AI Explained

Chapter 5 Learning AI

From the moment we are born, humans—and indeed all higher-order animals—are not equipped with all the knowledge, skills, and behaviours needed to live and thrive. The vast majority of what is required must be learnt over time. This learning is fundamentally experiential, involving processes such as trial-and-error, making mistakes, and gradually adapting to the environment. It is through these costly and sometimes error-prone processes that individuals acquire the abilities necessary for survival and success.

Just as infant children cannot begin life with all the knowledge, rules, and processes they will need, early AI researchers soon recognised that developing complete, preprogrammed systems was too ambitious a goal for most practical use cases. The demands of early approaches, such as symbolic AI, proved too difficult to meet in practice, as encoding every possible rule and piece of knowledge was both time-consuming and inherently limited.

Instead, researchers began to look towards the paradigm of adaptation through learning. Rather than attempting to specify everything in advance, they explored ways for AI systems to acquire knowledge, skills, and behaviours by interacting with data and their environment—mirroring the way living beings learn and adapt over time.

This shift laid the groundwork for learning AI, algorithms, and models that could improve themselves through “experience,” opening the door to more flexible and capable systems. The goal was ...