Scientists have investigated what makes humans and artificial intelligence (AI) work most efficiently together. The results suggest that AI sometimes is nothing but a roadblock.
Many industries are actively adopting AI, with 75% of surveyed workers saying they use it in their jobs and almost 80% of company leaders saying that AI should be adopted to stay competitive.
While some still fear that this technology has come to take our jobs, AI evangelists applaud its ability to define medical decision-making and boost scientific research. However, it’s evident that the future of AI is not just black and white.
A new study conducted by researchers from the Center for Collective Intelligence (CCI) at the Massachusetts Institute of Technology (MIT) revealed when humans and AI work most efficiently together.
The results showed that humans and AI underperformed when working together on tasks where decisions needed to be made. Such tasks included classifying deepfakes, forecasting demand, and diagnosing medical cases.
However, in creative tasks such as summarizing social media posts, responding to chats, or generating visual content, AI and human collaboration outperformed those working independently.
According to researchers, this could be explained by the dual nature of human thinking and AI algorithms. Creative tasks require a human touch. For example, image creation needs artistic inspiration, primarily a human prerogative, whereas AI is more adept when it comes to repetitive tasks, such as generating detailed images.
“There’s a prevailing assumption that integrating AI into a process will always help performance — but we show that that isn’t true,” said one author of the research paper, Michelle Vaccaro. “In some cases, it’s beneficial to leave some tasks solely for humans and some tasks solely for AI.”
Researchers are calling to assess and consider tasks in which AI can outperform humans or tasks where AI can work independently or alongside humans, depending on which approach best suits the task or could lead to better results.
“There is a lot of potential in combining humans and AI, but we need to think more critically about it,” said Vaccaro. “The effectiveness is not necessarily about the baseline performance of either of them, but about how they work together and complement each other.”
The methodology
Researchers selected 5,126 AI research papers published in various journals over the period of three years. All the selected papers reported the results of 106 unique experiments on AI and human collaboration. Researchers have identified 370 different results on how AI and humans perform together and separately.
All the studies compared three ways of performing tasks – human-only systems, AI-only systems, and human-AI collaborations. Later on, researchers applied a meta-analysis technique to draft the overall tendencies.
The study was published on October 28th in Nature Human Behaviour.
Your email address will not be published. Required fields are markedmarked