lpai.syllv6 - Minds & Machines Home

advertisement
03/08/16
Draft
1
LOGIC (PROGRAMMING) AND ARTIFICIAL INTELLIGENCE
CSCI 6967, PHIL 6240 (& PSYC)
Web page: http://www.rpi.edu/~brings/ilogprog.html
Mondays: 1/8/2001 – May 14, 2001 (w/ some “shift” days; see schedule),
Sage 4112
12PM – 3PM
A graduate seminar, Spr. 2001, Comp. Sci. (and Phil/Psych), Rensselaer
INSTRUCTORS
Selmer Bringsjord
selmer@rpi.edu
David Ferrucci
ferrucci@us.ibm.com
Guest Lectures: Steinar Flatland, DDC and Adam Lally, DDC
GENERAL ORIENTATION
Synopsis
This course will focus on the use of certain techniques in logic-based AI to produce
intelligent software systems that autonomously and semi-autonomously create and
produce documents. Some of these systems will be specifically related to literary
documents (as in the Brutus story generation system); others will relate to business
documents central to industry (e.g., proposals, statements of work, reports, exams, etc.).
Required Books
1. The Art of Prolog, Sterling & Shapiro (AOP)
2. Artificial Intelligence: A Modern Approach, Russell & Norvig (AIMA)
3. Artificial Intelligence and Literary Creativity, Bringsjord & Ferrucci (AILC)
“Background” Books
1. Language, Proof, and Logic, Barwise and Etchemendy (1st course math. logic;
worth adding to your library if you’re serious about logicist AI (LAI)). From
Seven Bridge Press.
2. Hyperproof, Barwise & Etchemendy (start of visual theorem prover; very helpful
system for representing diagrams and natural deduction style proofs) From
CSLI.
3. Mathematical Logic (2nd edition), Ebbinghaus, Flum, Thomas (2nd course math.
logic; worth adding to your library if your serious about LAI). This book covers,
in a new chapter for the excellent new edition, logic programming. From
Springer-Verlag.
4. Logical Foundations of AI, Geneserth & Nilsson (coverage of logic for
computation/AI beyond standard mathematical logic; again, a good book to
have, older now, but a classic)
03/08/16
Draft
2
Papers on Reserve (or online)
 A Taxonomy of the Part-Whole Relations, Winston & Chaffin (PWR)
 What IS-A is and isn’t, Brachman (ISA)
 Classes versus Prototypes: Some Philosophical and Historical Observations, Taivalsaari
 “Logic and Artificial Intellience: Divorced, Separated, Still Married?” Bringsjord
and Ferrucci, Minds & Machines 8: 273-308. Preprint available directly from
Bringsjord.
 “Creativity, the Turing Test, and (the better) Lovelace Test,” Bringsjord, Bello,
Ferrucci, forthcoming in Minds & Machines. Preprint available at
http://www.rpi.edu/~faheyj2/SB/SELPAP/DARTMOUTH/lt3.pdf)
 Others to follow, TBA.
 Also, a number of key documents are available on-line, and students should
obtain them ASAP on their own. These documents include the manual for the
OTTER system, and documents that explain John Pollock’s approach to logic
programming through natural deduction, and his OSCAR system. See the
relevant web sites for documents/info. Searching via the expected keywords
will be effective, but links are available to OTTER and John Pollock’s material
from the course web page.
Course Objective
The principal objective is for students to develop a deeper appreciation of the theoretical
and engineering challenges of formally representing and reasoning about commonsense knowledge for the purposes of simulating and assisting human activity in the
production of documents.
The instructors are committed to cutting-edge R&D in the intersection of logic, AI, and
document creation and production in multiple domains, and through this course offer
up the potential for future involvement for grad students in this R&D.
An overview of the theory and practice in knowledge representation and formal
reasoning and logic is provided with practical examples in logic programming.
Three document generation systems are examined in detail. The first, Brutus, is an
academic prototype developed to explore computational models of literary creativity.
The second, DocCon, is a commercial Knowledge-Based Interactive Document
Development (KIDD) system designed to assist in the representation, reuse and
application of content in the production of business documents. The third, eWriter, is a
prospective system for automatically and semi-automatically generating test items
appearing on “high stakes” standardized tests (e.g., GRE, LSAT). The eWriter system
will be analyzed in theory as a specialization of a document generation system.
