Preliminary Concepts
Chapter 1
Modules in this course
- 1. theoretical Foundations
- 2. Symbolic Knowledge Representation
- forgotten AI-- basically the models and procedures to operate with symbols
- 3. Sub Symbolic Knowledge Representation
AI is not about replicating human intelligence; we need to imitate human intelligence in order
solve complex problems that we cannot solve ((FOR THIS COURSE))
- problem solving by means of computational methods in order to obtain insights from
data
reasoning is just the formal manipulation of symbols representing a collection of
propositions---- the idea here is to produce new knowledge
we have two approaches for artificial intelligence
- symbolic reasoning // manipulating symbols in order to produce new knowledge //
good old fashioned AI // ++ we can understand how the manipulation goes ++
- sub-symbolic reasoning // goal is pretty much the same but here we’re not just
manipulating symbols-- we can manipulate more complex structures eg. numbers //
-- we can not always understand how the process works / black box metaphor --
Leibniz’s Idea
- by convention we say numbers are special symbols because we have different rules
to operate these symbols
in the case of “pure” symbols we have the rules of logic in order to manipulate them to
produce new pieces of knowledge
When it comes to symbolic representations we have to fulfill two main properties;
1- we should represent the problem domain very accurately--using precise facts
2- we should clearly express how to obtain new knowledge
granularity: level of detail which we will represent the knowledge
we should select the proper granularity for particular situations
- propositions are just facts we want to investigate// they can be either true or false
- they’re abstract facts // they’re ideas we want to investigate
- sentences are the ways we have to manipulate and express those propositions
,Why do we need reasoning?
- if we’re able to encode very precisely the knowledge about a particular universe we
should be able to search the information we need----but this is not the way it works
- because the information we have and the encoding of the problem will be
limited so we cannot expect to encode every single detail of the universe
we’re operating with. Therefore, we need to complete the missing pieces of
knowledge and this is the main goal behind reasoning
Knowledge based systems
- choosing the proper granularity level is crucial
- a knowledge based system is just a reasoning system that uses this knowledge in
order to solve the problem
- we have two main sources of knowledge; - human beings, -historical data
Sub-Symbolic
- Sometimes we have data and we don’t know the knowledge properly formalized or
described. Therefore we can apply machine learning methods in order to learn the
knowledge from the data. So we can obtain the knowledge structures we need from
the data automatically. eg. by using decision trees or neural networks(even more
powerful)
- Those models are able to learn internal representations but they’re very
complex -- we’re not always able to understand what knowledge actually
represents. so we know the internal knowledge is in there however we cannot
establish a direct mapping with the problem domain
- we want to combine the transparency of some algorithms while
getting the resources and the accuracy of others #Hybrid AI
Logical Entailment
Chapter 2
recap
- structures can be encoded using symbols
- According to Leibniz’s idea, every computational method would be symbolic as
numbers are also symbolic representations of abstract quantities
- however we decided to establish a clear difference between numerical
quantities and symbolic structures
- the idea here is to discover the relationship between those symbols in order
to draw conclusions or perform the symbolic reasoning
- Therefore we need a procedure for thinking--logical entailment
We want to manipulate the symbols, the different relationships and also the
representations we have in order to solve problems
Logic allows us to deal with abstract ideas in terms of concrete symbols, so the
manipulation of those symbols mirrors the relations among the ideas.
, And the logical entailment procedure has a strong foundation on mathematical logic
In particular we’re going to implement the back chaining procedure which is the
computational implementation of the logical entailment
Representing Knowledge
- Available knowledge
- In symbolic AI, this knowledge must be as precise as possible while avoiding
redundancy
- We’ll use IF-THEN rules
- Atomic Sentences: basic sentences, facts
- conditional sentences: with the form
- variables--capitalized
- constants--uncapitalized
we’re not allowed to make any claims that cannot be supported
by the knowledge included in the knowledge base
- Back-chaining is a logical procedure used to establish whether an atomic sentence 𝑄
(called query) is entailed by the available knowledge or not
The Prolog Language
Chapter 3
- Prolog is invented by Colmerauer for the purpose of natural language parsing
- Declarative programming is a programming paradigm (style of building the structure
and elements of programs) that expresses the logic of computations without
describing its control flow
- Imperative programming languages such as Python need well-defined steps
- Declarative programming tries to minimize or eliminate side effects by describing
what the program must accomplish in terms of the problem domain, rather than
describing how to accomplish it as a sequence of the language primitives and
instructions
- It is powerful because it uses high-level clauses describing what the program
should accomplish
- Prolog has its roots in first-order logic and formal logic. Unlike many other
programming languages, Prolog is intended primarily as a declarative programming
language: the program logic is expressed in terms of relations, represented as facts
and rules.
