Introduction to Artificial Intelligence

Overview of AI

• AI strives not only to understand intelligence but also to build intelligent entities.
• The field of AI was officially coined in 1956.
• AI includes a broad range of subfields, relevant to any intellectual task.

Defining AI

• Understanding how a computer can do the "right thing" is how rationality is key to designing intelligent systems.
• AI definitions vary along two dimensions: thought processes vs. behavior and human performance vs. rationality.
• Four categories of AI definitions include: Thinking Humanly, Thinking Rationally, Acting Humanly, and Acting Rationally.

Approaches to AI

• Combines observations/hypotheses about human behavior with rationalist ideas to form distinct AI development methods

The Turing Test

• The Turing Test was proposed in 1950 to operationally define intelligence.
• A computer would need natural language processing, knowledge representation, automated reasoning, and machine learning to pass the test.
• The total Turing Test includes computer vision and robotics.

Criticisms of the Turing Test

• Mimicking human behavior is insufficient for studying AI's underlying principles.

Cognitive Modeling

• Cognitive science combines computer models and psychological techniques to make precise and testable theories about the human mind.
• Three methods to determine how humans think are: introspection, psychological experiments, and brain imaging.

Thinking Rationally: Laws of Thought

• Aristotle's syllogisms aimed to codify "right thinking" and infer irrefutable processes.
• Logic was the field initiated by his study and patterns for argument structures.
• Challenges to this approach are difficulties in formalizing informal knowledge and solving problems in practice versus in principle.

Acting Rationally: Rational Agent Approach

• Agents must operate autonomously, perceive the environment, persist, adapt, and pursue goals.
• A rational agent acts to achieve the best outcome or the best expected when there is uncertainty.

Rationality vs. Inference

• Correct inference is part of being a rational agent, but rationality includes actions without provable correctness or inference.
• The rational-agent approach allows for scientific development because the standard and agent designs are mathematically defined.

Foundations of AI

• Explores how Al systems underpin the support of logical reasoning systems and knowledge representation to make decisions.
• Philosophy studies knowledge and is the origination point of fundamental questions.
• Key questions include:
• Can formal rules lead to conclusions?
• How does the mind emerge from the brain?
• What is the origin of knowledge?
• How does knowledge translate to action?
• Mathematics formalizes computation, logic, and dealing with uncertainty.
• Subsets include:
• Boolean logic and Frege's first-order logic form the basis of symbolic reasoning.
• Key algorithms provide the concept of step-by-step problem-solving were foundational for programming Al systems.
• Bayes' Rule introduced probabilistic reasoning, essential for Al systems managing uncertainty.
• Economics guides rational agents capable of making optimized decisions in dynamic environments.
• Subsets include:
• Utility theory formalizes how agents optimize choices.
• Game theory examines competitive and cooperative strategies.
• Decision theory combines probability and utility to guide decision-making amid uncertainty.

Psychology and AI

• Cognitive psychology inspired knowledge-based AI systems by suggesting that the brain is an information processor.
• Cognitive psychology influenced decision-making and learning systems.
• Behaviorism lacked the necessary depth related to understanding of thought processes.

Computer Engineering and AI

• Intelligence and an artifact are needed.
• AI has pioneered time-sharing, personal computers, and object-oriented programming.

Control Theory, Cybernetics, and AI

• Control theory studies how systems achieve goals autonomously.
• Subsets include:
• Feedback systems adapt to changing conditions
• Optimal control maximize performance

Linguistics and AI

• Chomsky's critique of behaviorism in "Verbal Behavior", caused a shift
• Computational linguistics/natural language processing became an intersection due to Modern linguistics and Al "born" at the same time.
• Linguistics examines language and cognition.
• Natural language processing enables AI computer systems to understand and mimic language while communicating with human interactions.

History and Evolution of AI

• Media in the early 1900's asked if it ways possible to create an artificial brain.

