VIRGINIA MONTECINO
Artificial Intelligence and Robotics
© Copyright 1999



DEFINITION OF ARTIFICIAL INTELLIGENCE : an attempt to model aspects of human thought on computers. It is also sometimes defined as trying to solve by computer any problem that a human can solve faster. 

See "The Machine that Changed the World: Part I Giant Brains," available at the Johnson Center Library and/or visit the companion Web site
 

``Machine intelligence will so fundamentally change existence that we   won't be able to conceive of going back to the way things are now.'' 

  ``It will transform life the way language did the Stone Age, mechanical inventions did the Industrial Age, electricity did the 20th century.'' 
  (Doug Lenat - Cyc project creator)

  How do researchers develop intelligent computers?

- create models of human intelligence 
- identify the fundamental components of human intelligence 
- omit non essential details 

symbol processing - an inference engine directs the computer to manipulate facts and rules in a knowledge base.

Some prominent figures in AI who would argue that computers can think:

  • Alan Turing  - -- 1912-1954.  (See Lecture 11) Turing is perhaps best remembered for the concepts of the Turing Test for Artificial Intelligence, the "acid test" of true artificial intelligence, and the Turing Machine, an abstract model for modeling computer operations. He said "a machine has artificial intelligence when there is no discernible difference between the conversation generated by the  machine and that of an intelligent person."  See an explanation of the Turing Test by The PT-Project, Illinois State University . See JAVA applet Turing test
  • Marvin Minsky-  has made many contributions to AI, cognitive psychology, mathematics, computational linguistics, robotics, and optics.  In 1951 he built the SNARC, the first neural network simulator. His other inventions include mechanical hands and other robotic devices. His recent work is to develop machines with the capacity for commonsense reasoning. 
Some prominent figures in AI who would argue that computers can't think:
  • John Searle - Chinese Room theory - Searle, a philosopher,  proposed a thought experiment outlining why computers can't think. He considers the following thought-experiment. Suppose that a person were given a set of purely formal rules for manipulating Chinese symbols.  The rules are a complete set of instructions that might be implemented on a computer designed to engage in grammatically correct conversations in Chinese. The person in the room, however, does not understand Chinese, yet can produce the correct symbols to give the correct response. See explanation of the Chinese Room theory, by The PT-Project, Illinois State University
  • Hubert Dreyfuss - Dreyfuss, a philosopher, says that computer games do nothing more than calculate which moves are the best. Wrote What Computers Still Can't Do : A Critique of Artificial Reason 
Question to ponder - But, if the computer can learn from the moves of the human chess player, for example, is it then capable of learning?  And is learning a sign of true intelligence? 

Games

  • Deep Blue Chess game- ACM sponsored match between World Chess Champion, Gary Kasparov and "Deep Blue" chess program. Deep blue won the first game.
  • Samuel's Checker's Player - Arthur Samuel's Checkers player  experiments (1959 and 1967) were the earliest success stories in machine learning .  His machine evaluated board positions in the game of checkers 
    TD - Gammon - Gerald Tesauro  was able to play his programs in a significant number of games against world-class human players. His TD-Gammon 3.0 appears to be at, or very near, the playing strength of the best human players in the world. TD-Gammon learned to play certain opening positions differently than was the convention among the best human players. Based on TD-Gammon's success and further analysis, the best human players now play these positions as TD-Gammon did. 
Natural Language Processing - Language is what sets us apart from the other members of the animal kingdom.  Challenges of getting computers to understand human language:
     
    Vocabulary, rules of grammar, syntax complex and changeable.  Words can be combined in many ways 
    Meaning nuance.  grammatically correct vs meaningful
    Ambiguity metaphor, sarcasm, irony, cultural context

    Natural Language Processing Machines:

    ELIZA - natural language processing machine. Joseph Weizenbaum invented ELIZA more or less as an intellectual exercise to show that natural language processing could be done. ELIZA is an automated psychoanalysis program based on the psychoanalytic principle of repeating what the patient says and drawing introspection out of the patient without adding content from the analyst.  Weizenbaum believed a computer program shouldn't be used as a substitute for a human interpersonal respect, understanding, and love. He rejected its use on ethical grounds. See the views on ELIZA: The Machine that Changed the World

    SHRDLU - pioneering natural language processing system. Could manipulate blocks based on a set of instructions and was programmed to ask questions for clarification of commands. 

