Artificial Intelligence Notes

Introduction to AI

Milestones in AI

  • The Rematch: Deep Blue vs. Garry Kasparov

    • Deep Blue defeated Garry Kasparov in chess.

    • Raises the question: Is playing chess a matter of intelligence?

  • Watson

    • IBM's Watson competed on Jeopardy!

    • Required extensive knowledge to answer questions.

  • Deep Learning & GO

    • Utilizes neural networks with input, hidden, and output layers.

  • Google DeepMind: AlphaGo

    • AlphaGo defeated Lee Sedol in a Go challenge match in 2016.

    • Demonstrates advancements in AI.

Success Stories

  • Chatbots: Alice

    • Example interaction with the Alice chatbot.

    • Illustrates the chatbot's ability to respond to various questions.

  • Is Alice Intelligent?

    • No, Alice is not truly intelligent.

    • It operates on approximately 150,000 trivial input-response rules.

    • Utilizes pattern matching, some stored knowledge, and randomness.

    • Lacks genuine reasoning capabilities.

    • Exhibits human-like behavior.

    • Won the Loebner Prize in 2000, 2001, and 2004.

    • Unable to pass the Turing Test.

  • Data Mining

    • Application of machine learning techniques.

    • Solves problems with large datasets that surpass human processing capabilities.

  • More Applications

    • Computer vision.

    • Speech technology (e.g., Amazon Echo Dot).

  • Autonomous Vehicles

    • Self-driving cars and related technologies.

About Intelligence

  • Defining Intelligence

    • When should a program be considered intelligent?

    • When does human creative activity require intelligence?

    • Possible answers: Never? Always?

  • Numeric Computation

    • Does numeric computation require intelligence for humans?

    • What about for computers?

    • When is a program considered "intelligent"?

    • Example: Xcalc3921,56×73,13286783,68Xcalc \quad 3921, 56 \times 73, 13 \quad 286783, 68

    • Could this exist in 1900?

  • Two Aims of AI

    • Long-term aim:

      • Develop systems with human-level intelligence (or better).

      • Not expected within the next 20-30 years.

    • Short-term aim:

      • Develop systems for specific tasks requiring intelligence.

      • Achieved for many tasks: Deep Blue, data mining, computer vision, etc.

  • The Singularity

    • Hypothetical point in time when technological growth becomes uncontrollable and irreversible, resulting in unforeseeable changes to human civilization

  • Reproduction vs. Simulation

    • Focus on reproducing the effect of intelligence rather than simulating human intelligence.

  • Defining AI

    • Many definitions exist.

    • What is important for a system to be intelligent?

A Brief History of AI

  • First use of "Artificial Intelligence"

    • By John McCarthy in 1955.

    • Goal: Develop machines that behave as though they were intelligent.

    • Case: Group of robotic vehicles with varying behaviors.

    • Question: Is this intelligent behavior?

  • Braitenberg Vehicles

    • Demonstrates simple agents with sensor-motor connections exhibit seemingly complex behaviors.

    • Highlights the difficulty in defining intelligence based solely on observed behavior.

  • Definitions of AI Over Time

    • Encyclopedia Britannica:

      • "AI is the ability of digital computers or computer-controlled robots to solve problems normally associated with the higher intellectual processing capabilities of humans…"

      • Is this sufficient? What about computers with large memory for storing and retrieving texts? Or multiplication?

    • Elaine Rich:

      • "Artificial Intelligence is the study of how to make computers do things at which, at the moment, people are better."

      • Long-lasting definition.

      • Includes computations, chess, entering an unknown room, making an inventory, planning actions.

      • Related to the field of autonomous robots.

    • Pitfalls:

      • AI is not only implementation of intelligent processes.

      • Requires understanding of human reasoning and intelligent action.

      • Neuroscience is important to AI.

  • History of AI Milestones

    • 1931: Kurt Gödel shows incompleteness in first-order predicate logic.

    • 1937: Alan Turing points out the limits of intelligent machines with the halting problem.

    • 1943: McCulloch and Pitts model neural networks.

    • 1950: Alan Turing defines machine intelligence with the Turing test and discusses learning machines and genetic algorithms.

    • 1951: Marvin Minsky develops a neural network machine.

    • 1955: Arthur Samuel (IBM) builds a learning checkers program.

    • 1956: McCarthy organizes the Dartmouth College conference, where the name "Artificial Intelligence" was first introduced.