Gestation of AI (1900-1950)

• 1921: Karel Čapek's play "Rossum's Universal Robots" introduced the word robots to describe 'artificial people'.
• 1929: Professor Makoto Nishimura built the first Japanese robot, Gakutensoku.
• 1949: Edmund Callis Berkley likened the current model of machines to "Giant Brains".

Birth of AI (1950-1956)

• The interest in AI really came when Alan Turing introduced the imitation game and computer machinery
• 1950 Alan Turing proposed a machine intelligence test called the "Imitation Game".
• 1956: John McCarthy's workshop at Dartmouth coined the term "artificial intelligence".

AI Maturation (1957-1979)

• Programming languages emerged.
• Both AI research grew while simultaneously struggling due to little government interest.
• 1958: John McCarthy created LISP (List Processing) a popular language, and among the first languages available.
• 1959: Arthur Samuel coined the term "machine learning".
• 1961: The first industrial robot, Unimate, started working at General Motors.
• 1965: Edward Feigenbaum and Joshua Lederberg created the first "expert system".
• 1966: Joseph Weizenbaum created ELIZA, the first chatbot.
• 1973: Lighthill's report led to less British Science support because milestones from scientists were not being met during early years.
• 1979: James L. Adams created The Standford Cart.
• 1979: The AAAI American Association of Artificial Intelligence was founded.

Early Enthusiasm and Challenges (1980-1987)

• The early 1980s included: AI boom from breakthrough + government funding + Expert System popularity = computers learn independently.
• 1980: The first commercial expert system XCON (expert configurer) assists computer systems.
• 1981: The Japanese government's Fifth Generation Computer focused on translation and human reasoning.
• 1984: The AAAI warned of a funding dip and challenges to research called the "AI Winter".
• 1985: AARON autonomous drawing program was demo'd at the AAAI conference.
• 1986: Ernst Dickmann was the first to demonstrate driverless transport.
• 1987: Alacrity by Alactrious Inc. commercial launch strategy advisory system.

A Dose of Reality: AI Winter (1987-1993)

• An AI Winter came as the AAAI warned of the end of funding and interest because costs were too high for seemingly lower returns on investment.
• 1987: Specialized LISP equipment failed due to competitor accessibility to offer IBM and Apple software.
• 1988: Rollo Carpenter invented Jabberwacky that was meant for conversation with humans

Emergence of Intelligent Agents (1995-Present)

• The 1990s introduced the first AI system that could beat reigning chess champions, followed by the first Roomba and speech recognition.
• 1997: IBM's Deep Blue beat chess champion Gary Kasparov as the first system to beat humans.

AI Progress Continues with Large Datasets (2001-Present)

• Emphasis has shifted from algorithms to available trillions English words + genomic sequencing
• Recent Al suggest machines prioritize what algorithms apply to specific dataset tasks.

Bottleneck Breaches (2012-Present)

• Knowledge bottleneck is being solved by learning tools

• AI tools for virutal assistents became surgant

• "AI Winter" may be yielding to a new Spring say reports

• 2012: Google's Dean and Nig trained a neural to recognize cats.

• 2015: Elon Musk and others banned autonomous weapons.

• 2016 Hanson Robotics created Sophia robot citizen with emotions communication

• Facebook created two AI for negotiation and conversation.

• 2018: Alibaba reading and comprehension AI better results than human scores.

• 2019: Google's AlphaStar reached grand master in StarCraft, and became a AI high performer

• 2020 Open AI testing for GPT for writing tasks.

• 2021 OpenAI developed DALL-E understanding of images to caption visual word.

Goals and Applications of AI in Modern Systems

• Present Day: Concise Answer, Robots, Language, Gaming
• Robotic vehicles detect objects and are traffic signal compliant.
• Tesla’s autopilot AI uses deep learning for both steering and parking.
• Speech recognition automated conversation guiding.
• Game Playing for immersive gaming experiences and real time interaction.
• 1997 IBM's Deep Blue world champion chess.

AI Present Day Logistics and customer support

• Spam Fighting: message classification spamming
• Logistic and Customer Support: provides personalized recommendation such chatbots responding questions and 1991 dynamics between the person.