Researchers from Massachusetts Institute of Technology and elsewhere have recently come up with some unexpected findings regarding training of AI. While engineers generally try to match the simulated training environment for artificial intelligence (AI) agents as closely as possible with the real world where they are deployed, new research now shows that sometimes training in a completely different environment yields a better-performing AI agent.
READ: This robotics startup has a sustainable solution for refurbishing discarded electronics (January 24, 2025)
The latest research revealed that less uncertainty or “noise,” enabled the AI agent to perform better than a competing AI agent trained in the same, noisy world they used to test both agents.
Researchers are calling this phenomenon “the indoor training effect.”
“If we learn to play tennis in an indoor environment where there is no noise, we might be able to more easily master different shots. Then, if we move to a noisier environment, like a windy tennis court, we could have a higher probability of playing tennis well than if we started learning in the windy environment,” explains Serena Bono, a research assistant in the MIT Media Lab and lead author of a paper on the indoor training effect.
Bono has worked with co-author Spandan Madan, a final-year computer science PhD candidate at Harvard, who said: “This is an entirely new axis to think about. Rather than trying to match the training and testing environments, we may be able to construct simulated environments where an AI agent learns even better.”
READ: China disrupts AI market with DeepSeek: A better, cheaper version of ChatGPT? (January 27, 2025)
Ishaan Grover, a PhD student at MIT, who is also the cofounder of Catalan.ai, has also contributed to the research, as have Mao Yasueda, a graduate student at Yale University; Cynthia Breazeal, professor of media arts and sciences and leader of the Personal Robotics Group in the MIT Media Lab; Hanspeter Pfister, the An Wang Professor of Computer Science at Harvard; and Gabriel Kreiman, a professor at Harvard Medical School.
The research will be presented at the Association for the Advancement of Artificial Intelligence Conference held between February 25 and March 4, 2025 in Philadelphia, Pennsylvania.