A new paper co-authored by Yale anthropologist Lisa Messeri and Princeton cognitive scientist M. J. Crockett highlights the dual-edged nature of artificial intelligence (AI) in scientific research.
While AI has the potential to significantly boost productivity, it also carries risks that could inadvertently narrow the scope of scientific inquiry and foster illusions of understanding among researchers.
Published in Nature, their Perspective article calls for a deeper conversation on how AI tools are utilized in science, urging caution against the blanket adoption of AI technologies without a thorough examination of their implications.
Messeri and Crockett introduce four archetypes of AI application that are currently generating excitement in the scientific community, each with its own promise and peril:
- AI as Oracle: Envisioned to efficiently comb through vast scientific literatures, these tools could help in formulating research questions but may also limit the breadth of inquiry to what the AI deems relevant.
- AI as Surrogate: Aimed at generating substitute data points, such as replacing human participants in studies, these applications risk overshadowing methods that are more complex but potentially richer in insights.
- AI as Quant: Tools designed to analyze large and complex datasets beyond human capabilities, which could overshadow traditional, nuanced analysis methods.
- AI as Arbiter: Intended to objectively evaluate the merit and replicability of scientific studies, potentially replacing the human peer-review process with what may be perceived as an unbiased AI judgment.
The authors caution against viewing these AI tools as more than mere instruments in the scientific process, warning that overreliance on AI could lead to “monocultures of knowing.” This phenomenon occurs when researchers prioritize AI-compatible questions and methods, thereby limiting the diversity of scientific exploration.
Such monocultures can foster illusions among scientists, making them believe they are exploring a wide range of hypotheses when they are, in fact, narrowly focusing on areas where AI can be applied.
Messeri and Crockett also discuss the danger of viewing AI tools as objective or superior to human scientists, which could lead to a “monoculture of knowers.”
This perspective risks ignoring the value of diverse scientific viewpoints and the inherent subjectivity of any tool, including AI, which reflects the biases and perspectives of its creators.
The paper stresses the importance of acknowledging science as a social practice enriched by diverse standpoints. Replacing the multiplicity of human perspectives with AI tools could undermine the progress made towards inclusivity in scientific research.
The authors urge scientists to consider not only the technical but also the social implications of integrating AI into their work, emphasizing the need for a balanced approach that leverages AI’s benefits while mitigating its risks.
The research findings can be found in Nature.
Copyright © 2024 Knowridge Science Report. All rights reserved.