Prerequisites: Intro to AI (or equivalent), and some previous mathematical logic.
03/08/16
Draft
3
Course Outline
Week 1 (1/8/2001): Overview and Background
Lecturer: Dave and Selmer
Content
 Objective and Course Outline
 Course Logistics
 History and Foundations of AI
 Euclid to “What is a proof? to mathematical logic (Frege) to computation
(Turing, Church, Kleene, Godel, etc.) to modern-day computers
 Introduction to Logic, AI, Creativity and Writing
 Including the stages from AILC text
 Introduction to Logic Programming (glimpses only)
 in “Prologish” form (Dave)
 via resolution-based theorem proving (foundations of OTTER; Selmer)
 via natural deduction using HYPERPROOF (foundations of OSCAR; Selmer)
 Application Overview: Goals Architecture and HCI
 Brutus: Computational Aspects of Literary Creativity
 KIDD: Knowledge-Based Interactive Document Development
 Test Authoring and Generation
Readings: AIMA (1,2)
Week 2 (01/15/2001): Knowledge Representation I
Lecturer: Selmer
Content
 Formal Representations and Inference
 Mathematical machinery needed to build first-order logic; first-order logic presented
as a so-called 'logical system.'
 Some traditional formalizations (e.g., ZF set theory), basic metatheoretical results.
 Review of FOL: Syntax, Semantics, Inference Formal Properties and their practical
implications
 Higher Order Logics. They Exist. What are they? How are they limited? Why would
anyone use them?
Readings: AIMA (6, 7, 9)
Week 3 (01/22/2001): Introduction to Logic Programming
Lecturer: Dave
Content
 Concepts
 Basic Constructs and Techniques in Prolog
 Discussion and examples
o Database programming
o Recursive Programming
o Searching, Non-deterministic Programming
o Meta-Programming
03/08/16
Draft
4
Readings: AIMA (10.1-10.3, 21), AOP (1,2,3)
Software: Prolog
Assignment I: Describe a classic programming problem, design a solution, Write a
Logic Program to implement the solution. Discuss the pros and cons of LP and a
programming paradigm and Prolog as an implementation language. (10%, 2 weeks)
Week 4 & 5 (1/29/2001, 2/05/2001): Applications in Logic Programming
Lecturer: Dave & (for 2nd week for last bullet below) Selmer
Content
 Review of Inference in First-Order Logic
 Mechanisms for Deduction: Forward Chaining, Backward Chaining, Resolution
 Examples in Prolog
 Simple Forward Chainer
 Simple Backward Chainer
 Knowledge Belief and the use of Negation: Mystery Solver
 Examples in OTTER and (possibly) OSCAR; TELL & ASK w/ LISP
Readings: AIMA (9, 10.4-10.5), AOP (4-6)
Assignment II: Describe a classic AI problem in terms of knowledge and inference.
Design a solution. Write a Logic Program to implement the solution. Discuss the pros
and cons of LP as an implementation language for this problem. (15 %, 2 weeks)
Week 6 (2/12/2001): Knowledge Structures and Domain Modeling
Lecturer: Dave
Content
 The world as objects and attributes
 Classic Links and Their Semantics
o Objects and their relationships
o Intension and Extension (“instance-of”)
o Classification (“is-a”, “sublcass” etc.)
o Aggregation (“part-of” and its variations)
o Classes and Prototypes
 The Meta-Modeling Problem – Spanning Domains
Readings: AIMA (6, 8, 10.6), PWR, ISA
Assignment III Option 1:
Design and Implement 3 knowledge structures and their inferences as a Logic
Program. Use this program to represent a sample domain and demonstrate its
ability to reason about that domain. (35%, 4 weeks, Due on Week 10, 15 Min
oral presentation required)
03/08/16
Draft
5
Week 7 (2/19/2001): Introduction to Software and Data Modeling
Guest Lecturer: Steinar Flatland, DDC
Content
 Introduction to Unified Modeling Language (UML)
o Examples in Software Design
 Introduction to the Entity-Relationship Data Modeling
o Examples in Data Modeling
 UML vs E-R
Readings: Handouts
Assignment III Option 2:
Discuss the objectives, strengths and limitations of UML and ER modeling as
domain modeling languages. Compare and contrast them to FOL as a domain
modeling language. How do they support the “knowledge structuring”
relationships discussed previously? Pick a domain and with examples show the
limitations of UML in accurately modeling the domain. What representation
structures would you use to more accurately model the domain and why.
