Introduction to Artificial Intelligence (CptS 440/540) Summary & Study Notes

🤖 What is AI?

Artificial Intelligence (AI) is the study and design of systems that exhibit intelligent behavior. Definitions vary by emphasis: some focus on systems that act like humans, others on systems that think like humans, and still others on systems that act or think rationally. AI blends ideas from computer science, cognitive science, mathematics, neuroscience, linguistics, and philosophy.

🎯 Why Study AI?

AI makes computers more useful and can have a huge societal impact. Building working AI systems forces us to refine theories into practical algorithms, yields cross-disciplinary benefits, and provides personal motivation through solving intriguing problems. AI research can produce immediate applied gains (e.g., fraud detection) and long-term scientific insights.

🧠 Four Approaches to Defining AI

• Acting humanly: Evaluated by the Turing Test — an agent passes if a human judge cannot reliably distinguish it from a human.
• Thinking humanly: Models that mimic human cognitive processes; aligns with cognitive science and brain-level modeling.
• Thinking rationally: Emphasizes correct reasoning and formal inference (Aristotle → modern logic).
• Acting rationally: Agents that choose actions expected to maximize goal achievement given their beliefs; a pragmatic approach used by many AI practitioners.

🧪 Turing Test and Critiques

The Turing Test measures human-like behavior through conversation. Practical variants include the Loebner Prize. Philosophical critiques like Searle’s Chinese Room argue that syntactic manipulation (producing correct outputs) does not entail genuine understanding or semantics; replies include the Systems Reply and Robot Reply which expand the locus of understanding beyond raw symbol manipulation.

🧩 Agents, Sensors, and Actuators

An agent perceives its environment via sensors and acts via actuators. The agent’s behavior is described by a function mapping percept histories to actions. Examples: humans (eyes, hands), robots (cameras, motors), webagents (web pages, hyperlinks).

📝 PEAS Framework

Use PEAS to describe tasks: Performance measure, Environment, Actuators, Sensors. Example: a taxi driver’s PEAS includes safety/comfort/profit (performance), roads and passengers (environment), steering/brake (actuators), and cameras/GPS (sensors).

🌍 Environment Properties

Key orthogonal properties used to classify environments:

• Fully observable vs. partially observable: whether current percepts give complete state info.
• Deterministic vs. stochastic/strategic: whether the next state is fully determined by the current state and action.
• Episodic vs. sequential: whether current decisions depend on previous ones.
• Static vs. dynamic: whether the environment changes while the agent deliberates.
• Discrete vs. continuous: whether state/action spaces are countable.
• Single-agent vs. multiagent: presence of other agents affects complexity. Examples: chess is fully observable and deterministic; taxi driving is partially observable, stochastic, dynamic, continuous, multiagent.

🧭 Types of Agents

Agents increase in generality and capability:

• Simple reflex agents: map current percept (or a small rule set) to actions using condition-action rules. Fast but short-sighted.
• Reflex agents with state: maintain an internal state to track unobserved aspects of the world.
• Goal-based agents: reason about future actions to achieve explicit goals.
• Utility-based agents: use a utility function to compare and choose among competing outcomes.
• Learning agents: improve performance over time by acquiring new knowledge or behaviors.

🧰 Example Systems & Techniques

Applications range from game-playing (TD-Gammon) to autonomous driving (ALVINN), medical diagnosis (Pathfinder), scheduling, webcrawlers, and robotics (soccer, rovers). Techniques include search, planning, constraint satisfaction, Bayesian networks, Monte Carlo methods, reinforcement learning, neural networks, genetic algorithms, and natural language processing.

📚 Foundations & History

AI draws on philosophy (rationality, mind), mathematics (logic, probability), economics (rational agents), neuroscience (brain computation), psychology (cognition and behavior), and linguistics (language learning). Modern CS-based AI traces back to the 1956 Dartmouth conference and pioneers like McCarthy, Minsky, Shannon, Newell, and Simon.

⚖️ Rationality vs. Intelligence

A rational agent acts to maximize expected performance according to a specified measure given its percepts and knowledge. Rationality is context-dependent: it depends on the performance measure, the environment, and the agent’s sensing/acting capabilities. Intelligent behavior is thus evaluated relative to these constraints.

🔍 Practical Notes

Designing AI systems requires explicitly modeling the task (PEAS), selecting the appropriate agent type, and choosing algorithms compatible with environment properties. Real-world constraints (partial observability, stochasticity, multiagent interactions) often motivate probabilistic and learning-based methods.