Mijn bachelor opleiding is afgerond met een eind project over de vertrouwens relatie tussen mensen en een artificial intelligence systeem. Specifiek gaat dit onderzoek om het effect van error nauwkeurigheid in advies van een AI systeem op verloop van het vertrouwen in het systeem over tijd.
In the last couple of years, Artificial Intelligence (AI) is used more extensively. A highly capable AI can be better than humans for difficult tasks, but trusting advice from a computer seems difficult. Benevolence, competence, and integrity are key factors that determine trust in an AI system. Trust over time is a trajectory in which you can gain trust at a slow pace and lose trust very quickly. In this report, the effect of error accuracy on the trust in the AI advisory system and the long-term effects are investigated. Participants showed more trust in a system that was more reliable than in a less reliable system. Over time a low-accurate system shows a decrease in trust and a highly accurate system a small increase in trust. This trust is measured by a self-assessment of the participants but also calculated based on the Weight of Advice. The result of this study can show how Artificial Intelligence advice can be made more trustful and how it can be used more in difficult tasks.
Keywords: trust, algorithms, artificial intelligence, accuracy, reliability, legal advice, recommendation