Artificial Intelligence 2074

Thinking humanly and Thinking rationally are concerned with thought process and reasoning.

Thinking Humanly

Defⁿ: The exciting new effort to make computers think machines with minds, in the full and literal sense.

If we are going to say that a given program thinks like a human, we must have some way of determining how humans think. We need to get inside the actual workings of human minds. There are two ways to do this: through introspection--trying to catch our own thoughts as they go by--or through psychological experiments. Once we have a sufficiently precise theory of the mind, it becomes possible to express the theory as a computer program. If the program's input/output and timing behavior matches human behavior, that is evidence that some of the program's mechanisms may also be operating in humans. 

Thinking Rationally

Defⁿ: The study of mental faculties through the use of computational models.

The laws of thought are supposed to implement operation of the mind and their study initiated the field called logic. It provides precise notations to express facts of the real world.

It also includes reasoning and "right thinking" that is irrefutable thinking process. Also computer program based on those logic notations were developed to create intelligent system.

A rational agent is an agent which has clear preferences and models uncertainity via expected value. It always performs right action, right action  means the action that causes the agent to be most successful in the given percept sequence.

Rational agent is capable of taking best possible action in any situation.

For each possible percept sequence, a rational agent should select an action that is expected to maximize its performance measure, given the evidence provided by the percept sequence and whatever built-in knowledge the agent has.

Episodic task environment vs Sequential task environment

In Episodic environment, the agent's experience is divided into atomic 'episodes' (each episode consists of the agent perceiving then performing a single action i.e. agent's single pair of perception & action) and every episode is independent of each other. The subsequent episodes do not depend on actions occured in previous episodes. For e.g. an agent sorting defective part in an assembly line, pick and place robot agent.

In Sequential environment, current actions may affect all future decisions. In sequential environment, an agent requires memory of past action to determine the next best action. For e.g. a taxi driving agent or Chess playing.

In goal formulation, we decide which aspects we are interested in and which aspects can be ignored. In the goal formulation process, the goal is to be set and we should assess those states in which the goal is satisfied. In problem formulation, we decide how to manipulate the important aspects, and ignore the others. So, without doing goal formulation, if we do the problem formulation, we would not know what to include in our problem and what to leave, and what should be achieved. So problem formulation must follow goal formulation. That means problem formulation must be done only after the goal formation is done.

In the state space representation of a problem, nodes of a graph correspond to partial problem solution states and arcs represent steps in a problem- solving process. An initial state, corresponding to the given information in a problem instance, forms the root of the graph. The graph also defines a goal condition, which is the solution to a problem instance. State space search characterizes problem solving as the process of finding a solution path from the start state to a goal state. Arcs of the state space correspond to steps in a solution process and path through the space represent solutions in varying stages of completion. Paths are searched, beginning at the start state and continuing through the graph until either the goal description is satisfied or they are abandoned.

E.g.

State Space representation of Vacuum World Problem:

States: two locations with or without dirt: = 8 states.

Initial state:  Any state can be initial

Actions:  {Left, Right, Suck}

Goal test: Check whether squares are clean.

Path cost:  Number of actions to reach goal.

Uniform cost searching algorithm is used for weighted state space. In this algorithm,

• Priority queue for storing nodes in state space is maintained, where least cost paths are given higher priority.
• Node at head of the queue is expanded first.
• The queue is updated at each expansion of nodes (deletion of node visited & insertion of node to be visited).

Given state space:

Alpha Beta pruning is needed to eliminate unnecessary nodes from state space. It has two values alpha and beta.

Alpha is the best (i.e. maximum) value found so far at any choice point along the path for MAX.

Beta is the best (i.e. minimum) value found so far at any choice point along the path for MIN.

Knowledge representation (KR) is the study of how knowledge about the world can be represented and what kinds of reasoning can be done with that knowledge. Knowledge representation is the method used to encode knowledge in Intelligent Systems.

Since knowledge is used to achieve intelligent behavior, the fundamental goal of knowledge representation is to represent knowledge in a manner as to facilitate inferencing (i.e. drawing conclusions) from knowledge. A successful representation of some knowledge must, then, be in a form that is understandable by humans, and must cause the system using the knowledge to behave as if it knows it.

Semantic Network Representation

Semantic networks can

    - show natural relationships between objects/concepts

    - be used to represent declarative/descriptive knowledge

Knowledge is represented as a collection of concepts, represented by nodes. Thus, semantic networks are constructed using nodes linked by directional lines called arcs

A node can represent a fact description

    - physical object

    - concept

    - event

An arc (or link) represents relationships between nodes. There are some 'standard' relationship types

    - 'Is-a' (instance relationship): represent class/instance relationships

    - 'Has-a' (part-subpart relationship): identify property relationships

Example:

The most specific hypotheses (i.e., the specific boundary SB) are the hypotheses that cover observed positive training examples, and as little of the remaining feature space as possible. These are hypotheses which if reduced any further would exclude a positive training example, and hence become inconsistent. These minimal hypotheses essentially constitute a (pessimistic) claim that the true concept is defined just by the positive data already observed: Thus, if a novel (never-before-seen) data point is observed, it should be assumed to be negative. (I.e., if data has not previously been ruled in, then it's ruled out.)

The most general hypotheses (i.e., the general boundary GB) are those which cover the observed positive training examples, but also cover as much of the remaining feature space without including any negative training examples. These are hypotheses which if enlarged any further would include a negative training example, and hence become inconsistent.

Candidate Elimination Algorithm is used to construct generalization specialization tree.

• Nodes in the generalization tree are connected to a model that matches everything in its subtree.
• Nodes in the specialization tree are connected to a model that matches only one thing in its subtree.

Method:

Machine vision is the ability of a computer to "see." A machine-vision system employs one or more video cameras, analog-to-digital conversion ( ADC ), and digital signal processing ( DSP ). The resulting data goes to a computer or robot controller. Machine vision is similar in complexity to voice recognition . The machine vision systems use video cameras, robots or other devices, and computers to visually analyze an operation or activity. Typical uses include automated inspection, optical character recognition and other non-contact applications.

Two important specifications in any vision system are the sensitivity and the resolution. Sensitivity is the ability of a machine to see in dim light, or to detect weak impulses at invisible wavelengths. Resolution is the extent to which a machine can differentiate between objects. In general, the better the resolution, the more confined the field of vision.

A typical machine vision system will consist of most of the following components:

• One or more digital or analogue cameras (black-and-white or colour) with suitable optics for acquiring images, such as lenses to focus the desired field of view onto the image sensor and suitable, often very specialized, light sources
• Input/Output hardware (e.g. digital I/O) or communication links (e.g. network connection or RS-232) to report results
• A synchronizing sensor for part detection (often an optical or magnetic sensor) to trigger image acquisition and processing and some form of actuators to sort, route or reject defective parts
• A program to process images and detect relevant features.

Distribution over Y can be obtained by summing out all the other variables from any joint distribution containing Y. This process is called marginalization.

    P(Y) =ΣP(Y, Z)

Z --> No. of variables where Y seems to be true.