Introduction to Artificial Intelligence and History of AI

In this presentation, I have presented an introduction to AI, foundation of AI, and History of AI.

The content is a summary of each topic of Chapter-1 of a very famous book on AI, "Artificial Intelligence, A Modern Approach by Stuart Russell and Peter Norvig ".

1. Introduction to AI By Sheetal Jain

2. Before We Begin Studying AI is Valuable but before delving into the Study of AI, defining is important. For years, we human thought how we humans think and why only a few living organisms can perceive, understand, manipulate, and predict.But AI goes beyond it, it doesn’t just understand but also builds intelligent entities. “Artificial Intelligence is a technology that can perform tasks which require human cognition”

3. 1.1 What is AI ? Scientists have approached artificial intelligence in various ways, looking at it from different angles. Some focus on making machines smart like humans, while others aim to create systems that can perform tasks intelligently. These approaches are as follows: Approaches to AI Thought Process and Reasoning Human Behavior Acting Rationally (The Rational Agent Approach) Acting Humanly (Turing Test) Thinking Rationally (The Laws of Thought) Thinking Humanly (Cognitive Modeling)

4. 1.1.1 Thinking H umanly : Cognitive Modeling This approach first requires to understand how human brain works on problems. We can understand how human brain works through introspection, psychological experiments, and brain imaging If we grasp how the human brain works, we can write a computer program on the same principles. Now, we need to ensure that the program input-output are correspond those of human brain. Experimental tools of Psychology Precise Theory of Human Brain AI Modeling

5. 1.1.2 Thinking Rationally : The laws of thought approach Aristotle was the first person to codify “Right Thinking Process” or “Irrefutable Reasoning Process”. He gave idea of Syllogism This study of laws of thought led to the birth to the term “Logic”. Logist tradition of AI hopes to build program to create intelligent system Challenges encountered in this approach are: (i) Notation of the informal knowledge in formal term using logical notation. (ii) Big difference between solving problems in principle and solving problem in practical

6. 1.1.3 Acting Humanly : Turing Test The Turing Test, devised by British mathematician Alan Turing, suggests that if a person cannot reliably tell whether they are communicating with a computer or a human through terminals, then the machine demonstrates human-like behavior. Following are the process under turing tests that a machine need to pass: Natural Language Processing: Understanding of Language Knowledge Representation: Storing information Automatic Reasoning: using stored data to draw conclusion Machine Learning: Adopting new situation and drawing patterns Computer Vision: To perceive object Robotics: To manipulate object physically

7. 1.1.4 Acting Rationally: The Rational Agent Approach The word agent is a latin word which means “To do” and computer agents are expected to do more: Operate Automatically, Perceive Environment, Persist over a prolonged period, adopt changes, create and pursue goals. The word “Rational” means “Act to achieve best or best expected outcome”. Combining above two points, the rational agent focuses on course of action. All the skills required in turing test also allow an agent to act rationally. Knowledge representation and reasoning will enable good decision making and then we require to generate natural general sentences to cope with complex society. Rational Agent approach has two advantages over other approaches to AI: The solution will be more general and this approach is acceptable to scientific development. This is because the standard of acting rationally is well defined mathematically and in more general way.

8. 1.2 The Foundation Of Artificial Intelligence This section provide a brief history of the disciplines that contributed ideas, viewpoints, and techniques to AI : Philosophy Mathematics Economics Neuroscience Psychology Computer engineering Control theory and cybernetics Linguistics

9. 1.2.1 Philosophy In ancient times (around 4th century B.C.), Aristotle created basic rules for thinking and made a system for logical reasoning called syllogisms. Aristotle's system allowed people to draw conclusions in a mechanical way from starting ideas. Ramon Lull later thought that machines could do logical thinking in a mechanical way. In the 17th century, Thomas Hobbes compared thinking to doing math, suggesting a link between mental processes and calculations. Around the year 1500, Leonardo da Vinci designed a working mechanical calculator, showing early progress in automation. Wilhelm Schickard (1623) and Blaise Pascal (1642) made machines that could do calculations, with Pascal saying they act a lot like thought. Gottfried Wilhelm Leibniz (1646–1716) made a machine that could do operations on ideas, going beyond basic math.

