Q1Artificial Intelligence Techniques

(a) Application Areas of Artificial Intelligence

Artificial Intelligence has found application across an exceptionally broad range of domains, reflecting its fundamental character as a general-purpose approach to automating tasks that traditionally required human intelligence and judgment.

• Expert systems: AI-based systems that emulate the decision-making and diagnostic capability of a human domain expert, widely used in medical diagnosis (assisting physicians in identifying diseases from symptoms and test results), financial analysis (credit risk assessment, fraud detection), and industrial fault diagnosis (identifying the likely cause of equipment malfunction from observed symptoms).
• Natural Language Processing (NLP): enabling computers to understand, interpret, and generate human language, powering applications such as machine translation, chatbots and virtual assistants, sentiment analysis, automatic text summarization, and voice-activated personal assistants.
• Computer vision and image processing: enabling machines to interpret and analyze visual information from images and video, used in facial recognition systems, medical image analysis (detecting tumors or abnormalities in X-rays/MRI scans), industrial quality-control inspection, and object detection/tracking for security and surveillance.
• Robotics: AI techniques provide the perception, planning, and decision-making capability underlying autonomous and semi-autonomous robots, used in industrial manufacturing automation, warehouse logistics, surgical assistance robots, and exploratory robots for hazardous or inaccessible environments.
• Autonomous vehicles: self-driving cars and drones rely heavily on AI for perception (interpreting sensor data from cameras, radar, and lidar), path planning, and real-time decision-making in dynamic traffic environments.
• Gaming: AI techniques (including classical search algorithms, minimax game-tree search, and modern reinforcement learning) power intelligent, adaptive non-player characters and have famously been used to develop programs capable of defeating human champions in games such as chess, Go, and various video games.
• Healthcare: beyond diagnostic expert systems, AI is used for personalized treatment recommendation, drug discovery (predicting molecular properties and candidate compounds), epidemic outbreak prediction, and analyzing large-scale genomic data.
• Speech recognition and synthesis: converting spoken language into text and generating natural-sounding synthesized speech, foundational to voice assistants, transcription services, and accessibility technology for visually or physically impaired users.
• Power system and industrial applications (of particular relevance to electrical engineering): AI techniques are increasingly applied to load forecasting, fault detection and diagnosis in power systems, optimal power flow and economic dispatch optimization, predictive maintenance of industrial machinery, and smart grid management, reflecting the growing convergence of AI methods with traditional electrical engineering domains.

(b) Machine Learning vs Expert Systems

Machine learning is an approach to building intelligent systems in which the system automatically learns patterns, relationships, and decision rules directly from a large body of training data, without being explicitly programmed with predefined rules for every possible situation — the system's internal parameters (weights, decision boundaries, or similar) are adjusted through a training/learning process (using algorithms such as gradient descent, decision tree induction, or similar) to minimize prediction error on the training data, and the resulting trained model is then used to make predictions or decisions on new, previously unseen data. Machine learning systems can, in principle, discover subtle patterns and relationships in data that even a human domain expert may not have explicitly recognized or articulated, and their performance generally improves as more training data becomes available.

Expert systems, by contrast, are built by explicitly encoding a human domain expert's knowledge into a structured knowledge base, typically expressed as a set of if-then production rules (or other knowledge-representation formalism, such as frames or semantic networks) capturing the expert's own reasoning process and domain-specific facts, combined with an inference engine that applies these encoded rules to a specific problem instance to arrive at a conclusion or recommendation, closely mimicking the step-by-step reasoning a human expert would follow. The knowledge in an expert system is typically elicited from human experts by a knowledge engineer through a deliberate, manual knowledge-acquisition process, rather than being automatically learned from data.

Key Differences

• Source of knowledge: machine learning derives its knowledge (in the form of learned model parameters) automatically from training data, while an expert system's knowledge is explicitly hand-crafted and encoded by knowledge engineers working with human domain experts.
• Adaptability: machine learning models can be retrained and improved as new data becomes available, automatically adapting to changing patterns, whereas an expert system's rule base must be manually updated by a knowledge engineer whenever new domain knowledge or exceptions need to be incorporated.
• Explainability: expert systems generally offer more transparent, explicit, human-readable reasoning (since their rules are directly stated in terms familiar to domain experts), whereas many modern machine learning models (particularly deep neural networks) operate as comparative 'black boxes,' making their specific decision rationale considerably harder to directly interpret or explain.
• Data dependency: machine learning fundamentally requires a substantial quantity of representative training data to learn effective, generalizable patterns, whereas an expert system can, in principle, be built and operate effectively even with very little or no historical data, relying instead entirely on the explicitly captured expertise of human specialists.
• Handling novel situations: a well-designed machine learning model can often generalize reasonably well to situations not exactly seen during training (provided they are not too dissimilar from the training distribution), while a purely rule-based expert system may fail entirely (or produce no useful output) when confronted with a situation not explicitly anticipated by its encoded rule set, unless the system incorporates additional uncertainty-handling or default-reasoning mechanisms.

In practice, modern intelligent systems frequently combine both approaches — using machine learning to automatically discover patterns and make predictions from data, while incorporating explicit rule-based expert-system components to encode known domain constraints, safety rules, or regulatory requirements that must be strictly and transparently enforced regardless of what the learned model alone might otherwise predict.

The generalization capability of a trained model, its ability to perform well on new data it has never encountered during training rather than merely memorizing the specific training examples, is the central goal and the central challenge of machine learning system design, and is directly why the held-out test dataset evaluation procedure described above is such an indispensable part of any rigorous machine learning workflow, rather than an optional afterthought.

This distinction between machine learning's data-driven, statistically-derived knowledge and an expert system's explicitly hand-coded, human-expert-derived knowledge is fundamental to correctly choosing between the two approaches for a given AI application.