Comprehensive Study Notes – AI Overview • Knowledge Representation • Programming Languages

Introduction & Scope of the Notes

• Covers material from p. 3–36 of the transcript (Chs. 1–3)
• Focus areas:
  • Overview, history and importance of Artificial Intelligence (AI) (Chapter 1)
  • Knowledge Representation & Reasoning (KRR) (Chapter 2)
  • Programming languages commonly used for AI development (Chapter 3)
• Embedded throughout: examples, timelines, interdisciplinary links, numerical data, ethical remarks, LaTeX-formatted logic where it appears in the source.

Core Ideas & Definitions of AI

• AI = “science & engineering of making intelligent machines” (John McCarthy, 1956; reiterated 2019).
• Two component words:
  • “Artificial” → human-created
  • “Intelligence” → capacity to learn, reason, adapt, perceive, decide, use language, etc.
• Working definition (Iyer 2018): ability of a system to calculate, reason, perceive relationships, learn from experience, solve problems, generalize, adapt.
• Data is the key raw material; Data Science = study of storing, recording & analysing data for societal benefit.
• AI ≈ simulation/replication of human cognitive processes with computer systems that think & act rationally.

Historical Milestones (condensed from Table 1.2)

• $1836$ Babbage & Ada Lovelace conceive programmable machine.
• $1923$ Word “robot” coined in Čapek’s play.
• $1943$ McCulloch & Pitts propose artificial neurons.
• $1950$ Alan Turing publishes Computing Machinery & Intelligence; Turing Test introduced.
• $1955$ Newell & Simon’s “Logic Theorist” (first AI program).
• $1956$ Dartmouth Conference; term “Artificial Intelligence” launched (McCarthy).
• $1966$ ELIZA chatbot.
• $1969$ Shakey robot (Stanford Research Institute).
• $1980$ Expert Systems boom (Edward Feigenbaum).
• $1997$ IBM Deep Blue defeats Garry Kasparov.
• $2011$ IBM Watson wins Jeopardy!
• $2018$ Google Duplex & IBM Project Debater demonstrate conversational AI.

Importance & Benefits of AI

• Automates high-volume, repetitive tasks with consistency (hardware-driven automation + learning layer).
• Progressive learning algorithms refine themselves via back-propagation & big data.
• Extracts hidden patterns through deep neural networks (DNNs) → higher accuracy (diagnose MRI scans, recommend products, etc.).
• AI augments rather than replaces humans; provides human–AI partnership benefits:
  • Enhanced perception/analytics (computer vision, time-series, NLP).
  • Bridges economic, language, translation barriers.
  • Builds predictive models, interactive interfaces, image/video understanding.

Cognitive Processes Involved in AI Systems

1. Reasoning – choosing best algorithm, inductive/deductive.
2. Learning – acquiring data, forming algorithms (rules) → actionable info.
3. Problem-Solving – selecting optimal alternative to reach goal.
4. Perception – sensing & interpreting environment (sensors / data fusion).
5. Self-Correction – continual refinement for accuracy.

AI as an Interdisciplinary Tool (Fig. 1.1 & 1.2)

• Draws from computer science, mathematics, psychology, biology, sociology, philosophy, neuroscience.
• Technology stack: Machine Learning (ML), Deep Learning (DL), Neural Networks (ANN), Natural Language Processing (NLP), Fuzzy Logic, Robotics, Expert Systems.
• Three classic ML paradigms: supervised, unsupervised, reinforcement learning.

Taxonomy: Types of AI (7-level schema)

1. Reactive Machines – no memory (IBM Deep Blue).
2. Limited Memory – short-term data retention (self-driving cars, chatbots).
3. Theory of Mind – social intelligence, emotion modelling (research stage).
4. Self-Aware – consciousness & self-preservation (conceptual; potential risks).
5. Artificial Narrow Intelligence (ANI / Weak AI) – performs one specific task.
6. Artificial General Intelligence (AGI / Strong AI) – human-level versatility (e.g., AlphaGo, Pillo Robot).
7. Artificial Super Intelligence (ASI) – surpasses human cognition (Alpha 2 humanoid prototype).

Advantages vs. Disadvantages of AI

• Pros: accuracy, speed, unbiased decision-making, reliability, 24 × 7 operation, multitasking, hazardous-task handling, resource optimisation, digital assistants.
• Cons: high cost, limited creativity, context-rigidity, absence of emotions, user dependence, potential misuse.

Contemporary Examples

• Alexa / Echo, Siri, Google Now – voice assistants.
• Flipkart / Amazon – recommender engines.
• Netflix OTT – personalised content.
• Roomba – domestic robot.

Applications Across Domains (Sec. 1.10)

• AI-as-a-Service (AIaaS) – IBM Watson, Amazon AI.
• Autonomous Vehicles – Tesla, lane keeping via computer vision.
• Agriculture – crop monitoring, predictive analytics.
• Banking & Finance – chatbots, credit scoring, algorithmic trading.
• Business & E-commerce – demand forecasting, customer support.
• Cybersecurity – malware detection (AEG, AI2).
• Education – adaptive tutoring, AI grading.
• Entertainment & Media – content recommendation.
• Government – policy analytics, traffic management.
• Health Care – diagnosis, surgical robots (Gaumard simulators), pandemic prediction (BlueDot).
• Law – document analysis, outcome prediction.
• Personal Assistants – global market projected $\text{USD }25\text{ billion by }2025$.
• Robotics & Cobots – industrial assembly, Kiva warehouse robots.
• Retail – inventory robots (Bossa Nova @ Walmart).
• Transportation – traffic flow optimisation, claim processing.
• Vision / Speech / Handwriting Systems – face recognition, speech-to-text, pen-input OCR.

