Defining Artificial Intelligence

• Introduction to CCAI 221: AI fundamentals

Outline

• The course will define:
  • AI
  • The foundations of AI
  • The history of AI
  • The state of the art in AI
  • Examples of AI Applications

What is intelligence?

• Intelligence is the computational part of the (human) ability to achieve goals in the world.
• Intelligent behavior involves:
  • Ability to act in complex environments
  • Ability to learn from experience
  • Ability to think and reason
  • Ability to perceive relations (in the world)
  • Ability to use tools
  • Varying kinds and degrees of intelligence
  • Varying kinds and degrees of intelligence occur in people, many animals, and even some machines

What is Artificial Intelligence (AI)? - Definitions

• AI definition according to John McCarthy is the science and engineering of making intelligent machines.
• AI definition according to Patrick Henry is the study of ideas that make people intelligent and incorporate those ideas into computers.
• AI definition according to Elaine Rich is the study of making computers do things which, at the moment, people are better.
• AI definition according to Anonymous is getting computers to do tasks that require human intelligence.

What is AI? - Measure success compared to human performance

• Thinking Humanly definition by Haugeland in 1985, "The exciting new effort to make computers think ... machines with minds, in the full and literal sense."
• Thinking Humanly definition by Bellman in 1978, "[The automation of] activities that we associate with human thinking, activities such as decision-making, problem solving, learning . . .”
• Acting Humanly definition by Kurzweil in 1990 is "The art of creating machines that perform functions that require intelligence when performed by people."
• Acting Humanly definition by Rich and Knight in 1991 is "The study of how to make computers do things at which, at the moment, people are better."

What is AI? - Measure of success against an ideal measure – rationality

• Thinking Rationally definition by Charniak and McDermott in 1985 is, "The study of mental faculties through the use of computational models."
• Thinking Rationally definition by Winston in 1992 is "The study of the computations that make it possible to perceive, reason, and act."
• Acting Rationally definition by Poole et al. in 1998 is "Computational Intelligence is the study of the design of intelligent agents."
• Thinking Rationally definition by Nilsson in 1998 is "AI ...is concerned with intelligent behavior in artifacts."

AI Approaches

• Four main approaches have been pursued:
  • Turing Test approach (act like human)
  • Cognitive modelling approach (think like human)
  • The Laws of thought approach (think rationally)
  • The rational agent approach (act rationally)
• The ultimate objective of these AI approaches is to build autonomous intelligent machines

Intelligence Test (Motivation)

• Computers can solve some difficult problems much more quickly than humans.
• Computing the GCD (Greatest Common Denominator) of two numbers can be done more quickly by computers.
• Solving complex integration problems can be done more quickly by computers.
• Computing product of, say, four numbers can be done more quickly by computers.
• Humans can solve some simple problems much more elegantly than computers.
• Navigating in a busy street can be done more elegantly by humans.
• Recognizing the voice and the image of a person can be done more elegantly by humans.
• The first law of AI: Easy problems are hard and hard problems are easy.

1. Acting humanly: The Turing Test approach

• An interrogator communicates with a person and a computer.
• The interrogator can't see the person or computer.
• The computer tries to fool the interrogator into believing that it is a human.
• The person also tries to convince the interrogator that it is human.
• If the computer succeeds in fooling the interrogator, then it passes the intelligence test – intelligent computer

What would a computer need to pass the Turing test?

6 things would need to happen for a computer to pass the turing test.

• Natural language processing to communicate with the interrogator.
• Knowledge representation to store and retrieve information provided before or during interrogation.
• Automated reasoning to use the stored information to answer questions and to draw new conclusions.
• Machine learning to adapt to new circumstances and to detect and extrapolate patterns.
• Computer Vision (for total Turing test) to recognize the interrogator's actions and various presented objects.
• Robotics (for total Turing test) motor control and other senses to manipulate objects and move about.

