Commonsense knowledge in AI refers to the generic, broad-spectrum knowledge that an average person tends to innately have about the world around them. It involves understanding everyday facts, assumptions, the functionality of objects, and the likely progression of events.


Imagine you’re playing with your toys. You know that if you roll a ball, it will move. You know that a teddy bear cannot drink milk. This is commonsense - understanding simple facts about the world. In the field of AI, commonsense knowledge is about teaching computers this kind of understanding.

In-depth explanation

Commonsense knowledge refers to basic information that a human mind generally assumes or knows about the world, often without formal teaching. This includes contextual facts about objects, properties, categories, events, time, and people’s beliefs and desires, among other things.

In the realm of AI, embedding such commonsense knowledge is a monumental task. The nuances and subjectivity associated with such knowledge make it challenging for machines to learn and understand. For instance, let’s consider the simple sentence, “The ice cream melted because it was hot.”. A human can instantly understand that ‘it’ refers to the environment and not the ice cream, but for an AI system to grasp this contextual reference, it must have some prior commonsense knowledge coded into it.

AI systems are often fed information in the form of structured databases, rules, or learned from large datasets. Commonsense knowledge graphs are a popular method of explicitly encoding this information, where nodes represent concepts/entities, and the edges characterize the relationship between these entities.

However, these methods often fail to capture the complete essence of commonsense knowledge due to the wide-ranging nature of the facts, assumption involved, and lived experiences humans hold, which inherently shape this commonsense knowledge.

Machine learning techniques, including Deep Learning and Reinforcement Learning, are increasingly being applied towards developing models that can learn and adapt to represent such nuanced and context-driven knowledge themselves.

However, despite several years of research and advances in the field, creating AI systems that can fully replicate or comprehend human commonsense knowledge remains one of the grand challenges of AI.

Knowledge Graphs, Semantic Networks, Machine Learning (ML),, Deep Learning, Reinforcement Learning (RL),, Natural Language Processing (NLP),, Contextual Understanding.