI Makes Rare Risks Look Common

BASE - When AI Makes Rare Risks Look Common

Our Nudge consists of a specially developed prompt for AI-assisted medical self-diagnosis. Users enter their symptoms using a predefined structure that guides the AI to consider not only the described symptoms but also the base probability of various conditions. As a result, diagnoses are not made solely on the basis of conspicuous symptoms but are also ranked according to their actual prevalence. The Nudge thus does not alter users’ freedom of choice but improves the way information is processed by the AI.

Accept the cookie setting to see Panopto content.

Unlock Panopto

What does the topic mean?

Artificial intelligence is increasingly being used in healthcare and is also gaining importance in medical self-diagnosis. Many people now use ChatGPT or similar AI systems to assess symptoms or identify possible illnesses. Studies show that general-purpose AI applications like ChatGPT are used more frequently than specialized symptom checkers provided by health insurance companies or medical institutions. At the same time, the decision-making processes of such systems are often difficult to understand. Users frequently place a great deal of trust in AI-generated medical recommendations, even though they have limited ability to assess their quality. This increases the risk that incorrect or biased diagnoses will be accepted without verification.

Goal of the nudge

The goal of the nudge is to promote more realistic and balanced AI diagnoses. The aim is to encourage the AI to consider not only how symptoms match specific diseases, but also to consider their actual prevalence in the population. This is intended to reduce misdiagnoses, avoid unnecessary anxiety, and improve the quality of AI-supported health counseling.

    Needs analysis

    • The use of AI for self-diagnosis is on the rise
    • One-quarter of the German population uses AI applications before seeing a doctor
    • LLMs do not always provide consistent or transparent results
    • Many people blindly trust the recommendations of AI systems (automation bias)
    • There is a need for mechanisms that improve the quality and transparency of diagnoses

    Cause analysis

    • The focus is on the bias known as base rate neglect

    • This bias occurs when basic statistical information is not sufficiently taken into account

    • People and AI systems may tend to view conditions that match symptoms as a diagnosis, even though the probability of having that condition is low

    • This can lead to unnecessary tests, incorrect treatments, or psychological distress

    • Our prompt counteracts this: AI is explicitly instructed to take the base probability of diseases into account

Target Group

The nudge is aimed at health insurance policyholders who use AI-powered applications for self-diagnosis. Health insurance companies provide the nudge through their existing information channels, such as websites, apps, or brochures. This helps policyholders make more informed and realistic assessments of potential illnesses when using AI systems.

Added value of the nudge

  • Reducing misdiagnoses

  • Avoiding unnecessary anxiety

  • Avoiding unnecessary anxiety

CONTACT US

Fabienne David

Fabienne.david@stud.hshl.de

Johanna Schreiner

Johanna.schreiner@stud.hshl.de

Rinkje Drost

Rinkje.drost@stud.hshl.de