Foundations of Artificial Intelligence: Search, Agents, and Problem Formulation
What Counts as Intelligence?
AI as an engineering and scientific discipline
Artificial intelligence studies systems that perceive, reason, act, and learn under uncertainty. At the research level, AI is not a single method but a family of formalisms for building agents that maximize expected utility or accomplish tasks in complex environments.
Core abstraction
An intelligent agent maps percept histories to actions; the quality of the mapping depends on the environment, objective, and available information.
Problem Formulation
Search problem components
Common search settings
| Setting | Challenge | Typical methods |
|---|---|---|
| Deterministic, fully observable | Large branching factor | BFS, DFS, A* |
| Stochastic | Outcome uncertainty | MDPs, dynamic programming, Monte Carlo |
| Partially observable | Hidden state | Belief-state search, POMDPs |
| Adversarial | Strategic opponents | Minimax, alpha-beta, game search |
Which component is necessary to define an optimal search problem solution?
Optimality is defined relative to a path cost or utility criterion, so a cost function is essential.
Correct answer: A path cost function
Why is heuristic search often preferred over blind search in large state spaces?
A good heuristic can dramatically reduce the effective branching factor while preserving completeness or optimality under certain conditions.
Correct answer: Because heuristics guide exploration toward promising states and reduce the number of expanded nodes.
A research workflow for search-based AI problems
-
1
Step 1: Formalize the task as states, actions, and costs.
-
2
Step 2: Identify whether the environment is deterministic, stochastic, adversarial, or partially observable.
-
3
Step 3: Choose a search or planning framework aligned with those assumptions.
-
4
Step 4: Design heuristics or abstractions to scale inference.
-
5
Step 5: Evaluate solution quality, optimality, and computational cost.