2. Thinking Humanly: Cognitive science-based approach

• Simulate human-like thinking in machines by:
  • Introspection: trying to catch our thoughts as they go
  • Psychological experiments: Observing a Person in action
  • Brain imaging: Observing the brain in action
• Develop theories and practice to build machines with human-like minds
• Emphasis is on the human reasoning process
• Cognitive science combines computer models from AI and experimental techniques from psychology to construct testable theories of the human mind

3. Thinking Rationally: The "Laws of thought" approach

• Develop systems that think rationally
• The focus is on logical rules and inference mechanisms that are provably correct and guarantee an optimal solution
• For exmaple:
  • Socrates is a man;
  • All men are mortal;
  • Therefore, Socrates is mortal.

4. Acting Rationally: "The rational agent" approach

• Rational behavior is doing the right thing.

• An agent is something that acts.

• Agents are expected to operate autonomously, perceive their environment, persist over a prolonged period, and adapt to change.

• Arational agent acts to achieve the best outcome or when there is uncertainty, the best-expected outcome.

• The focus is on systems that act sufficiently, if not optimally, in all situations.

• The rational-agent approach has two advantages over the other approaches:

  • It is more general than the "low of thoughts" approach, since correct inferences are just one of several possible mechanisms for achieving rationality.
  • It is more suitable for scientific development than approaches based on human behavior/thought.

The Foundations of AI

• Philosophy contributes to AI by:
  • Considering that a mind operates as a physical system and according to logical rules.
  • Establishing the source of knowledge by means of Rationalism: the power of reasoning to understand the world.
  • Viewing the mind as the connection between knowledge and action (thoughts).
• Mathematics contributes to AI by:
  • Using Theories of logic
  • Using Computation, algorithms, (un)decidability, (in)tractability
  • Using Formal representation and proof
  • Using Probability
• Economics contributes to AI by:
  • Providing Utility theory (make choices that leads to preferred outcomes)
  • Providing Decision theory (utility + probability)
• Neuroscience is the study of the nervous system.
  • Neurons are cells that lead to thought, action, and consciousness.
  • Brains and digital computers have different properties
• Psychology contributes to AI by:
  • Making Cognitive psychology view the brain as an information-processing device
  • The Applcation of knowledge-based agent
• Computer engineering contributes to AI by:
  • Providing the ability to build powerful computers that make AI possible
• Linguistic contributes to AI by:
  • Revelealing the meaning and structure of language (knowledge representation, grammar)
  • Providing Computational linguistic and NLP (Natural Language Processing)

History of AI

• The beginning of artificial intelligence (1943–1955)
  • 1943: McCulloch & Pitts created the Boolean circuit model of a brain
  • 1950: Turing wrote "Computing Machinery and Intelligence"
• The birth of artificial intelligence (1956)
  • 1956: At the Dartmouth meeting "Artificial Intelligence" was adopted.
• Early enthusiasm, great expectations (1952–1969)
  • 1950s: Early Al programs, including Samuel's Checkers program, Newell & Simon's Logic Theorist, Gelernter's Geometry Engine
  • 1965: Robinson's complete algorithm for logical reasoning
• A dose of reality (1966-1973)
  • 1966-73: There was a discovery of Al computational complexity, then Neural network research almost disappears
• Knowledge-based systems (1969-1979)
  • 1969-79: Early development of knowledge-based systems.
• Al becomes an industry (1980–present)
  • 1980: Al becomes an industry
  • 1980-88: Expert systems industry booms
• The Return of Neural Networks (1985–present)
  • 1985: Neural networks return to popularity
• Al adopts the scientific method (1987–present)
  • 1987: Al becomes a science
  • 1988: Resurgence of probability, general increase in technical depth: speech technology, handwritten character recognition
• The emergence of intelligent agents (1995-present):
  • 1995: Researchers started to look for a "whole agent" problem again
• There is increasing use of Big data(2001–present)
  • 2003: Human-level Al back on the agenda is developing.
• Deep Learning (20011–present)
  • Speech recognition and visual object recognition are progressing.

