Researchers at the Massachusetts Institute of Technology (MIT) say they have developed an artificial intelligence (AI) that can accurately predict human interactions.
The AI, created by MIT's Computer Science and Artificial Intelligence Laboratory, can watch a video and determine within a second if two people in the scene will hug, kiss, shake hands, or high-five. After five seconds, the AI also can predict the appearance of certain objects, such as anticipating that a coffee mug will appear after viewing a microwave in a scene.
The AI's machine-learning system relies on neural network-based algorithms to learn from large datasets, with sets of training data derived from more than 600 hours of shows such as "The Office" and "Desperate Housewives." Instead of the traditional method of examining individual pixels of a frame to predict future frames, the AI determines possible outcomes from entire freeze-frames.
Thus far, MIT's AI is successful in predicting interactions only 43% of the time, but the researchers say robots in the future could apply this knowledge to social exchanges with humans. "There's a lot of subtlety to understanding and forecasting human interactions," says MIT's Carl Vondrick. "We hope to be able to work off of this example to be able to soon predict even more complex tasks."
View Full Article
Abstracts Copyright © 2016 Information Inc., Bethesda, Maryland, USA
No entries found