10. Descartes (1596–1650) proposed rationalism, dualism, and materialism. The empiricism movement, led by Bacon and Locke, said knowledge comes from experiences with our senses. Logical positivism, created by the Vienna Circle, mixed ideas from thinking and experiences, connecting knowledge to what we observe. Rationalism Dualism Materialism The power of reasoning in understanding world There is a part of human brain which is out of nature Brain operation, according to the physics law, constitute brain Hence, the philosophical picture of the mind is constituted by connection between knowledge and actions

11. 1.2.2 Mathematics Philosopher gave fundamental idea of AI but formal science required mathematical formalization. There are three areas of focus under this discipline: Logic George Boole (1815–1864) initiated the development of propositional, or Boolean, logic, which serves as the basis for logical reasoning. In 1879, Gottlob Frege (1848–1925) expanded Boole's work, introducing first-order logic that is widely used today. Alfred Tarski (1902–1983) contributed by presenting a theory of reference, facilitating the connection between logical entities and real-world objects. Computation In AI, computation is fundamental, relying on logical operations and algorithms. By manipulating symbols and data, computers mimic intelligent behavior, facilitating problem-solving, learning, and decision-making across different fields. The essence of AI computation lies in mathematical models, algorithms, and data processing, replicating cognitive functions and advancing the creation of intelligent systems and machines. Probability Probability, the third big math idea in AI, started with Gerolamo Cardano and was developed by Blaise Pascal and others. It began with gambling but became crucial for dealing with uncertainty in sciences. People like James Bernoulli, Pierre Laplace, and Thomas Bayes improved the theory and introduced ways to use statistics. Thomas Bayes suggested a rule for updating probabilities with new evidence, which is a key part of how AI systems handle uncertainty today.

12. 1.2.3 Economics Adam Smith gave birth to economics by launching his book An Inquiry into the nature and the cause of wealth of nations . Smith was the first person to treat economic as science. Economists always thought that economy is always about money, but it was about how people make choices that lead to "preferred outcome". The mathematical treatment or Utility of "Preferred outcome" was formalised by Leon Walras, and was further improved by Frank Ramsey, John von Neumann and Oskar Morgenstern in the book, "The Theory of Games and Economic Behavior".

13. Decision Theory combines probability and Utility Theory, applicable to large economies where individual decisions have no bearing on others. In small organizations, individual decisions significantly impact others, leading to the development of Game Theory by Von Neumann and Morgenstern. Unlike Decision Theory, Game Theory doesn't prescribe clear actions. Probability Utility Theory Decision Theory

14. Economists haven't addressed the question of making rational decisions when payoffs result from a sequence of actions. This was pursued in the field of operations research, which emerged during WWII. Richard Bellman introduced a sequential process, Markov Decision Process, to formalize a class of sequential problems in the field of operations research. Pursued in

15. 1.2.4 Neuroscience Neuroscience is the study of brain. Though, exact working of human brain remained mysterious but later it was called a seat of consciousness Paul Broca’s study of speech deficit in a damaged-brain patient showed the existence of localized areas in brain responsible for different function. The left hemisphere of the brain is responsible for speech production The brain is made up of small nerves called neurons. Camillo Golgi developed a technique to study individual neurons, which Santiago used to understand the brain's neuron structure. Nicolas was the first to apply a mathematical model to comprehend how the brain's neurons work. Brain cannot use all of its neurons simultaneously but a computer can. But brain has advantage of storing unlimited information A simple collection of cells lead to thought, action, and consciousness or we can say brain causes mind.

16. The most information processing goes in cerebral cortex, the outer layer of the brain. When several neurons sends signal to each other and communicate at a junction, this happens because of electrochemical reaction. The signals control brain activity and this mechanism is thought to be the process of learning in the brain.

17. 1.2.5 Psychology Scientific psychology traces its origins to Hermann von Helmholtz (1821–1894) and his student Wilhelm Wundt (1832–1920). Helmholtz applied the scientific method to human vision, creating a fundamental treatise on vision. In 1879, Wundt established the first laboratory of experimental psychology at the University of Leipzig, emphasizing controlled experiments and introspection. Behaviorism, led by John Watson (1878–1958), rejected mental processes, focusing on objective measures of stimulus and response. Cognitive psychology, viewing the brain as an information-processing device, can be traced back to William James (1842–1910).

