Artificial Intelligence framework inspired by biological systems developed

Serdar HocamAuthor & Editor

Researchers have introduced a new approach that transforms the learning processes of artificial intelligence agents by equipping them with internal states.

◉ 0 views
A life-inspired framework for more autonomous and adaptive AI

Scientists have proposed a new framework called interoceptive artificial intelligence, inspired by how living organisms monitor their internal balances. This approach enables artificial agents to use internal states as context in their decision-making processes.

Background of the new framework

The research team developed the interoceptive artificial intelligence approach by drawing inspiration from the ability of living beings to regulate their internal conditions. Published in the journal Nature Machine Intelligence, this study aims to make artificial intelligence more autonomous.

Team and institutions

The study was conducted by a team led by Associate Professor Woo Choong-Wan from the Center for Neuroscience Imaging Research at the Institute for Basic Science, Professor Hong Seok Jun from Sungkyunkwan University, and Professor Karl J. Friston from University College London.

Difference from traditional methods

While traditional reinforcement learning relies on external rewards, the new framework provides artificial agents with clearly defined internal states. These states function as a continuous source of context for learning and decision-making processes.

EVAAA test environment

To examine this approach, the researchers developed a three-dimensional virtual survival benchmark called EVAAA. In this environment, as agents navigate among resources, obstacles, and predators, they regulate variables such as satiety, hydration, and body temperature.

Potential for robotics and physical domains

This developed framework can provide significant advantages, particularly in the fields of physical artificial intelligence and robotics. It allows internal conditions, such as battery levels or motor temperature, to directly influence learning.