Do humans and AI function better together or separately?

Just last week, the MIT Center for Collective Intelligence published a comprehensive meta-analysis related to the collaboration of humans and AI: “When Combinations of Humans and AI Are Useful.” It appeared in the journal Nature Human Behaviour and enlightened us on how to think about how humans and AI can work together.

Its finding is that although human input and AI systems collaborating to make a decision in most cases hardly meet expectations, on the contrary, human-AI partnerships in creative tasks are promisingly considered.

The following study was conducted by Michelle Vaccaro, Abdullah Almaatouq, and Thomas Malone to establish the conditions under which collaborations between humans and AI could be most effective and to provide recommendations on how organizations could design and implement rules and structures that allow human-AI collaborations to flourish.

The researchers carried out their analyses based on a total of 370 different task-oriented outcomes involving combinations of humans and AI. Of 106 published between January 2020 and June 2023, these studies found that human teams and AI generally outperformed individual humans but fell short of AI-alone performance. One critical observation from the research was what the authors have termed a “lack of human-AI synergy.” This means that, on the whole, the average performance of human-AI systems was not superior to the best performance that could be achieved by humans alone or AI systems alone on the metrics evaluated in this study. This would appear to indicate that, in those specific cases reviewed, either humans or AI on their own would have done better than the combined human-AI efforts that were tested.

Certain important factors were identified by the research team that affect the nature and degree of effectiveness of human and AI collaboration. For example, tasks requiring decision-making-needed tasks, such as the identification of deep fakes, the prediction of market demands, and diagnosing medical conditions, were studied. The results indicated that these collaborations, while superior in some aspects, generally fell short of standard performances. In dramatic contrast, human-AI teams outperformed the best results of humans or AI alone in many creative tasks: summing up information in social media, answering questions in conversational chat, and creating new content and images.

They believe this finding may impart critical lessons to organizations that seek to introduce AI technologies effectively in their work environments. They suggest that organizations should first identify whether human-AI partnership, in fact, produces superior performance compared to either human or AI-based work alone, and identify specific areas where human productivity would be enhanced using AI while they establish baseline rules and create solid frameworks for application and usage responsibly and effectively.

Thank you for reading this post, don't forget to follow my whatsapp channel