18. Helmholtz believed perception involved unconscious logical inference, a viewpoint later revived in cognitive psychology. Frederic Bartlett's Applied Psychology Unit at Cambridge fostered cognitive modeling, challenging behaviorism. Kenneth Craik (1943) outlined three key steps for a knowledge-based agent: translating stimulus, cognitive manipulation, and retranslation into action. Donald Broadbent continued Craik's work, modeling psychological phenomena as information processing. Cognitive science emerged in the U.S., influenced by computer modeling and key presentations in a 1956 MIT workshop by Miller, Chomsky, and Newell-Simon. ______________

19. 1.2.6 Computer Engineering Essentials for AI Success: AI success relies on combining intelligence with a computing artifact, with the computer being the primary tool. World War II Contributions: The first operational computers emerged during World War II, including Heath Robinson and Colossus by Alan Turing's team, Z-3 by Konrad Zuse, and ENIAC by John Mauchly and John Eckert.

20. Evolution of Computer Performance: Computer performance has evolved, emphasizing parallelism since 2005, after which the focus shifted from increasing clock speed to multiplying CPU cores. Calculating Devices Before Computers: Automated machines from the 17th century preceded electronic computers, with Joseph Marie Jacquard's programmable loom in 1805 and Charles Babbage's ambitious Analytical Engine in the mid-19th century.

21. Babbage's Unfinished Machines: In 1991, Charles Babbage designed the Difference Engine for mathematical computations. He also designed the Analytical Engine, the first artifact capable of universal computation. Ada Lovelace's Contribution: Ada Lovelace, Babbage's colleague, is considered the world's first programmer. She had written programs for the unfinished Analytical Engine. Debt to Computer Science: AI owes a debt to computer science for operating systems, programming languages, and tools. However, AI has also contributed significantly to mainstream computer science with ideas like time sharing, interactive interpreters, and more.

22. AI Pioneering Ideas: AI has pioneered concepts adopted in mainstream computer science, including personal computers with windows and mice, rapid development environments, linked list data type, automatic storage management, and key concepts of symbolic, functional, declarative, and object-oriented programming. Software Side Contribution: The software side of computer science has played a vital role in providing tools and languages for writing modern programs and papers about them. Reciprocal Impact: The relationship between AI and computer science is reciprocal, with both fields influencing and benefiting from each other's advancements.

23. 1.2.7 Control theory and Cybernetics Control theory and cybernetics contributed to AI by providing frameworks for understanding and regulating the behavior of systems. Cybernetics is the interdisciplinary study of the structure, function, and dynamics of systems, particularly those that involve communication and control. It explores the principles of feedback, information, and regulation in various types of systems, including biological, mechanical, and social systems.Cybernetic principles, such as goal-oriented feedback, played a crucial role in the development of learning algorithms in AI.

24. Control theory offered insights into system stability, helping AI engineers design robust and reliable autonomous systems. The integration of cybernetic ideas into AI allowed for the creation of self-regulating systems capable of adapting to changing environments. The study of control mechanisms in biological systems inspired the design of adaptive algorithms in AI, mirroring natural learning processes. Cybernetics influenced the development of intelligent agents, enabling them to perceive, reason, and act in a manner analogous to how living organisms interact with their environment. Control theory and cybernetics continue to shape AI research, providing theoretical foundations and practical tools for designing efficient and responsive artificial systems. __________________

25. 1.2.8 Linguistics B.F. Skinner's "Verbal Behavior" (1957) presented behaviorism in language learning. Noam Chomsky's critique questioned behaviorism's inability to explain language creativity. Chomsky's own theory, based on syntactic models, offered a programming potential. Modern linguistics and AI emerged simultaneously, forming computational linguistics. Language understanding complexity extends beyond sentence structure to context and subject matter. Early knowledge representation work in AI was closely linked to language and linguistics. The intersection of philosophy and language influenced linguistic research and AI development. _________________

