Artificial Intelligence_An Overview-Slide 2 (1).pdf

COSC 423 Artificial Intelligence and Applications Lecture 2: AI: An Overview Artificial Intelligence: An Overview What is Intelligence? A Brief History of AI Intelligent Agent? Alternative Perspectives on What is A...

COSC 423 Artificial Intelligence and Applications Lecture 2: AI: An Overview Artificial Intelligence: An Overview What is Intelligence? A Brief History of AI Intelligent Agent? Alternative Perspectives on What is Artificial Intelligence? Knowledge Representation Intelligence Computation, Computers, AI and Learning State of AI Research and Programs What is Intelligence? ▪ Defining intelligence is difficult, as no single scientific definition is universally accepted. Researchers in AI and cognitive science view intelligence as a collection of attributes like perception, reasoning, and learning. Current AI systems only exhibit narrow forms of intelligence in specific domains such as medical diagnosis or information retrieval. We will explore the following attributes; ▪ Perception: The process of manipulating, integrating, and interpreting sensor data to achieve purposeful, goal-directed understanding within a system. ▪ Action: Coordinating and controlling effectors to perform tasks, explore, and manipulate the environment, including designing and constructing tools. ▪ Reasoning: Using deductive, inductive, and analogical inferences to make decisions, justify conclusions, and adapt explanations in uncertain or changing situations. ▪ Adaptation and Learning: The ability to adjust behavior to environmental changes, discover patterns, build internal representations, and learn to differentiate, generalize, and apply knowledge across various domains. ▪ Communication: The exchange of information with other intelligent agents using symbols, language, or other media to convey goals, beliefs, and narratives. What is Intelligence? Cont’d ▪ Planning and Goal-Directed Problem-Solving: ▪ Formulating and evaluating sequences of actions to achieve goals, adapting to changes, and simplifying complex tasks. ▪ Autonomy: Setting and achieving goals independently, adjusting actions as necessary within environmental or physical constraints. ▪ Creativity: Exploring, modifying, and extending domains like language or mathematics by manipulating constraints or other methods. ▪ Reflection and Awareness: Being conscious of one's own internal processes, such as reasoning and goals, as well as those of others. ▪ Aesthetics: The articulation and application of aesthetic principles in various contexts. ▪ Organization: Forming social groups based on shared objectives, developing conventions to facilitate interaction and cooperation. The hallmark of intelligence is not just displaying certain attributes, but doing so across a wide range of unpredictable, context-dependent domains with varying constraints. Different systems, whether natural or artificial, may exhibit varying subsets of these attributes to different extents. What is Artificial Intelligence? ▪ AI aims to understand intelligence by creating models of both natural and artificial minds. Philosophers have long debated the nature of intelligence, but the digital computer in the 1950s shifted the focus to computer scientists. ▪ The development of computation theory provided tools to analyze, design, and evaluate systems that exhibit intelligent behavior. The Origins of AI ▪ The term Artificial Intelligence was coined by John McCarthy in 1956 at the Dartmouth Workshop. ▪ McCarthy contributed to AI by designing the LISP programming language and the first time-sharing operating system. Descriptions of AI ▪ AI is a science that formulates testable hypotheses about the structures and processes required for intelligent behavior. It studies computational models of perception, cognition, and action. ▪ AI explores possible and actual intelligent systems. It involves designing and analyzing intelligent agents. What is Artificial Intelligence? Cont’d AI as Engineering ▪ AI focuses on automating tasks that require human intelligence, such as theorem proving, facial recognition, disease diagnosis, and music composition. ▪ Like other engineering fields, AI relies on a strong scientific foundation. AI's Guiding Hypothesis ▪ The central hypothesis of AI is that cognition, or thought processes, can be modeled computationally. ▪ This idea traces back to thinkers like Helmholtz, Leibnitz, and Boole, who viewed thought as mechanical or computational. Functional View of Intelligence ▪ In AI, intelligence is seen as a functional capability, independent of the physical substrate supporting it. Examples include the physical symbol system hypothesis, the language of thought, and neural networks. AI’s Working Hypothesis ▪ AI’s hypothesis is useful for formulating models and testing ideas, but it is subject to revision based on experimental evidence. Computation, Computers, and Programs ▪ The formal concept of computation was developed in the 1930s by Alan Turing, influenced by Hilbert’s decision problem. ▪ Turing introduced the idea of an algorithm: a set of rules to solve specific problems, which laid the foundation for modern computer science. ▪ Turing invented the Turing machine, a hypothetical computer