Turing Test Approach

Cognitive Modeling Approach

Laws of Rational Thought Approach

Rational Agent Approach

Cognitive -Thinking human

Turing -Acting human

Rational Thought -Thinking rationally

Rational Agent - Acting rationally

• Pro: Empirical, does not define intelligence
• Con: Limited to symbolic tasks, human based, cannot have original thought, cannot create perfect rule-set

• Environment is multiagent if performance measures depend on other agent's behavior. If an environment is multiagent, it is classified cooperative or competitive

-Table Lookup

-Simple Reflex

-Reflex with memory

-Goal-based

-Utility-based

• Table Look-up - Look through table for option based on table of percepts
• Pro: Simplest, can be used with other agent types
• Con: Size of table, construction of table, not autonomous

Math - logic, numerical representations of conditions

Linguistics - Symbols, how data is represented/interpretted

Psych - Understanding of human mind, thoughts, processes

Philosophy - the mind body problem

(cond (test s-expression-sequence)

(test s-expression-sequence)
...

)

• (eq s1 s2): T if s1 and s2 point to the same object
• (eql s1 s2): T if s1 and s2 point to the same object, or if same number (and same type), or same characters
• (equal s1 s2): T if s1 and s2 evaluate to same value
• (equalp s1 s2): Same as equal, but returns T if s1 and s2 are same number regardless of type

• (map-fn-name applied-fn lists)
• lists must have same numbers of elements
• Must be 1 list for each argument of applied-fn

• (cons a b)
• Cons cell (dotted pair) - More primitive than list
• Syntax: (lobject . robject)
• Consists of a "left" pointer lobject and a "right" pointer robject (to S-expressions

• Pro: Thinks like a human, can be conducted through multiple avenues (introspection, experiments, brain imaging)
• Con: Requires understanding of how people think

• Pro: More general than Law of Thought, more applicable to scientific analysis
• Con: Finding a perfect/best COA is not always possible, and is impossible in any complex environment

• Pro: Formal, provable, mathematical approach
• Con: difficult language to use, limited computational resources

Performance

Environment

Actuators

Sensors

• Initial State - State from which problem solving begins
• Action - a step the agent can carry out
• Transition model - results of the action
• Goal test - tests to see if it is the goal state
• Path cost function - computes cost of path from initial state to current state

• Branching factor: move in direction of smaller branching factor
• Are actions reversible: cannot do goal-directed if not
• Number of targets: move in direction that has greater number of targets

• Start with k generated states
• Spawn a new generation
• Choose best chromosomes of the new generation
• Randomly combine pairs of good chromosomes
• Continue creation and combination of chromosomes until solution or some pre-set limit reached.

h_a dominates h_b if h_a(n) ≥ h_b(n)

Dominating heuristic will expand fewer nodes and be more accurate, making it the preferred choice.

A relaxed problem is one with fewer constraints, and the cost of the optimal solution to the relaxed problem is an admissible heuristic in the original.

EXAMPLE: The 8 puzzle

• Concise
• Unambiguous
• Can add to it

• Locational Info - Difficult to represent that something is only in one place at a time
• Relational Info - many real world relations preclude others (facing east means you can't be facing west)
• Actions - actions must have sufficient preconditions to prevent accidental simultaneous execution

• Constants to objects
• Predicates to relations among objects
• Functions to functions in the domain

If an object is substituted for x that makes P(x) false, that particular implication will be true; if the object makes P(x) true, then Q(x) is true (by modus ponens) and the implication is true again.

What are the semantics of

UNIVERSALLY QUALIFIED(x) P(x)

• Closed world assumption
• Closed domain assumption
• Unique names assumption

1. Eliminate implication
2. Reduce scope of NOT
3. Standardize variables
4. Eliminate existential
5. Move quantifiers to left
6. Drop prefix
7. Convert into conjunctions of disjuncts
8. Write as a set of clauses
9. Rename variables so each clause has unique variables

• Using more efficient data structures: Linked list, predicate indexing, multi-indexing in order from worst to best
• Using a more efficient search strategy for ASK

• Simple Reflex - exact cause and effect
• Pro: Simple, predictable
• Con: No internal representation of world, prone to infinite loops

• Reflex w/Memory - reflex based with knowledge of world
• Pro: utilizes knowledge, uses rules
• Con: No mechanism for making decisions

• Goal-based - incorporation of goals allows for decision making
• Pro: decision making, plans, more flexible than reflex based agents
• Con: slower than reflex as it must reason

• Utility-based - goal based with a a heurisitic about which decisions to choose
• Pro: chooses desirable way to goal
•  Con: slowest, most complex

Describe experiment (1 AI and 1 human are asked questions by a human interrogator, and AI completes test if interrogator is unable to determine responses are from a human)

Rational Agent - Agent acts intelligently, given a set of beliefs and goals, agent performs appropriate actions to achieve those goals

Laws of Rational Thought - Thinking rationally, based in math and seeks the best answer. may run forever if no solution exists