Sungkyunkwan University Cuts AI Assistant Costs and Improves Robot Learning Reliability

According to AI TIMES, a research team led by Professor Kwon Min-hye at Sungkyunkwan University has developed a technology that cuts the computing resources used by AI assistants to about one-third while raising task success rates to 78.3%. The study, introduced on October 13, presented a method that allows AI to ask for help only when needed instead of relying on external review at every step, along with a reinforcement learning technique that prevents robots from mistakenly learning accidental successes as stable strategies. The first technology focused on reducing the cost and speed problems of AI agents that handle complex tasks. The second technology is an approach aimed at lowering uncertainty in safety-critical areas such as robot control and autonomous driving by reflecting only repeatedly reliable behaviors in learning. The research team said the two methods address different problems, but both target the common challenge of autonomous AI that operates reliably while saving resources. The achievement is drawing attention as a foundation that could improve both AI operating costs and safety in real-world services and industrial settings.