Current AI algorithms primarily rely on pattern matching on a large scale, lacking inherent reasoning abilities. This approach leads to Moravec's paradox, where AI excels in specific tasks but struggles with basic real-world challenges like locomotion. The limitation arises from the difficulty of capturing infinite real-world variability within the confines of data.
Navigating a non-rule-based world demands a generalised understanding and reasoning, which AI lacks currently. To counter this, machines must learn to think like us — or at least understand how humans make decisions. But how? Maybe they can take some inspiration from biological intelligence.
We grow by gaining a generalised understanding of the world and enhancing our reasoning abilities, which allows us to quickly specialise in various domains.
Driving is essentially a derivative of locomotion capabilities which we develop during our infant years. That allows us to grasp the art of driving in a few days from a driving school.
Having that generalized understanding enables us to adapt our learning across vehicle form factors and geographies, without any need to re-learn and still be able to make safe decisions in adversarial scenarios.
The next-gen foundational model architecture inspired by the cognitive inference capabilities of the human brain that transcends beyond language and vision modalities to develop inherent understanding of the world enabling reasoning, adaptability and explainability to build truly generalised autonomous agents.
We humans have a world model that we build by observing the world to develop intuitive understanding and reasoning capabilities to guide on what is likely, what is plausible, and what is impossible. The world model is a cognitive ‘model’ of how the world works by capturing causality and intuitive physics for understanding the environment, agent’s intent and behavior's to make explainable decisions.
