Artificial Intelligence: History and Agents

• This lecture introduces the history and state of the art of Artificial Intelligence, as well as an overview of Agents.
• The material covered is from Chapter 1, Sections 1.3 and 1.4, and an overview of Agents from Chapter 2.
• PowerPoint slides for each lecture can be found on eFundi.
• The textbook should be studied in detail, since the PowerPoint slides only demarcate the scope of the work.
• An exception to relying on the textbook is Chapter 2 (Agents), for which the PowerPoint slides are enough for introducing terminology.

Lecture Outline

• The lecture covers: The History of Artificial Intelligence, the State of the Art, Intelligent Agents, and the Work Schedule(Quizzes and Assignments).

History of AI (1934 - 1955)

• Warren McCulloch and Walter Pitts proposed the first artificial neuron model.
• Hebbian learning laid the groundwork for neural network learning.
• Any computable function could be modeled by a set of neurons.
• In 1950, Alan Turing wrote "Computing Machinery and Intelligence," introducing the Turing test, reinforcement learning, and machine learning.

AI Inception (1956)

• A Dartmouth meeting was held to study AI.
• An AI program called "Logic Theorist" was developed to prove many theorems.

AI History (1952-1969)

• AI demonstrated capabilities in games, puzzles, and IQ tests, challenging the idea of human uniqueness.
• The General Problem Solver (GPS) was developed, mimicking human problem-solving strategies.
• Arthur Samuel's checkers program demonstrated early machine learning successes.
• The AI programming language LISP was defined.
• In 1965, Robinson discovered the resolution method for logical reasoning.

AI History (1966-1973)

• Computational intractability was discovered in many AI problems.
• Minsky and Papert's "Perceptrons" exposed the limitations of simple neural networks, causing neural network research to decline.

Knowledge-based systems (1969-1979)

• Domain knowledge was used to allow for stronger reasoning.
• From 1980 to present, AI became an industry.
• Digital Equipment Corporation sold R1 "expert system".
• Rule-based expert systems like MYCIN and DENDRAL emerged.
• The industry grew from a few million to billions of dollars in 8 years

Revival of Neural Networks and AI (1986-now)

• Neural networks made a comeback with the back-propagation algorithm.
• AI adopted a scientific method, basing theorems on previous work and rigorous evidence, rather than intuition.
• Speech recognition and Hidden Markov Models (HMM) improved.
• Intelligent agents emerged, powering search engines and recommender systems.
• Large datasets became available, shifting focus to data quality and quantity.

Deep Learning (2011 - present)

• Deep learning revolutionized AI, excelling in speech and image recognition.
• Victory in the 2012 ImageNet contest demonstrated superiority, extending to various domains.
• Advanced hardware and data fueled resurgence, sparking widespread AI optimism.

The State of the Art

• Robotics Vehicle - DARPA Challenge
• Speech Recognition
• Autonomous Planning and Scheduling with Remote Agent for spacecraft and MAPGEN for NASA rovers
• Game Playing
• Spam Fighting
• Logistic Planning with DART for dynamic analysis and the Gulf War 1991 example
• Robotics
• Machine Translation with statistical models
• Generative AI like ChatGPT, Gemini, and image generation
• Advances in AI include work in Robotics, Planning, and Language Processing
• Robotic Vehicles achievements include Waymo reaching 10 million miles without serious accidents
• Legged Locomotion includes development of robots with animal-like mobility like BigDog and humanoid robots like Atlas
• Autonomous Planning and Scheduling includes NASA's Remote Agent for spacecraft operations and EUROPA toolkit for Mars rover operations
• Online systems that provide machine translation can translate documents in over 100 languages, making information globally accessible.
• Microsoft's Speech Recognition system achieved a word error rate of 5.1%, on par with human performance.
• Recommendations algorithms power personalized suggestions across platforms like Amazon, Netflix, and Spotify.

