Chapter 3: Real World AI and Uncertainty (Part 2)

The Three Learning Paradigms

Machine learning allows computers to discover patterns in data without being explicitly programmed with static rules. We explore the three core paradigms that dominate the field:

• Supervised Learning: The algorithm is trained using labeled data containing both inputs and the correct target outputs. Over time, it learns a mapping function to predict outputs for entirely new inputs. This is analogous to an elder guiding a young farmer during Lavani (harvest), showing them labeled examples of dried Sangri pods suitable for storage versus fresh Gobaliya leaves best suited for livestock feed.
• Unsupervised Learning: The algorithm receives unlabeled data and must discover inherent groupings, clusters, or structures on its own without external guidance or pre-assigned categories. This mirrors how pastoralists and experienced farmers naturally group different soils across their region into distinct clusters, such as sandy Retili, gravelly Magra, or loamy Matiyar, purely by observing common characteristics like drainage, texture, and naturally co-occurring thorny bushes, without needing formal textbooks or external definitions.
• Reinforcement Learning: An agent learns to make decisions by performing actions in an environment and receiving delayed rewards or penalties. This trial-and-error learning is identical to how a young Todo (camel calf) learns to navigate rocky desert terrain under the guidance of a Raika pastoralist, gradually mastering physical balance and route-finding through environmental feedback.