that manipulates symbols on an infinite tape, demonstrating how algorithms could be processed. Turing’s Contributions and Universal Computation ▪ Turing showed that a universal Turing machine could compute anything any other machine could, provided sufficient memory and program instructions. ▪ He also proved that any describable processes, whether parallel or serial, could be executed by this universal machine or equivalent systems like lambda-calculus or neural networks. Limits and Evolution of Computation ▪ Some problems having no guaranteed solution, such as halting in certain logical systems. Recent developments in quantum computing and biological substrates like DNA and RNA open new frontiers in computation, beyond the traditional Turing model. A Brief History of AI Greek Mythology: Prometheus and the quest for intelligence. Vedic Texts: Ancient theories about existence and knowledge. Early Intellectual Contributions ▪ Aristotle (384–322 BC): Foundations of epistemology and logic; distinction between matter and form. ▪ Panini (350 BC): Formal grammar for Sanskrit; basis for syntactic models. ▪ Al Khowarizmi (825): Introduction of algorithms to the West; development of Arabic algebra. ▪ Philosophical and Mathematical Foundations ▪ Descartes (1556–1650): Cogito ergo sum; ideas about thought and reality. ▪ Leibnitz (1646-1716): General method for reducing truths to calculations. ▪ Boole (1815-1864): Study of logic and probability; laws of thought. The Rise of Modern AI and Key Developments ▪ Formalizing Logic and Computation: Russell, Frege, Tarski (1910-1950): Formal logic; theory of reference. Turing (1912-1954): Invention of the Turing Machine A Brief History of AI Cont’d ▪ Church, Kleene, Post (1930-50): Church-Turing thesis; Turing-equivalent models of computation. Early AI and Computational Advances ▪ McCarthy et al. (1956): Establishment of AI at Dartmouth College. ▪ Shannon (1948): Development of information theory. ▪ Chomsky (1956): Chomsky hierarchy of languages; formalization of computation. Key Innovations and Paradigm Shifts ▪ Internet (1974): Invention by Cerf. Neural Networks (1950s-1990s): Early models; renewed interest in biological inspiration. World Wide Web (1989-91): Invention by Berners-Lee. Recent Advances and Current Trends ▪ Applications: Adaptive information retrieval, data mining, smart devices, autonomous vehicles, healthcare systems. ▪ Nouvelle AI: Emergent behavior from interactions among autonomous entities; integration of symbolic and non-symbolic methods. ▪ Modern Research Focus: (Synthesis of Approaches) Combining traditional logic-based systems with soft computing technologies. ▪ Active Research Topics: Agent design, inter-agent communication, multi-agent Relation of AI to other disciplines AI and Computer Science ▪ Foundation: Digital and analog computers from the 1940s and 1950s; theory of computation, information theory, control. ▪ Contributions: LISP (early high-level programming language), multi-tasking operating systems, logic programming, constraint programming, heuristic search, neural networks. ▪ Impact: Stimulated research in parallel architectures, complexity of reasoning, and learning. AI and Psychology ▪ Focus: Psychology studies behavior in humans and animals; AI creates computational models of intelligent behavior. ▪ Influence: AI models have impacted contemporary psychology and neuroscience, providing new insights into cognitive processes. AI and Artificial Life ▪ Shared Concerns: Both explore intelligent behavior and adaptive systems, though AI is broader and more focused on computational models. Relation of AI to other disciplines Cont’d AI's Broader Implications and Applications ▪ AI as Applied Epistemology ▪ Focus: AI explores the nature of knowledge and introduces new questions in epistemology through computational models. AI in Engineering ▪ Practical Tools: Programs for system configuration, fault diagnosis, information retrieval. AI's Interdisciplinary Approach ▪ Integration: Draws from philosophy, psychology, linguistics, anthropology, engineering, neuroscience. ▪ Unique Contribution: Offers a computational perspective on intelligent behavior, contributing to Cognitive Science and raising new questions in various domains. AI in Art and Science ▪ Artistic Synthesis: Involves representing knowledge and theories of creativity. ▪ Scientific Exploration: Models scientific processes and methods, yielding insights into hypothesis formation, experimental design, and theory selection. ▪ 4o mini Goals of AI+Practical AI ▪ Understanding Cognitive Processes: AI research aims to enhance our understanding of perception, reasoning, learning, and creativity. ▪ AI as a Tool: AI extends human intellectual and creative abilities, comparable to how the internal combustion engine expanded physical mobility. ▪ Broader Impacts: Understanding AI’s mechanisms can help address social, environmental, and cultural issues. Practical AI ▪ Intelligent Agents: AI involves designing systems that perform tasks requiring intelligence, like problem-solving and planning. ▪ AI Tools: Includes search algorithms, knowledge representation, adaptation, and learning. ▪ Applications: AI has produced useful tools (e.g., game players, vision programs, tutoring