Knowledge based systems

What Are Knowledge-Based Systems?

A knowledge-based system (KBS) is a computer program that leverages a structured collection of information, known as a knowledge base, to solve complex problems and provide expert-level recommendations. These systems represent a significant branch within the broader field of Artificial Intelligence (AI), designed to emulate the decision-making abilities of human experts in specific domains. Unlike traditional programs that follow predefined algorithms, a knowledge-based system can reason and infer new facts from its existing knowledge, making it particularly useful in scenarios requiring specialized understanding or complex judgment.

History and Origin

The concept of knowledge-based systems emerged prominently in the artificial intelligence landscape during the 1970s and truly flourished in the 1980s, often used interchangeably with "expert systems." Early AI research faced limitations in general problem-solving; this led to a shift toward capturing domain-specific knowledge. One of the earliest and most influential projects was DENDRAL, initiated in 1965 by Edward Feigenbaum and Joshua Lederberg at Stanford University. DENDRAL was a chemical analysis system designed to deduce molecular structures from scientific data. Feigenbaum is widely recognized as a pioneer in this area, sometimes referred to as the "father of expert systems."²⁹, ³⁰, ³¹, ³²

The proliferation of these systems in the 1980s marked a period of rapid growth for AI, often dubbed the "AI boom."²⁷, ²⁸ Businesses adopted them for various applications, with some estimates suggesting that two-thirds of Fortune 500 companies utilized expert systems in their daily operations.²⁵, ²⁶ However, challenges, including the difficulty of acquiring and maintaining vast amounts of explicit knowledge, eventually led to a period known as the "AI winter" in the late 1980s and early 1990s, where funding and interest in this approach diminished.²², ²³, ²⁴

Key Takeaways

• Knowledge-based systems (KBS) are AI programs that solve complex problems using a dedicated knowledge base.
• They aim to replicate human expert decision-making by reasoning with stored facts and rules.
• The primary components are a knowledge base (data repository) and an inference engine (reasoning mechanism).
• KBS found significant applications in the 1980s, particularly in the form of expert systems.
• While their popularity waned, the underlying principles of knowledge representation continue to influence modern AI development.

Formula and Calculation

Knowledge-based systems do not typically rely on a single mathematical formula for their core operation. Instead, their "calculation" involves logical inference and symbolic reasoning within the inference engine, processing information stored in the knowledge base.

The process often involves:

• Rule-based Logic: Applying "if-then" rules to existing facts to deduce new facts or reach conclusions.
• Heuristics: Using "rules of thumb" or practical guidelines to guide the problem-solving process.
• Probabilistic Reasoning: In more advanced systems, employing statistical models to make decisions based on probabilities.

There isn't a universally applicable formula like those found in quantitative finance. The "calculation" is a symbolic manipulation rather than a numerical one.

Interpreting Knowledge-Based Systems

Interpreting the output of a knowledge-based system involves understanding the reasoning path it followed to arrive at a conclusion. Unlike some modern Machine Learning models that can be opaque, traditional knowledge-based systems, especially rule-based ones, are often designed to explain their reasoning. This means they can show the specific facts and rules used to reach a particular decision or recommendation.

For instance, in a financial context, a knowledge-based system designed for credit risk assessment might not just give a credit score but also explain why a loan was approved or denied by listing the criteria met or not met (e.g., "Loan denied because applicant's debt-to-income ratio exceeds threshold X and employment history is less than Y years"). This explainability is a key advantage, fostering trust and enabling human experts to validate or refine the system's logic.

Hypothetical Example

Imagine a small financial advisory firm, "DiversiPath Advisors," wanting to automate initial client suitability assessments for basic investment products. They decide to implement a simple knowledge-based system.

Scenario: A new client, Sarah, wants to open an investment account.

DiversiPath's KBS Rules (simplified):

• Rule 1: IF client age < 18 THEN ineligible for investment account.
• Rule 2: IF client age >= 18 AND client annual income >= $50,000 THEN eligible for standard investment account.
• Rule 3: IF client age >= 18 AND client annual income < $50,000 THEN recommend savings plan.
• Rule 4: IF client has prior bankruptcy THEN ineligible for standard investment account.

Sarah's Information:

• Age: 30
• Annual Income: $70,000
• Prior Bankruptcy: No

KBS Process:

1. The inference engine receives Sarah's data as inputs into the knowledge base.
2. It checks Rule 1: Sarah's age (30) is not less than 18. Rule 1 does not apply.
3. It checks Rule 2: Sarah's age (30) is >= 18 AND her annual income ($70,000) is >= $50,000. Both conditions are met.
4. It checks Rule 4: Sarah has no prior bankruptcy. Condition is met (no disqualification).