26. 1.3

27. 1943-55 The gestation of Artificial Intelligence 1956 The birth of artificial intelligence 1952-69 Early enthusiasm, great expectations 1966-73 A dose of reality 1969-79 Knowledge Based System 1980- Present 1986–present 1987 – present 1995– present 2001– present AI Becomes an Industry The return of neural network AI Adopts Scientific Method T he emergence of intelligent agent The availability of large data sets

28. 1.3.1 The Gestation of AI In 1943, Warren McCulloch and Walter Pitts laid the foundation for artificial intelligence (AI) by creating a model of artificial neurons inspired by brain physiology, propositional logic, and Turing's theory of computation. They demonstrated that networks of these neurons could compute any function and implement logical operations. Donald Hebb (1949) introduced Hebbian learning to modify connection strengths between neurons, a concept still influential today.

29. Alan Turing's 1950 article introduced key AI concepts, including the Turing Test, machine learning, genetic algorithms, and reinforcement learning. Turing also proposed the Child Programme idea, simulating a child's mind instead of an adult's. In 1950, Harvard students Marvin Minsky and Dean Edmonds built the first neural network computer, SNARC. Minsky later explored universal computation in neural networks at Princeton.

30. 1.3.2 The Birth of AI For the next 20 years, AI was shaped by these people and their connections at MIT, CMU, Stanford, and IBM. The Dartmouth proposal highlighted that AI focuses on imitating human abilities, using computer science as its method. AI became its own field because it had unique goals and methods, unlike control theory, operations research, or decision theory. In 1951, John McCarthy, an important person in AI, finished his PhD at Princeton. Later, in 1956, he organized a workshop at Dartmouth, which is considered the starting point of AI. The goal was to figure out how to make machines simulate human intelligence. Attendees included famous researchers like Allen Newell and Herbert Simon. The workshop didn't bring big breakthroughs, but it united key people.

31. 1.3.3 Early Enthusiasm, great expectations In the early days of AI, with basic computers, pioneers like John McCarthy and others amazed people by making computers do clever things. Allen Newell and Herbert Simon made the General Problem Solver, a program that solved problems like humans. It sparked the idea that intelligence involves manipulating symbols. Outline of General Problem Solver

32. At IBM, Herbert Gelernter and Arthur Samuel created AI programs. in 1958, McCarthy made Lisp, a key programming language for AI. McCarthy later started the AI lab at Stanford to emphasize logic. They explored "microworlds" like the blocks world to solve limited but smart tasks. Early work on neural networks, inspired by McCulloch and Pitts, also advanced. All these achievements set the stage for the future of AI. _________

33. 1.3.4 A dose of reality In 1 957, Herbert Simon said machines would think and learn fast. But early AI had problems. Translating languages failed because computers lacked knowledge. Thinking faster with better hardware didn't work for complex AI challenges. In 1973, the Lighthill report criticized AI, reducing support. I n 1969, Minsky showed that b asic structures for smart behavior had limit. New learning methods came later, but early AI struggled with big expectations and real-world difficulties. "AI Winter" symbolizes a period marked by reduced enthusiasm and backing for advancements in artificial intelligence.

34. 1.3.5 Knowledge Based System In early day of AI, AI researchers used weak methods or general searches for solutions. DENDRAL ,an expert system, broke ground using specific knowledge for molecular structure. It replaced exhaustive searches with chemists' pattern recognition, making it more efficient. DENDRAL was knowledge-intensive and used specialized rules. MYCIN, another expert system, was a backward chaining expert system that used AI to identify microorganisms causing severe diseases like bacteremia and meningitis and propose antibiotics based on patient weight . Since then, domain knowledge became crucial in natural language understanding. While early systems like SHRDLU had limitations, Roger Schank's work at Yale emphasized knowledge representation and reasoning for language understanding. Real-world applications led to different languages, from logic-based Prolog to Minsky's frame-based approach.

35. 1.3.6 AI Becomes An Industry In the early 1980s, the first successful commercial expert system, R1, operated at Digital Equipment Corporation, saving millions of dollars. By 1988, major corporations like DEC and DuPont had deployed numerous expert systems, resulting in significant cost savings. The AI industry grew rapidly, reaching billions of dollars with companies developing expert systems, vision systems, robots, and specialized software and hardware. However, the period known as the "AI Winter" followed, marked by companies failing to fulfill grand promises, leading to a downturn in the AI industry.

