Want Better GenAI Results? Try Speed Bumps
In this short video, Renée Richardson Gosline shares research on how to get more accurate results from generative AI without sacrificing speed.
Generative AI has vast potential to augment our work, now and in the future, but there’s a very real danger of human workers ceding too much control to machines and becoming complacent about mistakes. Maintaining a “human in the loop” is often touted as the antidote to catching errors, but MIT senior lecturer and research scientist Renée Richardson Gosline says that people frequently overestimate their ability to find flaws in GenAI-produced content.
In fact, her recent research indicates that we tend to “anchor,” or fixate, on generative AI’s answers, even when we’re aware of the likelihood of errors. According to recent research conducted by Gosline and Accenture, introducing the right kind of friction improves the overall accuracy of human work done in tandem with AI systems.
In this short video, Gosline speaks with MIT Sloan Management Review editor in chief Abbie Lundberg about the research findings and why putting cognitive speed bumps in place might be key to crafting the ideal human-AI work arrangement. Look for more on this topic in an article by Gosline coming later this spring from MIT SMR.
Video Credits
Renée Richardson Gosline is an MIT senior lecturer and research scientist.
Abbie Lundberg is the editor in chief at MIT Sloan Management Review.
M. Shawn Read is the multimedia editor at MIT Sloan Management Review.