KBS Output: "Client Sarah is eligible for a standard investment account."

This hypothetical example illustrates how a knowledge-based system applies predefined rule-based systems to client data to reach a specific recommendation, mirroring a basic decision-making process an advisor might follow.

Practical Applications

Knowledge-based systems, or their evolved forms, continue to have practical applications across various sectors, including finance, even as more advanced AI techniques like deep learning gain prominence.

In investing and markets, early applications included:

• Portfolio Management: Systems assisted in selecting assets based on predefined investment rules and client profiles.
• Risk Management: Helping identify and assess risks by applying rules related to market conditions, security characteristics, or client exposures.
• Fraud Detection: Rule-based systems were, and still are, used to flag suspicious transactions that deviate from established patterns.²¹

In banking, modern iterations of these systems contribute to:

• Customer Service: Chatbots often employ a form of knowledge base combined with Natural Language Processing to answer common queries.²⁰
• Loan Approvals: Automating parts of the loan application process by applying eligibility rules to applicant data.¹⁸, ¹⁹
• Regulatory Compliance: Assisting financial institutions in adhering to complex regulations by systematizing and checking rules against operational data, enhancing operational efficiency.¹⁶, ¹⁷

The CFA Institute, a global association for investment professionals, has emphasized the growing integration of AI and big data in the investment process, noting how these technologies augment investment and business outcomes.¹⁴, ¹⁵

Limitations and Criticisms

Despite their historical significance and utility, knowledge-based systems, especially in their early forms, faced several limitations:

• Knowledge Acquisition Bottleneck: Extracting and formalizing expert knowledge into a structured knowledge base proved to be an arduous and time-consuming task, often requiring specialized "knowledge engineers."¹³
• Brittleness: These systems perform well within their predefined domain but struggle with situations outside their programmed knowledge. They lack common sense or the ability to generalize, making them "brittle" when encountering unexpected scenarios.
• Maintenance Challenges: As knowledge domains evolve, maintaining and updating the rule sets and facts within a knowledge-based system can become complex and expensive.¹²
• Scalability Issues: For very large and complex problem domains, the number of rules and facts can grow exponentially, leading to performance issues and difficulty in managing the system.¹¹
• Lack of Learning: Traditional knowledge-based systems do not inherently learn from new data or experiences in the way modern Machine Learning models do. Their knowledge is explicitly programmed, not discovered.¹⁰

The National Institute of Standards and Technology (NIST) has developed an AI Risk Management Framework (AI RMF) to address broader concerns associated with AI systems, including issues of bias, transparency, and accountability.⁸, ⁹ These concerns are particularly relevant to knowledge-based systems, as even explicitly programmed rules can embed algorithmic bias if the human experts' knowledge contains implicit biases or if the underlying data analytics are flawed.⁷ The complexity of some AI systems can also make it challenging to identify the root cause of errors or explain certain decisions.⁶

Knowledge-Based Systems vs. Expert Systems

The terms "knowledge-based system" (KBS) and "expert system" are closely related and often used interchangeably, but there's a subtle distinction in their emphasis:

Essentially, all expert systems are knowledge-based systems because they rely on a formalized knowledge base and a reasoning mechanism. However, not all knowledge-based systems are necessarily expert systems. A KBS could be developed for tasks that don't strictly require "expert" human-level judgment, but rather systematic application of structured knowledge. The confusion arises because many of the earliest and most prominent knowledge-based systems were indeed expert systems.

FAQs

What are the main components of a knowledge-based system?

The two primary components of a knowledge-based system are the knowledge base and the inference engine. The knowledge base stores facts, rules, and heuristics about a specific domain. The inference engine processes this information, applying logical rules to deduce new facts or solve problems.², ³

How does a knowledge-based system differ from traditional software?

Traditional software typically follows a set of fixed instructions (algorithms) to perform tasks, with knowledge implicitly embedded in the code. A knowledge-based system, however, separates the knowledge from the processing logic. It explicitly represents knowledge in a structured format, allowing the system to reason and make inferences, often with the ability to explain its decisions, making it more flexible for problems requiring expertise.¹

Can knowledge-based systems learn and adapt?

Traditional knowledge-based systems primarily rely on explicitly programmed knowledge and do not inherently learn or adapt from new data in the way modern Machine Learning or Predictive Models do. While their performance can improve through iterative refinement of their knowledge base by human experts, they lack autonomous learning capabilities. However, contemporary AI often integrates knowledge-based approaches with machine learning for enhanced capabilities.