36. 1.3.7 The Return of Neural Network In the 1980s, researchers rediscovered a learning algorithm called back-propagation, first found in 1969. They applied it to solve learning problems in computer science and psychology. Some thought that connectionist models, which emphasize neural networks, could challenge symbolic and logic-based approaches in AI. There was a debate about whether manipulating symbols played a crucial role in human thinking. Nowadays, we see both connectionist and symbolic approaches as working together, not competing. Current neural network research has two branches: one focuses on designing effective systems, and the other studies the properties of real neurons.

37. 1.3.8 AI Adopts Scientific Method In recent years, there has been a significant shift in artificial intelligence (AI) towards building on existing theories, rigorous experimentation, and real-world applications. AI, once isolated, is now integrating with fields like control theory and statistics. The scientific method is firmly applied and, now, AI requires hypotheses to undergo empirical experiments and statistical analysis. R ecent dominance by hidden Markov models (HMMs) is due to their rigorous theory and training on real speech data. S imilar trends are seen in machine translation and neural networks, which now benefit from improved methodology and theoretical frameworks.

38. Judea Pearl's work in probabilistic reasoning led to a new acceptance of probability and decision theory, with Bayesian networks dominating uncertain reasoning in AI. Normative expert systems, acting rationally based on decision theory, have become prominent. Similar revolutions have occurred in robotics, computer vision, and knowledge representation, as increased formalization and integration with machine learning prove effective in solving complex problems. ______________

39. 1.3.9 The emergence of Intelligent Agent Researchers are looking again at the "whole agent" challenge in AI, like the SOAR architecture. The Internet is a big deal for smart agents, used in things like search engines. Creating complete agents shows the need to shake up AI fields and handle uncertainties in sensory systems. AI now works closely with areas like control theory and economics, especially in things like controlling robotic cars.

40. Despite successes, some AI leaders like McCarthy, Minsky, Nilsson, and Winston weren't happy. They wanted AI to go back to its original goal of making human-like AI (HLAI), focusing on machines that think, learn, and create. Another idea was Artificial General Intelligence (AGI), aiming for a universal way of learning and acting in any situation rightly and making sure AI is friendly and not a worry in this journey. __________________

41. 1.3.10 The availability of large data sets. In the past 60 years of computer science, people mostly focused on creating algorithms. But now, in AI, we're realizing that for many problems, it's more useful to focus on the data instead of getting too caught up in which algorithm to use. This change is because we have a lot of data available, like trillions of English words or billions of web images.

42. An important study by Yarowsky showed that, for tasks like figuring out the meaning of a word in a sentence, you can do it really well without human-labeled examples. Another study by Banko and Brill found that having more data is often more helpful than choosing a specific algorithm. For instance, Hays and Efros improved a photo-filling tool by using a bigger collection of photos. This shift in thinking suggests that in AI, where we need a lot of knowledge, we might rely more on learning from data instead of manually coding everything. With the rise of new AI applications, some say we're moving from "AI Winter" to a new era, “AI Summer”, as AI becomes a fundamental part of many industries, as noted by Kurzweil.

43. 1.4 The State of The Art AI today does various tasks: Robotic Vehicles: Driverless cars like STANLEY navigate terrains using cameras and sensors. Speech Recognition: Systems guide conversations, like booking flights with an automated phone system. Autonomous Planning: NASA's Remote Agent autonomously plans spacecraft operations. Game Playing: IBM's DEEP BLUE beat the world chess champion, Garry Kasparov.

44. 5. Spam Fighting: Learning algorithms classify over a billion messages daily to identify and filter spam. 6. Logistics Planning: During the Persian Gulf crisis, AI tools like DART automated complex logistics planning for the U.S. forces. 7. Robotics: iRobot's Roomba vacuum and PackBot handle various tasks, from cleaning homes to hazardous materials disposal. 8. Machine Translation: Programs translate languages, like Arabic to English, using statistical models trained on vast text examples. These are real applications of AI, showing its practical use in today's world.

45. Thank You