0%
Caution: JavaScript execution is disabled in your browser or for this website. You may not be able to answer all questions in this survey. Please, verify your browser parameters.

Understanding and trust in AI: a study on the impact of explanations

This survey is part of a study aimed at understanding how people interpret decisions made by an artificial intelligence (AI) model.

During the test you will be shown instances (the students under study), for which we will ask you to:

  • Predict what decision the AI model will make;

  • Assess your level of confidence in that prediction;

  • Observe an explanation provided by the AI;

  • Re-evaluate your prediction and your level of confidence.

The test will take about 10-15 minutes.
Thank you for your collaboration!

BEFORE PROCEEDING READ THIS SECTION

An explanation serves to make us understand why an artificial intelligence system has made a certain decision or given a certain result.

Imagine that an AI says: ‘In my opinion, this student will perform high in the exam.’

You might ask, "Why high? What led to this conclusion?"

The explanations help to answer this question.

 

 

 

Global explanation: Look at all the IA decisions and try to explain how it thinks in general. (e.g. on average, studying matters more than sleeping, TV has little influence, etc.).

Global explanation example:

 

Local explanation: show which characteristics weighed most (e.g. studied a lot, slept well, etc.) but for the specific case. E.g. for one student the hours of sleep or study impacted more than for another student and so on.

Local explanation example:

 

 

Counterfactual explanation: Shows you what would have changed if certain aspects had been different. For example, a student might score higher if he/she used less social media.

Counterfactual explanation example:


Study hours: 2.2 to 3.5

Netflix hours: 3.5 to 2.0



 

This survey is anonymous.

The record of your survey responses does not contain any identifying information about you, unless a specific survey question explicitly asked for it.

If you used an identifying token to access this survey, please rest assured that this token will not be stored together with your responses. It is managed in a separate database and will only be updated to indicate whether you did (or did not) complete this survey. There is no way of matching identification tokens with survey responses.