Knowledge Representation & Reasoning (KRR) Basics (Ch. 2)

• KRR = methods for describing real-world facts so machines can understand, learn & reason.
• Central to tasks like theorem proving, gaming, medical imaging, NLP.
• AI pipeline (Perception → Learning → KRR → Planning & Execution).

Types of Knowledge (Fig. 2.1)

• Declarative – facts & objects (“what”).
• Procedural / Imperative – rules & strategies (“how”).
• Heuristic – rule-of-thumb expertise.
• Structural – relationships between entities (part-of, instance-of).

Intelligence–Knowledge Link

• Intelligent behaviour rests on prior knowledge; absence of knowledge ⇒ poor decision-making.
• Explosion of unstructured data ⇒ need for big-data analytics & cognitive computing to extract knowledge.

Knowledge Life-Cycle in AI (Fig. 2.3)

Perception → Learning → KRR → Planning → Execution (feedback loop).

Representation Approaches

1. Relational / Tabular – simple tables.
2. Inheritable / Frame-based – hierarchical with slot-value pairs (Fig. 2.5).
3. Inferential / Logical – well-formed formulas (wff); e.g.,
   $\text{Lady}(\text{Daph}) \land \forall x\,[\text{Lady}(x) \Rightarrow \text{Mortal}(x)] \;\Rightarrow\; \text{Mortal}(\text{Daph})$
4. Procedural – code fragments & IF–THEN rules.

Requirements for a KR System

• Accuracy, Semantic clarity, Expressiveness, Scalability, Naturalness, Robustness, Portability, Cost-effectiveness, GUI tools, Foreign-system interfaces, Knowledge-entry support.

Major KR Techniques

• Logical Representation – propositional & predicate logic; precise syntax/semantics.
• Semantic Networks – graph of nodes & arcs (IS-A, kind-of relations).
• Frames – slot–facet structures; aka slot-filter.
• Production Rules – IF THEN ; recognise–act cycle, conflict resolution.

Real-Time Challenges

• Identifying key attributes & relationships.
• Choosing proper granularity.
• Representing sets of objects unambiguously.
• Structuring huge volumes for efficient retrieval.

Programming Languages for AI (Ch. 3)

Overview

• Selection criteria: execution speed, library ecosystem, ease of learning, portability, memory manage-ment, community support.

Java

• Strengths: portability via JVM, robust OOP, automatic memory management, libraries (TensorFlow Java, Deeplearning4j, OpenNLP).
• Weakness: slower than C++, higher latency.

C++

• Strengths: fastest execution, fine-grained memory control, useful for search engines, real-time systems, game AI; supports object-oriented paradigms.
• Weakness: weak multitasking, no built-in garbage collection, bottom-up complexity.

Python

• Strengths: simple syntax, huge ecosystem (PyBrain, Theano, MXNet, PyTorch, TensorFlow, Scikit-Learn), cross-platform, integrates with C/Java, ideal for ML & DL.
• Weakness: interpreter-driven ⇒ slower; mediocre for mobile apps; can create language-switching barrier.

LISP (LISt Processing)

• Invented by McCarthy; dynamic, macro-rich, rapid prototyping, automatic garbage collection.
• Inspired later languages (R, Julia).
• Drawback: sparse modern libraries, unconventional syntax, needs heavy configuration.

Prolog

• Logic-programming paradigm; relies on pattern matching & backtracking; perfect for rule-based expert systems (ELIZA chatbot).
• Supports both symbolic & statistical AI.
• Drawback: steeper learning curve, limited standardization across platforms.

R

• Created for statistics & data analysis; strong in numeric computation, vector ops, visualisation, AI packages (gmodels, tm, RODBC, OneR).
• Interfaces with C/C++/Fortran.
• Drawbacks: high memory use, weaker security for Web, slower execution, tougher for programming newcomers.

Ethical, Philosophical & Practical Implications

• Job displacement vs. new skill creation – need for reskilling.
• Bias & fairness – training data must be representative.
• Autonomy & accountability – who is liable for AI decisions?
• Privacy & surveillance – face recognition, data mining require regulation.
• Super-intelligence risk – self-aware AI may pursue self-preservation.

Quick Reference Timeline & Key Figures

• Charles Babbage & Ada Lovelace – mechanical computing $1836$.
• Alan Turing – Turing Test $1950$.
• John McCarthy – father of AI, coined term, created LISP.
• Newell & Simon – Logic Theorist $1955$.
• Feigenbaum – Expert Systems $1980$.
• Geoffrey Hinton et al. – modern deep learning boom.

Suggested Further Reading

• Bellman (1978) An Introduction to Artificial Intelligence.
• Nilsson (1998) Principles of Artificial Intelligence.
• Russell & Norvig (latest ed.) Artificial Intelligence: A Modern Approach.
• Sharma & Garg (2020) Artificial Intelligence: Technologies, Applications, and Challenges.

End-of-Chapter Takeaways

• AI integrates data, algorithms, and computational power to emulate human cognition.
• Effective Knowledge Representation is foundational for intelligent behaviour.
• Choice of programming language must align with application constraints (speed, libraries, portability).
• Ethical foresight is essential to harness AI’s power responsibly.