The ingredients of an NSAI system
With AI gaining more traction, in August of 2019, strong research efforts began to enable common-sense and reasoning abilities in AI systems by reverse engineering the brain of human babies. As the name implies, the recipe of neuro-symbolic programming involves two main ingredients: NNs and symbolic programming. We will explore these two ingredients using the Compositional Language and Elementary Visual Reasoning (CLEVR) example case. CLEVR is a dataset of 100,000 computer-generated scenes portraying 3D shapes (https://cs.stanford.edu/people/jcjohns/clevr/). The objective of this dataset is for AI to reason about these images and be able to answer questions regarding the said images—for example: How many spheres are in the image?
The symbolic ingredient
Motivated by their observations, the researchers highlighted one key aspect of the reasoning abilities of humans (and other organisms, for that matter): world knowledge. We can reason about...
