AI Advanced 4

In the context of AI definitions, what key insight differentiates rational action from mere imitation of human behavior?

• Rational action always requires a model of human cognition

• Rational action avoids heuristics in decision-making

• Rational action is constrained to perfect environmental knowledge

• Rational action seeks optimal outcomes based on well-defined performance measures, rather than just mirroring human behavior

While early AI research focused on symbolic reasoning, modern approaches also incorporate probabilistic methods. Why is this integration critical?

• Deterministic logic can represent all forms of uncertainty perfectly

• Symbolic reasoning is inherently incompatible with statistical inference

• Probabilistic methods replace logic, making classical inference unnecessary

• Probabilistic techniques allow AI systems to handle incomplete, noisy, or ambiguous information that purely symbolic methods struggle with

Consider an AI system designed to interact naturally with humans. Why might “acting like humans” be an insufficient design criterion on its own?

• It ensures optimal results in all complex tasks

• It guarantees the system will solve intractable problems efficiently

• It prioritizes human-like errors and biases over rational decision-making

• It may lead to suboptimal solutions because human behavior is not always rational or optimal

In formulating a search problem, which aspect ensures that the search process has a clear stopping condition and evaluation criterion?

• Defining arbitrary state transitions without goals

• Using heuristics that never relate to the goal

• Ignoring path costs entirely

• Establishing a well-defined goal test that determines when a goal state is reached

Depth-First Search (DFS) is memory-efficient but can become trapped exploring infinite or very deep paths. What strategy counters this drawback while retaining DFS’s advantages?

• Relying solely on breadth-first expansions

• Employing an uninformed heuristic that limits expansions

• Using no search tree at all

• Iterative Deepening Search, which incrementally increases depth limits to maintain completeness and efficiency

Informed search algorithms leverage heuristics. Under what condition does a heuristic preserve A* search’s optimality?

• Whenever it returns random values

• If it underestimates the actual cost for some states but overestimates for others

• If it is equal to or greater than the true cost at all times

• If it never overestimates the true remaining cost to the goal (admissibility)

Why might one prefer a heuristic that dominates another (i.e., is never smaller than a weaker heuristic for any state)?

• Dominating heuristics always increase search cost

• Dominating heuristics must produce incorrect solutions to guide exploration

• Dominating heuristics are easier to compute than weaker heuristics

• Dominating heuristics generally lead to fewer node expansions and more efficient searches

How does bidirectional search reduce the time complexity in certain problem spaces?

• By expanding only one direction from the start state

• By avoiding the use of heuristics entirely

• By ensuring identical branching factors in both directions

• By simultaneously searching forward from the start and backward from the goal, dramatically shrinking the search frontiers

In knowledge representation, why is mapping symbols to real-world entities systematically essential?

• Arbitrary symbols provide maximum flexibility

• Real-world correspondences make logic unnecessary

• Without such mapping, inference is purely syntactic and meaningless

• It ensures that symbol manipulation reflects real semantics and enables meaningful reasoning

Reasoning systems rely on inference rules. What advantage does a sound and complete inference procedure provide?

• It restricts the agent to only trivial conclusions

• It allows derivation of false conclusions if no alternatives exist

• It ignores relevant knowledge outside the knowledge base

• It guarantees that all truths derivable from the knowledge base can be found (completeness) and no falsehoods are derived (soundness)