Researchers have recently succeeded in developing a new artificial intelligence technology capable of understanding human preferences, marking an important step toward safer and more harmonious interaction between humans and robots across various aspects of daily life.
Researchers from the Korea Advanced Institute of Science and Technology (KAIST) in South Korea have taken a significant step toward expanding real-world AI applications by developing a new technology that allows machines to autonomously learn human preferences and evaluation criteria from a limited number of video clips.
Professor Chang Tae-you, from the School of Electrical Engineering, presented a new approach called VOTP (Video-based Optimal TransPort Preference), which aims to enable intelligent systems to understand human preferences without requiring large volumes of pre-labelled data.
Key applications of this approach include industrial robots operating in hazardous environments, self-driving vehicles capable of navigating complex traffic conditions, and surgical robots.
Until recently, the development of robot behaviour faced a significant obstacle: the need for AI to understand which behavioural patterns humans consider appropriate or inappropriate.
To achieve this, developers relied on collecting tens of thousands of human evaluations.