Gaming and Vision

• In Game Playing, AI surpassed human players in Go, Chess, and various video games
• In Image Understanding, AI excels in object recognition and image captioning, though challenges remain in complex interpretations
• In Medicine, AI matches or exceeds expert doctors in diagnosing diseases from images, enhancing diagnostics and treatments
• Climate Science’s Deep learning models uncover detailed information on extreme weather events

Science and AI

• New scientific tools advance and transform science and are exemplified by work in Biochemistry
• “The Most Useful Thing AI has Done" https://www.youtube.com/watch?v=P_fHJIYENDI&t=0
• The video contains the following information:
• 0:00 How to determine protein structures
• 3:50 Why are proteins so complicated?
• 5:34 The CASP Competition and Deep Mind
• 9:08 How does Alphafold work?
• 12:06 3 ways to get better AI
• 14:24 What is a Transformer in AI?
• 17:15 The Structure Module
• 18:35 Alphafold 2 wins the Nobel Prize
• 20:36 Designing New Proteins - RF Diffusion
• 22:58 The Future of AI

Intelligent Agents - Framework for designing AI systems

• The 4 basic types of Agents in order of increasing generality are: Simple reflex agents, Reflex agents with state, Goal-based agents, and Utility-based agents.
• Agents terminology is used in the textbook to demonstrate Al topics.
• A Reflex Agent with State makes decisions based on its current perceptions and its internal state
• The Environment is the external world in which the agent operates
• The agent receives input from the environment sensors
• The agent then decides on an appropriate action and sends output through its effectors
• Unlike simple reflex agents, Reflex Agents maintain an internal state.
• Reflex Agents keep track of events and understand how the world changes.
• Sensors capture information from the environment.
• The agent maintains an internal state, keeping knowledge about the world.

Properties of Task Environments

• Task Environments can have several dimensions, like:
• Fully observable vs. partially observable
• Single agent vs. multiagent ((competitive vs. cooperative multiagent))
• Deterministic vs. stochastic
• Episodic vs. sequential
• Static vs. dynamic
• Discrete vs. continuous
• Known vs. Unknown
• The hardest case is partially observable, multiagent, stochastic, sequential, dynamic, continuous, and unknown.

Examples of task environments and their characteristics

• Crossword puzzle: Fully observable, single agent, deterministic, sequential, static, discrete
• Chess with a clock: Fully observable, multi agent, deterministic, sequential, semi-dynamic, discrete
• Poker: Partially observable, multi agent, stochastic, sequential, static, discrete
• Backgammon: Fully observable, multi agent, stochastic, sequential, static, discrete
• Taxi driving: Partially observable, multi agent, stochastic, sequential, dynamic, continuous
• Medical diagnosis: Partially observable, single agent, stochastic, sequential, dynamic, continuous
• Image analysis: Fully observable, single agent, deterministic, episodic, semi-dynamic, continuous
• Part-picking robot: Partially observable, single agent, stochastic, episodic, dynamic, continuous
• Refinery controller: Partially observable, single agent, stochastic, sequential, dynamic, continuous
• Interactive English tutor: Partially observable, multi agent, stochastic, sequential, dynami

Intelligent Agents Summarized

• An agent perceives and acts in an environment and specifies the action taken in response to any percept sequence.
• The performance measure evaluates the behavior of the agent in an environment.
• A rational agent acts to maximize the expected value of the performance measure, given the percept sequence it has seen.
• Agents can improve their performance through learning.
• A task environment specification includes the performance measure, the external environment, the actuators, and the sensors.
• In designing an agent, specify the task environment as fully as possible.
• Task environments vary in how observable they are, single agent and multiagent dynamics, and whether are stochastic or episodic
• If performance measure is unknown or hard to specify correctly, the design should reflect uncertainty about the true objective.
• variety of basic agent program designs reflecting the kind of information made explicit and used in the decision process.
• Designs program vary in efficiency, compactness, and flexibility, all depending on the nature of the environment.
• Simple reflex agents respond directly to percepts, whereas model-based reflex agents maintain internal state to track aspects of the world that are not evident in the current percept. Goals are achieved with Goal-based agents
• Utility-based agents try to maximize their own expected "happiness."

Work Schedule

• Quiz 1 in class (Ch.1 sections 1.1 and 1.2)
• Assignment 1 Topics: Compare chatbots on CMPG topics and Computational resources for Intelligence
• Assignment 2 (prac) using Teachable machine.
• Quiz 2 in class (Ch.1 sections 1.3, 1.4, 1.5)
• Assignment 3 (prac) using Lobe.ai