Problem Characteristics
7 standard characteristics of an AI problem are defined in classical AI theory (often used to analyze
the "Problem Space).
These characteristics help developers choose the right AI architecture, whether it's a simple search
algorithm or a complex neural network.
1. Is the problem Decomposable?
Can the problem be broken down into smaller, independent sub-problems?
Explanation: If a problem is decomposable, we can solve the parts separately and combine
the results. This makes the overall task much easier to manage.
Example: Symbolic Integration.
Non-Decomposable Example: The Rubik’s Cube. You cannot solve one face without affecting
the others. Every move is interconnected.
2. Can solution steps be Undone or Ignored? (Recoverability)
How much risk is involved in taking a single step toward a solution?
Ignorable: Mistakes can be discarded without penalty (e.g., Theorem Proving).
Recoverable: Mistakes can be undone by "backtracking" (e.g., 8-Puzzle). You can move a tile
back to where it was.
Irrecoverable: Mistakes are permanent (e.g., Robotic Surgery). You cannot "undo" a physical
cut. This requires the AI to be extremely cautious and perform high-level planning.
3. Is the Universe Predictable? (Certainty)
Does the environment change randomly, or is it stable?
Deterministic: The outcome of an action is 100% certain. For example, in Chess, if you move
your Knight to a specific square, it stays there.
Stochastic (Non-deterministic): The outcome is uncertain. For example, in Bridge or Poker,
the next card dealt is a matter of probability. Self-driving cars operate in a stochastic
universe because they cannot predict exactly what a human pedestrian will do.
4. Is the solution Absolute or Relative? (Optimality)
Is any path to the goal fine, or do we need the best one?
Absolute: You just need to reach the goal. In the Water Jug Problem, if you need 2 liters, it
doesn't matter if you got there in 5 steps or 10 steps.
Relative: The quality of the solution depends on the cost. In the Traveling Salesman
Problem, finding any path between 20 cities is easy, but finding the shortest path is the
actual challenge.
5. Is the solution a State or a Path?
Is the goal the final configuration of the world, or the series of actions taken to get there?
State: We only care about the result. In Natural Language Processing (NLP), when translating
a sentence, we only care that the final output is correct; we don't care about the order in
which the AI processed the grammar rules.
Path: The sequence of steps is the actual answer. In a Bank Robbery or Security Simulation,
the "path" (the sequence of actions to bypass security) is what the user wants to see.
6. What is the Role of Knowledge?
How much "extra" information is required to solve the problem?
Low Knowledge: The rules of the game are enough (e.g., Chess). You don't need to know
history or politics to play chess.
High Knowledge: Requires a massive database of facts (e.g., Medical Diagnosis). An AI
cannot diagnose a disease just by "searching" through symptoms; it needs specific, expert-
level knowledge about biology, drugs, and patient history.
7. Does the task require Interaction with a Person?
Is the AI working in a vacuum, or is it part of a human conversation?
Solitary: The AI solves the problem and gives the answer (e.g., a Sudoku solver).
Conversational: The AI must understand human intent, handle interruptions, and explain its
reasoning (e.g., ChatGPT or a Diagnostic Assistant). This adds a layer of complexity regarding
natural language and social intelligence.