Benefits and Risks of Using AI Agents in Research

AI agents are the newest addition to our repertoire of intelligent machines. Powered by large language models and connected to other tools and robots, AI agents not only perform tasks, but they can design a workflow and employ other available tools and digital resources like datasets to carry out a compound task. AI agents are also being employed by researchers to conduct multi-step experiments or stimulate human behavior to perform social science studies. AI agents can transform various research endeavors through efficiency gains, increased productivity and reducing human errors in research and education. However, given their limitations like errors and biases, they also introduce specific risks. In this paper we highlight six specific risks such as errors and biases, loss of research jobs, loss of critical thinking and deskilling scientists, increased possibility of machine-driven misconduct and the increased rate of unverified research results. We conclude that researchers need training in areas such as algorithmic and AI literacy, to be able to identify biases, errors, and other problems in tasks conducted by AI agents. Institutions need to pay more attention to the development of these technologies and provide guidance about their use.