Develop your example to illustrate your argument. Prepare to present and
discuss your example. (35%, Due on Week 10, 2 person groups encouraged, 15
min oral presentation required)
Week 8 (2/26/2001): Automatic Story Generation I
(Tuesday follow Monday’s Schedule)
Lecturer: Selmer
Content
 System Objectives
 Historical Perspective
 The problem of interestingness
 Computational Elements of Creativity
 A new conceptual approach: Anchoring the Theme (rather than the plot)
 Formal Representations of Betrayal in First Order Logic
Readings: AILC (1-4)
Note: Assignment II Collection
Assignment IV Option 1:
Pick a literary theme and develop a formal representation for it similar to the one
presented for Betrayal. Build a thematic relation and extend the Micro Brutus.1
representation structures to instantiate your new theme and generate a new
story. (50%, Due on Week 14)
Week 9 (03/5/2001): Automatic Story Generation II (Dave)
Lecturer: Dave
Content
 The Brutus Architecture for Story Generation
 Idealized Implementation
03/08/16



Draft
6
Representation and Reasoning Features of FLEX
o Forward and Backward Chaining: Planning and Acting in Brutus,
o Sentence Generation in Brutus
Overview of an Brutus Implementation in FLEX
Discussion: Engaging the Human in the Creative Process
Readings: AILC (6.3-6.7)
Software: Micro Brutus.1 will be distributed.
***************3/12/2001 Spring Break*****************
Week 10 (03/19/2001): Domain Modeling Presentations and Discussions
7 X 20 Min Team Presentations
Note: Assignment III Collection
Week 11 (03/26/2001): Introduction to KIDD
Lecturer: Dave
Content
 Knowledge-Based Interactive Document Development
 Background and Motivation
 Solution Concept
 Technical and Business Issues
 Representing Document and Domain Knowledge
 Challenges for an Interactive Reasoning System
 Introduction to DocCon a KIDD system implementation
 Introduction to the DKE Process and the DK Design Document
 Examples in Drafter
 Behind the scenes of the “Marketing Letter”
Readings: Handouts
Software: DocCon will be Distributed on CD or via website.
Assignment IV Option 2: Pick a document class and several instances. Design a
document and domain model. Develop a DK Design Document. Implement a SmartTemplate and associated knowledge-base in DocCon. (50%, Due on Week 14)
Week 12 (04/02/2001): Document Knowledge Representation in DocCon
Guest Lecturer: Adam Lally, DDC
Content
 Concepts
 Document Templates and their Parts
 Document Components, Classifications and Roles
 The Domain Model: A Classic OO Modeling Approach
 Linking the Domain Model and the Document Model through CVs
 Gathering Domain Data
03/08/16


Draft
Smart Document Behavior
o Rule Control Logic
o Passive Behavior through Constraints
o Active Behavior through AutoCorrect Rules
o Aggregating Behavior in Guidelines
Simple Template Example – The Easy Path
Readings: Handouts
Week 13 (04/09/2001): Applications in DocCon
Guest Lecturesr: Steinar Flatland & Adam Lally, DDC







Designing and Building the Domain Model
o Objects, Attributes, Classifications and Roles
Document Structure
Document Instance Analysis
o Content Variables
o Reusable Document Components
o Document Library Design and Implementation
Using Template Starter
Manual Carving
Adjusting Range Constraints
Adding Behavior
Week 14 (04/16/2001) Test Authoring and Generation
Lecturer: Selmer
 “High Stakes” Standardized Tests
o SAT, GRE, LSAT, etc.
 AR, LR, RC (item types on e.g. SAT, GRE, LSAT)
 Formal Representation of AR and LR Items
 Implementation of some items in OTTER/OSCAR/HYPERPROOF
 eWriter Project
o Architecture Sketch
o Inductive Logic Programming
 Test Authoring in DocCon
Readings: AIMA (21)
Week 15 (Last Class - 04/23/2001): Final Presentations and Discussions
20 Minute Team Presentations
This class should be held on 4/30th during study week instead (Dave will not be
available)
Study Week (04/30/2001)
Finals Week (05/07/2001)
7
Download