T

Comprehensive Notes — Intelligent Machines Seminar

Page 1: Foundations of AI and Human Intelligence

  • What intelligent machines can do:
    • Improve performance and capabilities:
    • Improving accuracy
    • Improving reliability
    • Improving relevance
    • Achieve greater scale and higher levels of productivity in tasks
    • Enable continuous monitoring and auditing
    • Extract more value from data
    • Provide assistance to those in need
  • The quest of intelligent machines is to understand intelligence.
  • Question: What is intelligence?
  • Why study human intelligence when building AI?
    • To create artificial intelligence, we should understand what makes up human intelligence.
  • Core aspects generally considered part of human intelligence:
    • Perceive and understand the outside environment
    • Learn from experience
    • Adapt to new situations
    • Use reasoning to solve new problems
    • Apply learning to different contexts
    • Understand and handle abstract concepts (including emotions)
    • Use that knowledge to manipulate one’s environment
    • Goal-directed future: alive does not automatically mean intelligent
  • Examples and contrasts:
    • Plants and bacteria vs higher-order animals; higher-order animals vs humans
    • Humans express creativity and imagination; respond quickly and successfully deal with complex situations
    • Discern what is important from what is not important
    • Handle ambiguous situations and incomplete information

Page 2: How Intelligence Works, Definitions, and Core Aspects

  • We can guess and be inspired about how learning happens:
    • Trial and error (experimentation and feedback)
    • Observing and discovering patterns
    • Being taught by others (instruction)
  • We can look at the brain’s structure and try to mimic it, but:
    • We don’t really know exactly how intelligence works to handle all the broad things our brains do.
    • All attempts to build intelligent machines are based on guesses about how intelligence works.
  • Definition by Cognilytica (as presented):
    • "Artificial Intelligence is machine behavior and function that exhibits the intelligence and behavior of humans."
  • The Core Aspects of Intelligence (as listed in the material):
    • Perception / Sensing
    • Understanding / Processing lots of inputs
    • Prediction / Understanding possibilities
    • Thinking ahead / Planning
    • Adapting to new environments / Determining outcomes
    • Self-learning / Feedback loop

Page 3: Quiz Preview

  • Questions covered in this page:
    • What is AGI?
    • What is Narrow AI?
    • What are the core aspects of intelligence?
    • What is Cognilytica’s definition of AI?
    • There is not a fully accepted definition of AI.

Page 4: Turing Test, Myths, and Cognitive Technology

  • Turing Test (Alan Turing):
    • A human, a machine, and an interrogator engage in a conversational setting.
    • If the interrogator cannot distinguish the machine from the human, the machine passes the test and is considered to have crossed a critical threshold of intelligence.
    • Modern view: The Turing Test may be overly simplistic for judging true intelligence.
  • Common AI myths (as presented):
    • Cognitive Technology / Cognitive Computing: umbrella term for AI or ML used for specific applications, avoiding stigma of the term AI; often used for narrow applications.
    • AI is all about superintelligent machines.
    • AI is all about building robots.
    • AI is all about building autonomous vehicles.
    • We don’t need AI; analytics and statistics suffice.
    • This is just math — nothing special.
    • Augmented Intelligence: AI plus human collaboration to enhance human capabilities.
  • Additional myths and clarifications:
    • Automation and intelligence are the same thing.
    • You need Google-scale data to do AI well.
    • AI, ML, and DL are all the same thing.
    • Neural networks are dead (not true).
    • You can buy AI from a vendor (not a simple off-the-shelf purchase; needs adaptation).
  • The “best of both worlds” concept:
    • Augmented Intelligence: humans and machines work together to perform information tasks that were previously impossible or too time-consuming.

Page 5: Quiz – Statements About AI Concepts

  • Turing Test: "The Turing Test is the primary method for determining whether a machine or system is ‘intelligent’." Answer: not true (False).
  • Definition for cognitive technology: "Cognitive technology is an alternative to the term artificial intelligence when applied for narrow applications without the potential confusion or stigma applied to the term AI." Answer: (as presented) is part of the slide content; note that the slide marks this as a defined concept.
  • AI is all about application development: Answer: False (the slide marks this as false).