    Common Sense-Lenat, "Cyc" project, University of Texas, Austin - Cyc is a very large, multi-contextual knowledge base and inference engine developed by Cycorp, Inc., at Austin, Texas. The goal of the Cyc project is to construct a foundation of basic "common sense" knowledge base of terms, rules, and relations that will enable a variety of knowledge-intensive products and services. Cyc is intended to provide a "deep" layer of understanding that can be used by other programs to make them more flexible. Cyc has provided the foundation for ground-breaking pilot applications in database browsing and integration, captioned image retrieval, and natural language processing. Demonstration of Cyc's "intelligence":
     

    Cyc ... demonstrated it couldn't be fooled into blaming a terrorist act on a suspect who, it had been previously informed, had died. 

    ``He couldn't have done it,'' Cyc responded ``Dead people can't commit terrorist acts.'' 

    Computer News Daily. From Tod  Ackerman's article  c.1997, 
    Houston Chronicle. Retrieved WWW 4/15/99 (http://computernewsdaily.com/132_051297_100008_32235.html) 

    Case Based Reasoning -  (CBR) views reasoning as a process of remembering one or a small set of concrete instances or cases and basing decisions on comparisons between the new and old situation. The problem can then be solved by using the knowledge based on the earlier situation and adapting it. 

    The steps of CBR generally involve: 
    -- Retrieve the most similar case
    -- Reuse the information in the retrieved case
    --  Revise or adapt the case to solve the current problem 
    -- Retain the solved problem as another case (to be used to help solve another problem).

    A case may not be entirely suitable for a new problem and must be adapted. 

    Examples of CASE systems -  Cyrus,  HYPO, CASEY. 
    See Case Based Reasoning on the Web

    Pattern Recognition - 

    Computer Vision - pattern recognition using cameras for eyes, microphones for ears:

    Optical Character Recognition (OCR)

    Some examples of computer-vision applications:

    -  Satellite photo interpretation
    -  Facial characteristics detection 
    -  Digital searching of videos,  based on content 
    -  Obstacle detection systems for aircraft navigation
    -  Automatic analysis of multi-dimensional radiological images 
    -  Machine vision grading of quality of produce (apples, etc). 
    -  Shape recognition and analysis of machined parts 

    See: 
         GMU Computer Vision and Neural Network Lab
         Carnegie Mellon U. Vision and Autonomous Systems Center
         Computer vision online demos
         Carnegie-Mellon Computer Science Computer Vision Home Page
         Computer Vision Handbook, by Dr. Margaret Fleck

    Virtual Reality (VR) - immerses viewers in virtual worlds even though they are physically present in the real world. Each viewer moves independently and freely throughout this world, allowing the participants to see events from his or her own perspective. Participants enter a 3-D graphical environment and control graphical objects in the environment with body movements.  A glove was the first the input device. Computer vision and robotics technologies can be used to support practical, useful virtual environments. Applications: flight simulators, virtual surgery, virtual museums.  NASA used virtual reality to design the Pathfinder mission to Mars.
     

    Knowledge Engineering/Expert Systems

    What are Expert Systems? 
    Conventional programming languages, such as FORTRAN and C, are designed to manipulate data, such as numbers. Humans, on the other hand, can solve complex problems using very abstract, symbolic approaches,   not well suited for conventional programming languages. Abstract information can be modeled in conventional programming languages, but significant effort is needed to transform the information to a usable format which deals with high levels of abstraction, more closely resembling human logic. The programs which emulate human logic are called expert systems. 

    The expert system tool provides a mechanism, called the inference engine, which automatically matches facts against patterns and determines which rules are applicable. The if portion of a rule applies to the situation (if "such and such" happens or changes"). The then portion of a rule is the set of actions to be executed when the rule is applicable. The inference engine then selects another rule and executes its actions. This process continues until no applicable rules remain. 

    Fuzzy Logic -  Fuzzy logic is a superset of conventional (Boolean) logic that has been extended to handle the concept of partial truth - truth values between "completely true" and "completely false".  Dr. Lotfi Zadeh, Professor Emeritus at Berkeley,  father of "Fuzzy Logic" , introduced the theory in the 1960's. Interview with Zadeh

    Genetic Algorithms - use the principles of  Charles Darwin's natural selection :

    Natural selection - Some traits in a species cause a member of that species to be better suited to its environment than some other traits.  The members of the species with the characteristics that give it the strongest possibility to survive, pass those traits on to offspring.   The species with the stronger characteristics mate and pass the traits on in the process called natural selection.
    Crossover is the term for natural selectionin genetic algorithms.  In crossover, natural selection is accomplished when the genetic algorithm: 
    1. Selects the set of best possible solutions to a problem 
    2.  Selects the best candidates among the set of best possible solutions. 
    3. Selects pairs of solutions and the best parts of each solution to create a new solution - called crossover

    Artificial Life - computer organisms that reproduce and adapt to their environment, mimicking the natural selection process which occurs with biological organisms. 

