What is Neuro-Symbolic AI?
Machine learning is a popular term that everyone knows these days. Even if you know that machine learning is a branch of AI, you may not have a clear idea about symbolic AI, if you are a beginner. Symbolic AI, also known as Good, Old-Fashioned AI (GOFAI) is an approach that uses tools such as logic programming, production rules, semantic nets and Frames, and learns to understand its environment. Symbolic AI involves the explicit embedding of human knowledge and behaviour rules into computer programs. The main difference between machine learning and symbolic AI is how the learning happens.
In machine learning, the algorithm use data to learn rules. In symbolic AI, the rules are created with human intervention and then hard-coded those rules into a static program.
In this article, I will give a brief idea regarding Neuro-symbolic artificial intelligence (NSAI). NSAI is an area of AI research that combines symbolic AI approaches with deep learning (DL) techniques. Here features can be extracted from data using DL approaches, and then symbolic systems are used to manipulate those features. The ultimate goal of a neuro-symbolic model is to close the loop between the data and insight. Insight is the capacity to gain an accurate and deep understanding of something while data is defined as facts or figures, or information that’s stored in or used by a computer. Here data insights are extracted, and external data about the environment or domain knowledge is added. Then it provides a better understanding also with the help of a query model which is a language model used in information retrieval.
According to literature, Neuro-symbolic models performed better than pure DL models, for image and video question answering tasks. Also, those models are accurate even for a small amount of training data. Therefore, neuro-symbolic models are also a good choice for scenarios where the availability of training data and training time is limited.
For instance, we used neural networks to identify things like what an object is and what kind of shape or colour a particular object is. Using symbolic reasoning can take it a step further and tell more exciting features about objects or their environment.
Let’s see an example for a better understanding…
The shapes in Figure 2 can be identified using a neural network model. But think about it if you want to ask questions like “What shape is farthest from the red square?”. Then the neural network alone cannot answer this question. Therefore, the neuro-symbolic system uses both logic and language processing to answer this question, similar to how a human would respond.
So this is the very basic idea of neuro-symbolic AI. Let’s deep dive into neuro-symbolic AI in future articles and will discuss an interesting project. So if you have enjoyed this article please hit the clap icon. Also, don’t forget to share your thoughts and knowledge by putting your valuable comments. :)