Page 6: True/False Corrections

  • Neural networks are dead — False (they are not).
  • Automation and intelligence are the same — False (they are not the same).
  • Definition of augmented intelligence: Having the human and AI work together to accomplish a task or achieve a goal. (This is the correct definition as per the content.)

Page 7: Symbolic Approaches, Expert Systems, Fuzzy Logic, Why AI Now?

  • Symbolic approaches to machine learning:
    • Uses logic and constructs similar to human reasoning.
    • Not statistically based in the way deep learning and current ML algorithms are.
    • Use human-understandable concepts (names, descriptions, semantic relationships) to reason and deduce meanings.
    • Might handle higher levels of the DIKUW pyramid (Data, Information, Knowledge, Understanding, Wisdom) where statistical approaches struggle (e.g., common sense and machine reasoning).
    • History: Became popular when neural networks and statistical approaches fell out of favor in the early 1970s; later regained interest.
  • Expert systems:
    • AI systems that mimic human expert decision-making using symbolic, logic-based approaches.
    • Components:
    • Knowledge base: stores learned information.
    • Rules base: maps how different inputs lead to outputs.
    • Inference engine: generates results using knowledge base and rules.
    • Often include explanation facilities to show how inputs lead to outputs.
    • Popular in the late 1980s but declined due to complexity and brittleness.
  • Fuzzy Logic:
    • A cognitive technology that provides probabilistic system behavior by allowing multiple possible truth values, enabling non-binary outcomes.
    • Example: washing machines adjusting cycles based on dirtiness (imprecise inputs).
  • Why AI Now?
    • Experience and infrastructure to deal with Big Data (data fuels AI).
    • Rapid evolution of ML algorithms, especially deep learning.
    • Almost limitless, cost-efficient compute power.
    • Mega-tech companies investing heavily in AI.

Page 8: AI Winters and Drivers of Resurgence

  • Quiz question: How many AI winters have occurred previously? Answer: 2.
  • Factors contributing to resurgent interest in AI:
    • Experience and infrastructure to handle big data.
    • Rapid evolution of machine learning algorithms, especially deep learning.
    • Almost limitless, cost-efficient compute power.
    • Heavy investment by mega-tech companies.
  • AI can help break the digital transformation logjam.
  • AI and machine learning require a data-first approach.

