Active Inference

Active Inference

Active Inference is a theoretical framework that combines perception, action, and learning. It is a concept derived from the Free Energy Principle, a theory proposed by Karl Friston, which suggests that all adaptive systems, including the brain, resist a state of disorder. Active Inference is a process that allows systems to maintain a state of order by minimizing the difference between their predictions and sensory inputs.


Active Inference is a process where an agent actively samples its environment to reduce uncertainty about its state. It is a predictive coding framework where the agent uses its internal model to predict sensory inputs and then updates this model based on the difference between the predicted and actual sensory inputs. This difference is known as the prediction error. The agent then uses this updated model to guide its actions, aiming to minimize future prediction errors.

How Active Inference Works

Active Inference operates on the principle that agents seek to minimize their Free Energy, a measure of surprise or uncertainty. The agent does this by updating its internal model to better predict sensory inputs and by acting in ways that bring its sensory inputs in line with its predictions.

The process involves two key steps:

  1. Model Updating: The agent updates its internal model based on the prediction error. This is done using Bayesian inference, a statistical method that updates the probability of a hypothesis as more evidence becomes available.

  2. Action Selection: The agent selects actions that are expected to minimize future prediction errors. This is done using a process called planning as inference, where the agent infers the most likely sequence of actions that will lead to desired outcomes.

Applications of Active Inference

Active Inference has wide-ranging applications in various fields, including neuroscience, artificial intelligence, and robotics. In neuroscience, it provides a unified theory of brain function, explaining how the brain predicts sensory inputs and updates these predictions based on actual sensory data. In artificial intelligence and robotics, it provides a framework for designing systems that can learn from their environment and adapt their behavior based on this learning.

Importance of Active Inference

Active Inference offers a comprehensive understanding of how agents interact with their environment. It provides a theoretical foundation for understanding perception, action, and learning as a unified process. This makes it a valuable tool for researchers and practitioners in fields like neuroscience, artificial intelligence, and robotics.

Active Inference vs. Reinforcement Learning

While both Active Inference and Reinforcement Learning are frameworks for understanding how agents learn from their environment, they differ in several key ways. Active Inference is a model-based approach, where the agent maintains an internal model of the world and uses this model to guide its actions. Reinforcement Learning, on the other hand, is typically model-free, where the agent learns a policy directly from rewards and punishments without maintaining an explicit model of the world.

Active Inference also differs from Reinforcement Learning in its treatment of uncertainty. In Active Inference, the agent actively seeks to reduce uncertainty by sampling its environment. In Reinforcement Learning, uncertainty is typically treated as a source of exploration, where the agent takes random actions to discover new information.

Further Reading

For those interested in a deeper understanding of Active Inference, the following resources are recommended:

  • Friston, K. (2010). The free-energy principle: a unified brain theory? Nature Reviews Neuroscience, 11(2), 127-138.
  • Friston, K., FitzGerald, T., Rigoli, F., Schwartenbeck, P., & Pezzulo, G. (2017). Active inference: a process theory. Neural computation, 29(1), 1-49.
  • Friston, K., & Frith, C. (2015). Active inference, communication and hermeneutics. Cortex, 68, 129-143.