Artificial General Intelligence

11Reinforcement Learning for AGI Model–Based versus Model-Free Approaches

Shivamma D.^1*, Shaila S.G.¹, Ramesh Chundi² and Monish L.¹

¹School of Engineering, Dayananda Sagar University, Bangalore, India

²School of Computer Applications, Dayananda Sagar University, Bangalore, India

Abstract

Artificial general intelligence (AGI) is attempting to develop a system that can function autonomously with respect to human-like levels of cognition and the capacity to make decisions and adapt to varied environments. Reinforcement learning (RL) underpins AGI as it provides ways to scoop up those behaviors by having the agents learn through interaction with its surroundings. These are model- free and model-based. RL plays a key part in how AGI systems learn and act in order to deal with different problems, suitable for reasoning and strategic problems that can be solved with planning and forward-looking predictions. Model-based RL creates a representation of the environment. Environmental models to build up a representation of the world to aid in making decisions is the underlying idea behind model-based RL. The prediction of system may wrong sometimes, The construction and nonlinearity of the environmental models. This work is not scalable when trying high-dimensional and dynamic environments. Model-free RL learns policies directly from the experience and thus avoids modeling the environment explicitly. It is well-suited to more complex, unstructured environments, also further scalable ...