Can multi-agent AI systems help us make better decisions by balancing out our biases?
This experiment explored whether the complexity of a problem affects the impact of social influence, and how much social information is actually useful for the accuracy of results. It then went on to test whether an AI system in the form of a multi-agent system (MAS) was able to mitigate the negative impact of social bias and lead to more accurate decision-making in a group.
The experiment found that more information about other group members’ thinking increased group accuracy on complex tasks, but decreased it on simpler ones. The experiment also found that the group decisions mediated through a MAS were more accurate, irrespective of the task difficulty. The MAS helped to reduce the negative effect of herding by slowing down the process of decision-making and encouraging individual participants to continue exploring all available options.
From healthcare to democracy, platforms are being created to harness the collective intelligence of crowds. In medical diagnostics, for example, online platforms have emerged that connect patients to a network of doctors worldwide. This has great promise for opening up healthcare globally. But any group making collective decisions can be negatively influenced by social biases. For example, participants might make a wrong judgement about how competent another group member is, or be overly-confident of their own opinions. This can lead to important information being overlooked, and less accurate decisions being made.
Follow the project lead on Twitter: @vitotrianni