What is Symbolic AI?
Symbolic AI is artificial intelligence that focuses on processing symbols using logic rules in traditional programming. In the early days of AI research, engineers focused on manipulating abstract symbols to represent knowledge so that software could reason about it. However, later on, advances in neural networks, machine learning, and deep learning gained prominence and shifted focus. Present-day AI research combines symbolic AI with neural networks to combine the best of both worlds.
What is the difference between symbolic AI and other AI approaches?
Symbolic artificial intelligence (AI), also known as classical AI or good old-fashioned AI, involves manipulating symbols and logical reasoning to solve problems. In contrast, sub-symbolic AI combines machine learning and neural networks to mimic human reasoning and analyze incomplete or ambiguous data. Generally, we can differentiate symbolic and sub-symbolic AI by these characteristics.
Learning method
Symbolic AI stores information with human-readable symbols and engineers hard-code the rules and human knowledge into them. Conversely, sub-symbolic AI uses numerical distribution to represent the pattern it learns. Training on existing datasets is required to learn and identify patterns.
Working
The symbolic AI system evaluates each symbolic representation logically, based on the pre-defined rules, to provide an outcome. For example, you can create expert systems with symbolic reasoning that grant users access to buildings based on identification, time, and other rules.
In contrast, sub-symbolic AI exposes neural networks to massive datasets to learn and identify complex rules. Instead of applying fixed rules, the neural network stores learned knowledge using numerical representation, such as parameter weights. For example, scientists use sub-symbolic AI to create computer vision applications to identify dogs from various images.
Interpretability
Symbolic AI is more interpretable. Every decision a symbolic AI system makes can be traced to a logical function. On the other hand, sub-symbolic AI is less interpretable. Data scientists often face difficulties explaining why a neural network produces a specific result because of its black box-like mechanism.
Flexibility
Symbolic AI tools are only applicable to the problem they are developed for. Meanwhile, you can repurpose a sub-symbolic AI system for a related task by training the neural network with new data.
What are the benefits of the symbolic AI approach?
Symbolic AI forms the foundation of symbolic manipulation, reasoning, and representation in artificial intelligence research. Numerous AI fields, like natural language processing, industrial automation, and robotics, have benefited from classical AI research and development. For example, in natural language understanding, symbolic AI allows computer programs to analyze text by representing it with semantics, lexicons, and rules. It established the foundation for modern text processing and generative AI applications.
Symbolic AI excels at solving problems constrained by specific procedural requirements. It is more cost-efficient and faster to develop for certain problem cases. For example, engineers manipulate symbols with classical AI to raise alerts when sensors in a facility are triggered. Developers can create symbolic AI tools for various applications provided they can fully define the rules and parameters surrounding the problem.
What are the methods used in symbolic AI?
Data scientists use symbolic AI methods to represent abstract concepts with symbols that computers understand. We give several examples below.
Logic programming
Logic programming is a programming approach that applies formal logic to solve problems. The system can reason with facts and support the results with pre-written rules. The system applies reasoning statements, such as if pet A is a cat, then pet A is a mammal, to deduce a problem.
Production rules
Production rules govern how computers generate a string from a symbol and comply with the formal grammar. It replaces a starting symbol with sequences of terminal and non-terminal symbols. The process continues until the resulting string contains only terminal symbols.
Semantic reasoning
Semantic reasoning connects concepts and visualizes their relationship with semantic frames. A semantic frame provides a structured approach to understanding and processing information. AI systems can develop and apply human-like cognitive reasoning through the interconnected nodes.
Frames
Frames are a special data structure that stores information and attributes in AI processing. They are similar to how humans organize their thoughts into smaller compartments. Frames allow knowledge representation in a more structured manner, enabling AI systems to analyze data more effectively and accurately.
What are the limitations of symbolic AI?
Symbolic AI is versatile for solving well-defined problems but difficult to generalize for a wider problem range.
Limited problem solving
Symbolic AI software struggles to analyze and interpret ambiguous data. For example, it’s unsuitable for data that cannot be represented simply as true or false. Similarly, symbolic AI solutions cannot operate with incomplete information. If the circumstances of the problem change, engineers must rewrite the system's rules.
Difficult to adapt
You cannot train a symbolic AI tool to adapt to new data. Because symbolic AI works according to pre-determined rules, modification is required to adjust the software to evolving requirements.
Difficult to scale
Scaling symbolic AI solutions for complex use cases is challenging. Engineers must add more rules, which risk affecting the functionality and accuracy of existing ones.
How can you integrate symbolic AI systems with other AI techniques?
Despite its limitations, symbolic AI remains an essential part of machine learning. There are ongoing efforts to create AI systems that benefit from both approaches.
Neuro-symbolic AI
Neuro-symbolic AI combines machine learning algorithms with symbolic systems to create computational models with continuous learning and cognitive capabilities. The idea is based on the concept that humans think intuitively and also explicitly.
Neuro-symbolic AI can overcome the propositional fixation problem by creating a hybrid architecture. Propositional fixation refers to the challenge neural networks face in handling tasks that require logical statements or rules. Neural networks excel at making predictions based on patterns, but they struggle with tasks that involve understanding and manipulating discrete, structured information.
Interpretable machine learning
Symbolic AI can help make machine learning models more interpretable. Machine learning processes information through layers of neural networks that are not transparent to the observer. By integrating rule-based reasoning, data scientists can easily trace the result to the decision node.
How can AWS help?
AWS Bedrock is a fully managed service that provides organizations with leading foundational models for implementing and scaling generative AI applications. You can use AWS Bedrock to customize foundational models for specific use cases responsibly and safely. With AWS Bedrock, organizations can overcome scalability and data complexity issues that symbolic AI struggles with.
To support ongoing research and developmental machine learning works, use AWS SageMaker. Engineers build, train, and deploy custom machine learning models with the tools that support the machine learning development lifecycle. SageMaker also provides pre-built templates that help you implement text classification, image classification, and other models to complement symbolic AI algorithms.
Expanding symbolic AI usage on AWS by creating a free account today.
Browse all cloud computing concepts
Browse all cloud computing concepts content here:
Did you find what you were looking for today?
Let us know so we can improve the quality of the content on our pages