Neural Symbol AI: Innovation through AI Composition

2020-01-28

With the development of the world, new era technologies are bound to maintain their latest forms. It is not enough for these forms to be considered new technologies; on the contrary, they need to evolve to self-renew as demand grows. Many experts in the field of artificial intelligence believe that it needs to change over time. There are many types of AI technologies, some of which are at the forefront, while others are at certain levels below (such as neural networks and symbolic AI). AI innovation certainly does not mean bringing changes to the entire field. The fusion of small combinations and different AI technologies can also bring inventions, innovations, and better implementations into the fast-paced world.


According to David Cox, a researcher at MIT and IBM Watson AI Lab, AI urgently needs to change. However, his suggestion for this change is an obscure term called "neuro-symbolic AI."


As he pointed out, neuro-symbolic AI is not a completely new way of AI, but a combination of two existing approaches to building thinking. The "symbolic AI" part refers to the first mainstream approach to creating artificial intelligence, while the "neuro" part refers to deep learning neural networks.


Neural networks work differently from symbolic AI because they are data-driven. In order to explain certain content to a symbolic AI system, a person needs to explicitly provide all the information for it to correctly identify, while training a neural network to perform this task only requires showing it thousands of images. Once it becomes smart enough, it will be able to recognize the object and compose its own similar objects. Such objects may never have actually existed in the real world.However, the idea behind neuro-symbolic AI is to combine these methods to integrate learning and logic. It is evident that neural networks will simplify the world into symbols instead of relying on human programmers to do it for them, making symbolic AI systems smarter. Additionally, symbolic AI algorithms will help transform common sense reasoning and domain knowledge into deep learning. This could subsequently lead to significant advancements in AI systems in handling complex tasks, involving everything from autonomous vehicles to NLP, while requiring less data for training.


AI researcher David Cox said, "Neural networks and symbolic ideas are really perfect complements. Because neural networks give you the answer to go from the messy state of the real world to a symbolic representation of the world, where you can find all the correlations in the image. Once you have symbolic representation, you can do some really magical things in terms of reasoning."

RECENT AI NEWS

RECENT AI TOOLS