Reflections on Chapter 7-10

Chapter 7: First-Order Logic

The chapter focuses how first-order logic, a general-purpose representation language that is based on an ontological commitment to the existence of objects and relations in the world, can be utilizes as the representation language for a knowledge-based agent. Constant symbols and predicate symbols name objects and relations, respectively using function symbols. Atomic sentence consists of a predicate applied to one or more terms. Complex sentences use connectives like propositional logic and quantified sentences allowing the expression of general rules. Using first-order logic, an agent that reasons can be defined wherein it needs to: react to what it perceives; extract abstract descriptions of the current state from percepts; maintain an internal model of relevant aspects of the world that are not directly available from percepts; express and use information about the desirability of actions in various circumstances; use goals in conjunction with knowledge about actions to construct plans. Using the conventions of situation calculus, the knowledge about the actions and their effects can be characterized in which this knowledge enables the agent to keep track of the world and to assume the effects of plans of action. Causal rules are often more flexible and entail a wider range of consequences, but can be more expensive to use in inference.

Chapter 8: Building a Knowledge Base

In this chapter shows the process of representing knowledge of a domain goes through several stages. Starting with the informal stage which involves deciding what kinds of objects and relations need to be represented or what we called ontology. After that a vocabulary is chosen, and employ to encode general knowledge of the domain. After encoding specific problem instances, automated inference procedures can be used to solve them. Good representations eliminate irrelevant detail, capture relevant distinctions, and express knowledge at the most general level possible. Constructing knowledge-based systems has advantages over programming. Special-purpose ontologies, such as the one constructed for the circuits domain, can be effective within the domain but often need to be generalized to broaden their coverage. A general-purpose ontology needs to cover a wide variety of knowledge, and should be capable in principle of handling any domain. This chapter presented a general ontology based around categories and the event calculus and covered structured objects, time and space, change, processes, substances, and beliefs. It also presented a detailed analysis of the shopping domain, exercising the general ontology and showing how the domain knowledge can be used by a shopping agent. Lastly, it is important remembering that the nature of a good representation depends on the world being represented.

Chapter 9: Inference in First-Order Logic

In this chapter, an analysis of logical inference in first-order logic is presented and a number of algorithms for doing it. A simple extension of the prepositional inference rules allows the construction of proofs for first-order logic. Unfortunately, the branching factor for the quantifier is huge. The use of unification to identify appropriate substitutions for variables eliminates the 5 instantiation step in first-order proofs, making the process much more efficient. A generalized version of Modus Ponens uses unification to provide a natural and powerful I inference rule, which can be used in a backward-chaining or forward-chaining algorithm. The canonical form for Modus Ponens is Horn form: p\ A ... A pn => q, where pt and q are atoms. This form cannot represent all sentences, and Modus Ponens is not a complete proof system. The generalized resolution inference rule provides a complete system for proof by refutation. It requires a normal form, but any sentence can be put into the form.

Chapter 10: Logical Reasoning Systems

This chapter gives an association between the conceptual foundations of knowledge representation and reasoning and the practical world of actual reasoning systems. It highlights that real understanding of these systems can only be obtained by trying them out. It has explains the implementation techniques and characteristics of four major classes of logical reasoning systems: (1) Logic programming systems and theorem provers; (2) Production systems; (3) Semantic networks; (4) Description logics. It also has seen that there is an exchange between the expressiveness of the system and its efficiency. Compilation can provide significant improvements in efficiency by taking advantage of the fact that the set of sentences is fixed in advance. Usability is enhanced by providing a clear semantics for the representation language, and by simplifying the execution model so that the user has a good idea of the computations required for inference.

Chapter 7: First-Order Logic

First-Order Logic is a logic that makes a stronger set of ontological commitments. It has been so important to mathematics, philosophy and artificial intelligence precisely because those fields – and indeed, much everyday human existence – can be usefully thought of as dealing with objects and the relations between them. It can also express facts about all of the objects in the universe. Although it commits to the existence of objects and relations, it does not make an ontological commitment to such things as categories, time, and events, which also seem to show up in most facts about the world. Constant symbols and predicate symbols name objects and relations respectively. We have a choice of writing diagnostic rules that reason from percepts to propositions about the world or casual rules that describe how conditions in the world cause percepts to come about.

Chapter 8: Building a Knowledge Base

The process of building a knowledge base is called knowledge engineering. A knowledge engineer is someone who investigates particular domain, determines what concepts are important in that domain, and creates a formal representation of the objects and relations in the domain. One does not become a proficient knowledge engineer just by studying the syntax and semantics of a representation language. It takes a lots of examples before one can develop a good style in any language, be it a language for programming, reasoning, or communicating. To explain the general principles of good design, we need to have an example. Every knowledge base has two potential consumers: human readers and inference procedures. The main advantage of knowledge engineering is that it requires less commitment, and thus less work. The knowledge engineer specifies what is true, and the inference procedure figures out how to turn the facts into a solution to the problem. One cannot understand knowledge representation without doing it, or at least seeing it. Good representations eliminate irrelevant detail, capture relevant distinctions, and express knowledge at the most general level possible. Finally it is worth recalling that the nature of an appropriate representation depends on the world being represented and the intended range of uses of the representation.

Chapter 9: Inference in First-Order Logic

Logical inference was studied extensively in Greek mathematics. The type of inference most carefully studied by Aristotle was the syllogism. A simple extension of the propositional inference rules allows the construction of proofs for the first-order logic. The generalized resolution inference rule provides a complete proof system for the proof by refutation. It requires a normal form, but any sentence can built into the form. Thus, we have serious difficult, in the form of a collection of operators that give long proofs and a large branching factor, and hence a potentially explosive search problem. The canonical form for Modus Ponens mandates that each sentence in the knowledge base be either an atomic sentence or an implication with a conjunction of atomic sentences on the left hand side and a single atom on the right. The generalized Modus Ponens rule can be used into ways; Forwarding chaining and Backward Chaining. Chaining with resolution is more powerful than chaining with Modus Ponens, but it is still not complete.

Chapter 10: Logical Reasoning Systems

It is a good idea to build agents as reasoning systems – systems that explicitly represent and reason with knowledge. The main advantage of this is a high degree of modularity. All knowledge-based systems rely on the fundamental operation of retrieving sentences satisfying certain conditions – for example, finding an atomic sentence that unifies with a query, or finding an implication that has given atomic sentence as one of its premises. Recently, much of the effort in logic programming has been aimed toward increasing efficiency by building information about specific domains or specific inference patterns into the logic programming language.