Examining the Effect of AI and Human Guidance on Human Decision Making
Activity: Talk or presentation types › Oral presentation
Artificial Intelligence (AI) is increasingly used to assist decision-making in public and private sector organisations. However, there is a lack of research and understanding about whether humans use algorithms effectively when making decisions. The current work examined whether humans successfully utilise guidance purportedly provided by humans or an algorithm in a decision-making task. Participants (N = 295, Mage = 33.79) provided judgements on the authenticity of 80 faces (40 real, 40 artificial) presented alongside guidance ostensibly from human experts or an algorithm that was correct only 50% of the time. Analyses revealed that individuals were equally likely to make judgements consistent with guidance from humans or an algorithm but, encouragingly, were more likely to respond consistently with guidance when it was correct. Signal detection analyses identified a tendency to classify faces as real, alongside a reduced ability to discriminate between real and synthetic faces with greater self-reported trust in AI, when participants received algorithmic guidance. These results are promising as they suggest humans can follow AI guidance when it is correct and disregard it when it is not. However, further work is required to understand human-algorithm interactions when guidance accuracy is higher and therefore is considered more trustworthy.
Co-author(s): Sophie Nightingale; Grace-Rose Whordley (DSTL); Itiel Dror (Cognitive Consultants International)
Title | Behavioural and Social Sciences in Security 2024 Conference |
---|
Abbreviated title | BASS24 |
---|
Date | 16/07/24 → 18/07/24 |
---|
Location | St Andrews University |
---|
City | St Andrews |
---|
Country/Territory | United Kingdom |
---|
Degree of recognition | International event |
---|