    • 1958: Newell and Simon present the Logic Theorist.

    • 1959: McCarthy invents LISP.

    • 1961: The General Problem Solver (GPS) by Newell and Simon imitates human thought.

    • 1963: McCarthy founds the AI Lab at Stanford University.

    • 1965: Robinson invents the resolution calculus for predicate logic.

    • 1966: Weizenbaum's program Eliza carries out dialog with people in natural language.

    • 1969: Minsky and Papert show limitations of perceptrons.

    • 1972: Alain Colmerauer invents Prolog.

    • 1976: British physician de Dombal develops an expert system for diagnosis of acute abdominal pain.

    • 1981: R1 expert system saves Digital Equipment Corporation money.

    • 1982: Renaissance of neural networks.

    • 1990: Bayesian networks gain popularity.

    • 1992: Tesauros TD-gammon demonstrates reinforcement learning.

    • 1993: Worldwide RoboCup initiative begins.

    • 1995: Vapnik develops support vector machines.

    • 1997: IBM's Deep Blue defeats Garry Kasparov.

    • 2003: First international RoboCup competition in Japan; service robotics becomes a major AI research area.

    • 2006: Autonomous robots improve behavior through learning.

    • 2009: First Google self-driving car drives on the California freeway.

    • 2011: IBM's Watson beats human champions on Jeopardy!

    • 2015: Daimler premiers the first autonomous truck on the Autobahn.

    • 2016: AlphaGo beats European champion 5:0 and Lee Sedol 4:1; deep learning enables good image classification; AI becomes creative.

  • Timeline of AI Areas

    • Numeric, symbolic, probabilistic reasoning, and hybrid systems.

    • Includes events like Gödel's work, Turing's contributions, the Dartmouth conference, and advancements in neural networks and deep learning.

What is A.I.?

  • Emulates human performance via learning, reasoning, understanding complex content, natural dialogs, enhancing human cognition (cognitive computing), or replacing humans in nonroutine tasks.

  • Applications: autonomous vehicles, speech recognition, detecting novel concepts.

Intelligence in AI

  • Ability to interact with the world (speech, vision, motion, manipulation).

  • Ability to model and reason about the world.

  • Ability to learn and adapt.

Four Main Approaches to AI

  • Systems that act like humans.

  • Systems that think like humans.

  • Systems that think rationally.

  • Systems that act rationally.

1. Acting Humanly

  • Creating machines that perform functions requiring intelligence when done by people (Kurzweil, 1990).

  • Tested by the Turing Test: Can a computer's response in a natural language conversation be distinguished from a human's?

2. Thinking Humanly

  • Automating activities associated with human thinking, such as decision-making, problem-solving, learning (Bellman).

  • Goal: Build systems that function internally similar to the human mind.

  • Cognitive science tries to model the human mind based on experimentation.

  • Cognitive modeling approach: act intelligently while internally doing something similar to human mind.

  • Watson

    • Designed by IBM.

    • AI computer system capable of answering questions in natural language.

    • Played Jeopardy against human players and won.

    • Accessed 200 million pages of content, including Wikipedia, but was not connected during the game.

3. Thinking Rationally

  • AI involves the computations that enable perception, reasoning, and action (Winston).

  • Grounded in logic.

  • Knowledge representation and deduction.

  • Deals with uncertain and informal knowledge.

    • Example: "I think I know you."

4. Acting Rationally

  • AI is the branch of computer science concerned with the automation of intelligent behavior (Luger and Stubblefield).

  • Intelligent agent approach.

  • Agent: perceives and acts.

  • Emphasis on behavior.

  • Focus on solving hard problems rather than imitating humans.

What AI is For Me

  • Computers/algorithms making decisions/predictions in real-world problems.

    • Apply, formulate, solve.

Real-World Example: Nurse Rostering Problem (NRP)

  • Assigning shifts to qualified nurses.

  • Considering various constraints.

The Current Hype

  • Gartner Hype Cycle for Emerging Technologies (2017, 2019)

    • Graphs depicting the maturity, adoption, and social application of specific technologies.

  • Gartner Hype Cycle for Artificial Intelligence (2019)

    • Virtual Assistants, Machine Learning, Deep Learning, etc.

  • Relationship between Business Rules/Expert Systems, Machine Learning, Neural Networks, Artificial Intelligence, and Deep Learning.