systems, stock-market analysts). State Space Search Problem-Solving as State Space Search ▪ Search Paradigm: Problem-solving in AI often boils down to searching a state space to find a solution. ▪ State-Space Definition: Each state represents a step in solving a problem, and operators transform one state into another. ▪ Heuristics & Efficiency: Efficient search strategies are crucial as exhaustive search is computationally expensive. Computational Challenges in Search ▪ Intractability: Search problems grow exponentially with problem size, making them hard to solve efficiently. ▪ Heuristic Search: AI often uses heuristics (rules of thumb) to narrow down the search space and find feasible solutions. ▪ AI’s Focus: Designing algorithms to efficiently search vast state spaces and find near-optimal solutions. Knowledge Representation ▪ Core Role: AI systems must represent knowledge about their environment to function effectively. ▪ Representation Techniques: AI uses various formal languages (e.g., logic, production rules, lambda calculus) to represent knowledge. ▪ Representation Flexibility: Knowledge can be expressed in multiple forms depending on the task or system’s needs. Nature of Knowledge Representation ▪ Logical Systems: Logic allows AI systems to represent and manipulate knowledge through inference. ▪ Expressive Power: First-order logic is powerful but may be inefficient for certain kinds of knowledge (e.g., spatial relations). ▪ Trade-offs: Practical systems often limit the expressive power of representations to balance efficiency and precision. AI and Learning ▪ Role of Learning: AI systems must learn from the environment to improve their performance over time. ▪ Representation & Learning: Learning processes help AI systems develop better internal representations, solving tasks more efficiently. ▪ Challenges: Initial representations must be carefully chosen, as poor representations can hinder learning. Evolution and Learning ▪ Task-Specific Representations: Evolutionary processes in biological systems provide a model for AI to develop task-specific representations. ▪ Flexible AI Systems: Some AI systems can compose functions and representations on demand for specific tasks. ▪ Learning's Importance: AI systems improve through learning by storing results of past inferences for future use. AI and Learning Cont’d Semantic Grounding of Representations ▪ Grounding Problem: Representations need to be connected to the physical world through sensory inputs and motor actions to be meaningful. ▪ Embodied Intelligence: Autonomous robots provide a useful testbed for studying AI, as they require grounded representations to operate in the physical world. Intelligent Agent? How Do We Know if an Agent is Intelligent? ▪ Turing Test for Intelligence: ▪ A system is intelligent if its performance is indistinguishable from a human in the same task. ▪ The interrogator asks questions to both a human and an AI, unable to tell which is which. ▪ If the AI wins as often as the human, it passes the Turing Test. Limitations of the Turing Test (Challenges to the Turing Test): ▪ Is it necessary for AI to mimic human weaknesses (e.g., slow calculations)? ▪ Does intelligence require a human-centric evaluation? ▪ The test may need to include more complex tasks beyond verbal answers. Alternative Perspectives on Intelligence Intelligence in Broader Contexts: ▪ Intelligence can also be seen in other species (e.g., chimps, dogs, parrots). ▪ Should we measure intelligence based on survival in changing environments? ▪ Perhaps the question of intelligence itself is too complex and subjective. Tasks vs. Intelligence ▪ Alternative Measures of Intelligence: ▪ Instead of asking whether an agent is intelligent, ask: ▪ What tasks can it perform well? ▪ What inference mechanisms can it handle? ▪ How does it acquire knowledge? State of AI Research ▪ Lessons from AI Research: ▪ Problems once thought easy (e.g., face recognition) are very hard to mechanize. ▪ Problems thought hard (e.g., proving theorems) are easier to automate. ▪ AI research raises exciting philosophical questions (e.g., what is cognition?). AI's Ongoing Journey ▪ Quo Vadis: Where is AI Heading? ▪ The simplicity of AI’s questions mirrors physics in Newton’s time (e.g., “How do we recognize objects?”). ▪ AI is a field shaped by both scientific progress and changing trends, such as: ▪ Early AI and expert systems (1960s-70s). ▪ Neural networks and genetic algorithms (1980s). ▪ Intelligent agents and distributed AI (1990s). AI’s Broader Importance ▪ AI touches on understanding intelligence, societies, and cultures. ▪ Its development offers intellectual, scientific, and technological advancements that define the 20th and 21st centuries. References Honavar, V. Symbolic Artificial Intelligence and Numeric Artificial Neural Networks: Toward a Reso- lution of the Dichotomy. In: Computational Architectures Integrating Symbolic and Neural Processes. Sun, R. & Bookman, L. (Ed.) New York, NY: Kluwer, 1994. Uhr, L. & Honavar, V. Artificial Intelligence and Neural Networks: Steps Toward Principled Integra- tion. In: Artificial Intelligence and Neural Networks: Steps Toward Principled Integration. Honavar, V. & Uhr, L. (Ed). New York, NY: Academic Press, 1994.