    Neural Networks - Artificial Intelligence systems that attempt to duplicate the physical functioning of the human brain by using a biological model of intelligence. 

    Three (3) parts of a neural network:

    - input layer corresponding to the 5 human senses: sight, hearing, touch, smell, taste 
    - processing layer (hidden) corresponding to neurons in the brain 
    - output layer corresponding to parts of the body that act on signals from the brain (muscles, etc.) 
     
     

    Input layer ----------------- >  Processing layer ---------- >  Output layer ---------------- >
      (Hidden layer)   
    cameras, microphones, Computers plus printers, screens, robot arms,
    data gathering equipment programs and functions chemical dispensers

    NNs "learn" from examples and exhibit some capability for generalization beyond the specific example. Knowledge is acquired by the network through a learning process. 

    Where can neural network systems help? 

         - where we can't formulate an algorithmic solution. 
         - where we can get lots of examples of the behavior we require. 
         - where we need to pick out the structure from existing data. 

    Real human brains, however, are orders of magnitude more complex than any artificial neural network so far developed. 

    Existing computer "logic is not good at interacting with "noisy" data, and adapting to unexpected or unusual circumstances.

See: Genetics Algorithms Archives
        Hitchhiker's Guide to Evolutionary Computation

Herbert. A. Simon, Allen Newell & J.C. Shaw: In 1957 devised a logic theory machine (first proof by machine) the General Problem Solver (GPS). The method for testing the theory involved developing a computer simulation and then comparing the results of the simulation with human behavior in a given task. 

Simon believes that brain activities, as well as computer processing activities can be explained in terms of information processing. Creativity can be automated, he believes,  by having the computer do selective searches, then recognize cues that index knowledge in given situations.  For example, he says, his his chess playing computer can separate the important moves from the unimportant ones, for a given chess configuration, and even know when the opposing player makes an error.

Herbert A. Simon and AI 
Simon, H.A. Interview. (1994, June). Omni Magazine, 16(9), 70-89. 

Robotics -  Intelligent robots are in use today for innovative uses in entertainment, commerce, industry, and advanced research. - everything from interactive toys to robots that go down oil wells to animated simulations of humans in museum displays. 
 

  • The term robot comes from a play written by K. Capek, RUR , Czech novelist and playwright.
  • Leonardo da Vinci designed a "robot" in the late 15th century.
  • First "arm" that could be programmed to perform tasks developed by George Devol in 1954.
  • Stationary (manufacturing)
  • Mobile (surveillance)
  • Applications too dangerous for humans: industrial activities, planetary rovers, locating sunken ships, exploring active volcanos....
  • Edward Tufte - Visual Explanations : Images and Quantities, Evidence and                     Narrative - book explores how visual evidence influences computer interfaces, design strategies, and how information is transferred and represented, including the arts and science. 
  • Nanotechnology  - As we discussed in the last lecture, nanotechnology is an emerging new field which is attempting to break the barriers between engineered and living systems. K. Eric Drexler, 43, the founding father of nanotechnology, envisioned the idea of using individual atoms and molecules to build  living and mechanical "things" in miniature factories.  His vision is that if scientists can engineer DNA on a molecular, why can't we build machines out of atoms and program them to build more machines? The requirement for low cost creates an interest in the "self replicating manufacturing systems," studied by von Neumann in the 1940's. These "nanorobots, " programmed by miniature computers smaller than the human cell, could go through the bloodstream curing disease, perform surgery, etc.   If this technology comes about the barriers between engineered and living systems may be broken. Researchers at various institutions and organizations, like NASA and Xerox,  are working on this technology. 

  • See: 
    information on Nanotechnology
    nanoManipulator
J. von Neumann -- (1903-1957). A child prodigy in mathematics, authored landmark paper explaining how programs could be stored as data. (Unlike ENIAC, which had to be re-wired to be re-programmed.). Virtually all computers today, from toys to supercomputers costing millions of dollars, are variations on the computer architecture that John von Neumann created on the foundation of the work of Alan Turing's work in the 1940s.  It included three components used by most computers today: a CPU; a slow-to-access storage area, like a hard drive; and secondary fast-access memory (RAM ). The machine stored instructions as binary values (creating the stored program concept) and executed instructions sequentially - the processor fetched instructions one at a time and processed them. The instruction is analyzed, data is processed, the next instruction is analyzed, etc. Today "von Neumann architecture" often refers to the sequential nature of computers based on this model.  Nanotechnology creates new interest in self replicating manufacturing systems studied by von Neumann in the 1940s.

VIRGINIA MONTECINO