Observational Learning and Machine Learning

Researchers find breakthrough in neuroscience that will contribute to machine learning.

Researchers from the University of California Los Angeles (UCLA) and the California Technical Institute (Caltech) studied the activity of individual neurons when humans are learning through observation. The study used abstract levels of computational models that reflected in the activity of individual neurons and supported human behavior and interaction. Michal Hill, the study’s research group leader, called the study groundbreaking in that it “transcend[s] different levels of neuroscience.”

The study called for ten patients suffering from epilepsy who were instructed to play a card game. When the patients observed other players, the neurons created a complex learning language and the electrodes reflected the changes in neural behavior caused by observing the other players. When observing others, the neurons in the rostral anterior cingulate cortex (rACC) registered the expected value of an observed choice and the prediction error after the outcome was revealed. The parameters were used by the brain to learn from others’ experiences rather than learning from their own mistakes.

Machine learning is largely created the same way; machines are taught by observations and reactions. Machine learning is expensive to develop, and building robots that can move on their own with required special sensitivity is a time-consuming task. However, many large organizations consider it a worthwhile undertaking and many are willing to share their information with developers willing to use it. Elon Musk opened a lab for developers to work on AI as part of a nonprofit this summer. Facebook is launching Facebook Artificial Intelligence Research (FAIR) in order to help machines register two-dimensional images. 

Leave a reply