State of the art

• Robotic vehicle:
  • In 2005, A robotic car named STANLEY completed autonomously a 132-mile desert track at 22 mph in the DARPA challenge
  • In 2007, vehicles drove on streets with traffic on the Urban challenge
  • In 2018, Waymo test vehicles passed the landmark of 10 million miles driven on a public road without a serious accident, with the human driver taking over control only every 6000 miles.
  • Soon after, the company began offering commercial robotic taxi services
  • Autonomous fixed-wing drones have been providing cross-country blood delivery in Rwanda since 2016
  • A Quadcopter can explore building while constructing 3D maps
• Legged locomotion
  • BigDog, closely resembles an animal and can move in irregular terrain and recover when slipping on an icy puddle (2008)
  • Atlas, a humanoid robot, not only walks on uneven terrain but jumps and does backflips (2016)
• Autonomous planning and scheduling:
  • During the Gulf War, US forces deployed an AI logistics planning and scheduling program that involved up to 50,000 vehicles, cargo, and people at a time (1991)
  • NASA's onboard autonomous planning program controlled the scheduling of operations for a spacecraft (2000)
  • SEXTANT system allows autonomous navigation in deep space beyond the global GPS (2017)
  • Uber and Google provide directions taking into account current and predicted future traffic conditions every day

State of the art - current AI achievements

• Machine translation:
  • Online machine translation systems now enable the reading of documents in over 100 languages
• Speech recognition:
  • Alexa, Siri, Cortana, and Google offer assistants that can answer questions and carry out tasks for the users. (ex. Restaurant reservation).
• Recommendations: -Amazon, Facebook, Netflix, Spotify, YouTube, and Walmart use machine learning to recommend what you might like based on your experience and those of others like you
  • Spam filtering can be considered a form of recommendation
• Game playing:
  • Alpha Go defeated the world chess champion Garry Kasparov in 1997
  • Humans champions have been beaten by AI in the Jeopardy game in 2010
• Image understanding:
  • AI not content with exceeding human accuracy on the challenging ImageNet object recognition task
  • Current systems are far from perfect ex: "a refrigerator filled with lots of food and drinks" turned out to be "no parking sign partially obscured with lots of small stickers “
• Medicine:
  • AI algorithms now equal or exceed expert doctors at diagnosing many conditions, particularly when the diagnosis is based on images (2016)
  • ex: Alzheimer's disease, metastatic cancer, skin diseases.
• Climate science:
  • Machine learning can be used to tackle climate change (2018)
• AIs can currently:
  • Play a decent game of table tennis
  • Play a decent game of Jeopardy
  • Win against humans at Chess
  • Drive safely along the highway
  • Buy a week's worth of groceries on the web
  • Translate spoken Chinese into spoken English in real-time
  • Converse successfully with another person for an hour

Examples of AI applications

Game Playing:

• Deep Blue chess program beat world champion Gary Kasparov Self-driving car:
  • Self driving cars an AI application system Natural language Understanding
• AI Translators – spoken to and prints what one wants in foreign languages
• Natural language understanding (spell checkers, grammar checkers) is an AI application Diagnostic Systems
• WebMD Symptom Checker:A web medical diagnostic system Robotics (SOFIA) is an AI application:
• Robotics are becoming increasingly important in various areas like games, to do tedious jobs among other things

Benefits of AI

• Free humanity from menial repetitive work
• Dramatically increase the production of goods and services
• Help with finding cures for diseases
• Help with finding solutions for climate changes

Risks of AI

• Risks of AΙ:
  • Lethal autonomous weapons
  • Surveillance and persuasion
  • Biased decision making
  • Safety-critical applications
  • Cyber security

Summary

• AI is the science of building intelligent machines.
• Main AI approaches can be classified as Turing test-based approach (acting humanly), Cognitive science-based approach (thinking humanly), Laws of thought-based approach (thinking rationally), and Rational agent-based approach (acting rationally).
• Some disciplines that contributed ideas, viewpoints, and techniques to AI include psychology, mathematics, linguistics...
• The history of Al has cycles of introducing new creative approaches and systematically refining the new ones.
• Some main application areas of Al include game playing, natural language processing, speech recognition, machine vision, robotics, and expert Systems.

Examine core definitions and approaches to AI, including perspectives from John McCarthy, Patrick Henry, and Elaine Rich. Explore capabilities, rational agents, and feasibility assessing algorithms. Understand the 'Laws of thought' approach and its relevance.