- such that a computation (reasoning) is initiated by running a query over these
relations
Chapter 1
Modules in this course
- 1. theoretical Foundations
- 2. Symbolic Knowledge Representation
- forgotten AI-- basically the models and procedures to operate with symbols
- 3. Sub Symbolic Knowledge Representation
AI is not about replicating human intelligence; we need to imitate human intelligence in order
solve complex problems that we cannot solve ((FOR THIS COURSE))
- problem solving by means of computational methods in order to obtain insights from
data
reasoning is just the formal manipulation of symbols representing a collection of
propositions---- the idea here is to produce new knowledge
we have two approaches for artificial intelligence
- symbolic reasoning // manipulating symbols in order to produce new knowledge //
good old fashioned AI // ++ we can understand how the manipulation goes ++
- sub-symbolic reasoning // goal is pretty much the same but here we’re not just
manipulating symbols-- we can manipulate more complex structures eg. numbers //
-- we can not always understand how the process works / black box metaphor --
Leibniz’s Idea
- by convention we say numbers are special symbols because we have different rules
to operate these symbols
in the case of “pure” symbols we have the rules of logic in order to manipulate them to
produce new pieces of knowledge
When it comes to symbolic representations we have to fulfill two main properties;
1- we should represent the problem domain very accurately--using precise facts
2- we should clearly express how to obtain new knowledge
granularity: level of detail which we will represent the knowledge
we should select the proper granularity for particular situations
- propositions are just facts we want to investigate// they can be either true or false
- they’re abstract facts // they’re ideas we want to investigate
- sentences are the ways we have to manipulate and express those propositions
,Why do we need reasoning?
- if we’re able to encode very precisely the knowledge about a particular universe we
should be able to search the information we need----but this is not the way it works
- because the information we have and the encoding of the problem will be
limited so we cannot expect to encode every single detail of the universe
we’re operating with. Therefore, we need to complete the missing pieces of
knowledge and this is the main goal behind reasoning
Knowledge based systems
- choosing the proper granularity level is crucial
- a knowledge based system is just a reasoning system that uses this knowledge in
order to solve the problem
- we have two main sources of knowledge; - human beings, -historical data
Sub-Symbolic
- Sometimes we have data and we don’t know the knowledge properly formalized or
described. Therefore we can apply machine learning methods in order to learn the
knowledge from the data. So we can obtain the knowledge structures we need from
the data automatically. eg. by using decision trees or neural networks(even more
powerful)
- Those models are able to learn internal representations but they’re very
complex -- we’re not always able to understand what knowledge actually
represents. so we know the internal knowledge is in there however we cannot
establish a direct mapping with the problem domain
- we want to combine the transparency of some algorithms while
getting the resources and the accuracy of others #Hybrid AI
Logical Entailment
Chapter 2
recap
- structures can be encoded using symbols
- According to Leibniz’s idea, every computational method would be symbolic as
numbers are also symbolic representations of abstract quantities
- however we decided to establish a clear difference between numerical
quantities and symbolic structures
- the idea here is to discover the relationship between those symbols in order
to draw conclusions or perform the symbolic reasoning
- Therefore we need a procedure for thinking--logical entailment
We want to manipulate the symbols, the different relationships and also the
representations we have in order to solve problems
Logic allows us to deal with abstract ideas in terms of concrete symbols, so the
manipulation of those symbols mirrors the relations among the ideas.
, And the logical entailment procedure has a strong foundation on mathematical logic
In particular we’re going to implement the back chaining procedure which is the
computational implementation of the logical entailment
Representing Knowledge
- Available knowledge
- In symbolic AI, this knowledge must be as precise as possible while avoiding
redundancy
- We’ll use IF-THEN rules
- Atomic Sentences: basic sentences, facts
- conditional sentences: with the form
- variables--capitalized
- constants--uncapitalized
we’re not allowed to make any claims that cannot be supported
by the knowledge included in the knowledge base
- Back-chaining is a logical procedure used to establish whether an atomic sentence 𝑄
(called query) is entailed by the available knowledge or not
The Prolog Language
Chapter 3
- Prolog is invented by Colmerauer for the purpose of natural language parsing
- Declarative programming is a programming paradigm (style of building the structure
and elements of programs) that expresses the logic of computations without
describing its control flow
- Imperative programming languages such as Python need well-defined steps
- Declarative programming tries to minimize or eliminate side effects by describing
what the program must accomplish in terms of the problem domain, rather than
describing how to accomplish it as a sequence of the language primitives and
instructions
- It is powerful because it uses high-level clauses describing what the program
should accomplish
- Prolog has its roots in first-order logic and formal logic. Unlike many other
programming languages, Prolog is intended primarily as a declarative programming
language: the program logic is expressed in terms of relations, represented as facts
and rules.
- such that a computation (reasoning) is initiated by running a query over these
relations