Page 9: Comprehensive Study Notes — AI Overview

  • What is Artificial Intelligence (AI)?
    • The science and engineering of making intelligent machines that can perform tasks requiring human-like intelligence.
    • No single universally accepted definition; various experts define it differently:
    • John McCarthy: “The science and engineering of making intelligent machines.”
    • Rodney Brooks: “A collection of practices and pieces people put together.”
    • Max Tegmark: “Intelligence that is not biological.”
    • Cognilytica: AI is machine behavior and function that exhibits the intelligence and behavior of humans.
  • Core Aspects of Intelligence (reaffirmed): Perception, Prediction, Planning.
  • Types of AI:
    • Artificial General Intelligence (AGI): Machines that can perform any intellectual task a human can do. Not yet achieved.
    • Narrow AI (Weak AI): AI systems designed for specific tasks (e.g., image recognition, chatbots, recommendation systems). All current AI is narrow AI.
  • Human Intelligence vs. Machine Intelligence:
    • Human: learning, adapting, creativity, emotional understanding, handling ambiguity.
    • Machines: excel at processing large data, probabilistic thinking, repetitive tasks; lack intuition, emotional IQ, and true creativity.
  • Key Terminology:
    • Turing Test: Test whether a machine exhibits human-like intelligence by whether a human evaluator confuses machine with human responses.
    • Symbolic Systems: AI using logic and rules to represent knowledge.
    • Expert Systems: Mimic human expert decision-making using a knowledge base, rules base, and inference engine; include explanations.
    • Fuzzy Logic: Logic allowing degrees of truth for imprecise situations.
    • Cognitive Technology / Cognitive Computing: AI/ML applied to specific applications.
    • Augmented Intelligence: Humans and machines working together to enhance task performance.
  • AI Myths Debunked (summary):
    • AI is not solely about super-intelligent machines or robots.
    • AI is not only about autonomous vehicles.
    • AI is more than just math or automation.
    • You don’t always need massive data for AI.
    • AI, ML, and DL are related but not the same.
    • Neural networks are not obsolete; they power much of today’s AI.
    • You cannot simply buy AI; it must be built and tailored.
  • History of AI:
    • AI predates modern computers (initial ideas from cybernetics and feedback systems).
    • Notable early work: W. Grey Walter’s tortoises (robotic navigation), Claude Shannon’s Theseus mouse (maze-solving).
    • The term “artificial intelligence” was coined in 1956 at the Dartmouth conference.
    • Two AI winters: mid-1970s–1980s and late 1990s–early 2000s (periods of reduced funding/interest due to unmet expectations).
  • Why AI Now?:
    • Advances in big data, ML algorithms (especially deep learning), and affordable computing power.
    • Heavy investment from tech companies and governments.
    • Demand for always-on, personalized, and relevant services.
  • Practical Applications of AI:
    • Cybersecurity: 24/7 threat monitoring.
    • Medical diagnostics: analyzing X-rays, MRIs, and other data.
    • Autonomous vehicles: self-driving cars.
    • Personalized recommendations: streaming services, online shopping.
    • Continuous monitoring: manufacturing, fraud detection, compliance.

Page 10: Learning Approaches, Transformation, Challenges, and Distinctions

  • Learning approaches in AI:
    • Trial and error: Reinforcement learning (e.g., AI playing games).
    • Observation: Learning from data patterns (e.g., medical image analysis).
    • Instruction: Supervised learning with labeled data.
  • Digital Transformation and AI:
    • Digital transformation is using digital technologies to make organizations more agile and responsive.
    • AI helps overcome the “log jam” of traditional processes by enabling intelligent automation and decision-making.
  • Key Challenges in AI:
    • Understanding and replicating the brain’s complexity.
    • Handling ambiguous, incomplete, or imprecise information.
    • Managing bias in training data and AI systems.
  • Important Distinctions:
    • Automation vs. Intelligence: Automation is repetitive and rule-based; intelligence involves learning and adapting.
    • Machine learning and deep learning are subsets of AI, not synonyms.
  • Seminar examples cited:
    • Washing machines using fuzzy logic.
    • Industrial robots (automation, not intelligence).
    • IBM’s Deep Blue (an expert system).
    • Chatbots, recommendation systems, and autonomous vehicles (narrow AI).
  • Summary:
    • AI is a broad, evolving field focused on creating machines that can perform tasks requiring intelligence.
    • True AGI remains a goal, but current AI excels in narrow, specific domains.
    • Understanding history, terminology, and applications — as well as myths and challenges — provides a foundation for further study and engagement with AI.

Page 11: Additional Synthesis and Practical Takeaways

  • The landscape of AI is built on a spectrum from automation to intelligent systems, including symbolic and statistical approaches.
  • The practical value of AI today lies in augmented intelligence and domain-specific (narrow) AI that can automate, enhance decision-making, and personalize experiences.
  • Foundational ideas to remember for exams:
    • Distinguish between AGI and Narrow AI.
    • Understand the Turing Test and its limitations as a benchmark for intelligence.
    • Recognize symbolic vs. statistical approaches (Symbolic Systems, Expert Systems, Fuzzy Logic).
    • Know why AI is prominent now (data, algorithms, compute power, investment).
    • Be able to describe augmented intelligence and its ethical/practical implications (human-AI collaboration).
    • Be aware of common myths and debunk them with evidence from the material.