Introduction to Intelligent Systems
Uncategorized
Introduction to Intelligent Systems
Uncategorized

Introduction to Intelligent Systems

August 5th

Next I provide the most relevant notes and thoughs of a lecture of Intelligent Systems.

The lecture is an introduction to the topic, it defines and explores the types of AI. Additionally, it explored the most important disiplines involved in AI, as well as some addiotional deffinitions.

Definitions

The obvious start point for an introduction to this topic is to define concepts like system and intelligence.

System:

  • “Collection of connected elements organized for a common purpose.” A lovely definition, btw.

Intelligence:

  • It is more difficult to define.
  • We think it is the combination of
    • Abilities: learning capability, reasoning, etc.
    • Behaviors: planning, application of knowledge, etc.

Intelligent systems:

  • “System based on artificial intelligence (AI) techniques to perform more accurate and effective operations for solving problems that require intelligence.”
  • Technically, an intelligent system includes any system that has the capacity to behave intelligently. Systems that occur naturally like our own brain, some complex populations should be included. But, as a basic computer sciences course, it will focus on studying the non-natural systems.
  • Then, the definition is kind of redundant but a very approximate definition.

 

Thinking and Acting Humanly Square divides AI into four cathegories depending on the objectives to achieve.
Thinking and Acting Humanly Square

Artificial Intelligence:

  • Currently, we do not have a completely intelligent machine, but there are relevant advances, if you do not believe this try to chat with an AI chatbot like GPT.
  • A classical classification of the AI is the Thinking and Acting Humanly Square, which divides the AI in four classes based on two classes

Thinking

  • Thinking Humanly:
    • Includes cognitive models that aim to replicate how the human cognition works.
    • Utilizes the next inspiration:
      • “Through introspection: trying to catch our own thoughts as they go by.”
      • “Through psychological experiments: observing a person in action.”
      • “Through brain imaging: observing the brain in action.”
    • Example: ACT-R(Adaptive Control of Thought-Rational), which aims to replicate human cognitive processes, like memory and reasoning. 
  • Thinking Rationally:
    • Acts in a rational manner according to the rules of logic and decision-making.
      • Modus ponens and modus tollens are examples.
    • Example: An expert systemlike MYCIN, used in medicine to diagnose bacterial infections and recommend treatments based on logical rules and medical knowledge.

Acting: It requires agents (see agents below)

  • Humanly:
    • AI that behaves in a way that would be considered human, although it does not necessarily use the same mental processes.
    • The main components to achieve this AI are listed in the Turing test section below.
    • Example: Asistentes Virtuales: Siri, Alexay Google Assistant.
  • Rationally:
    • AI that acts effectively to achieve a given goal, following a logical strategy that may or may not imitate human behavior.
    • Notes
      • “Making correct inferences is sometimes part of being a rational agent. However, correct inference is not all of rationality. In some situations, there may not be correct thing to do, but something must still be done. Like in ethical dilemmas or application of instincts.”
      • “Achieving perfect rationality is not feasible in complex environments since computational demands are just too high. In most cases, there is not enough time to do all the computations one might like.”
    • Example: Adaptive Control Systems like Aircraft controllers
  • From the previous square we notice that depending on your definition of intelligence, you might optimize different functions to develop an intelligent system.

Acting humanly: Turing test

  • It is a classical algorithm in computer sciences, developed by our god Alan Turing in 1950.
  • It was designed to provide a satisfactory operational definition of intelligence based on the class acting humanly.
  • Based on this, we can infer a machine needs the next to be intelligent:
    • Natural language processing to successfully communicate with the human.
    • Knowledge representation to store what it knows.
    • 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.
  • A computer passes the test if a human interrogator cannot tell whether the written responses come from a person or a computer after making some written questions.
  • There is an issue, for example, if the interrogator asks whether the entity to which they are talking is a computer, the computer must lie (give an incorrect answer) to pass the test.
  • It is a classical test, but outdated. Currently, we are usually more interested in developing rational machines.

Intelligent agents: All computer programs do something, but computer agents are expected to do more:

  • Operate autonomously.
  • Perceive their environment.
  • Persist over a prolonged time period.
  • Adapt to change.
  • Create and pursue goals.

AI is hard because it is difficult to state directives to a computer in a manner that it considers all the complexity of intelligence.

More concepts:

  • Human-level artificial intelligence (HLAI). A machine able to learn and do anything a human can do.
  • Artificial super intelligence (ASI). An artificial intelligence that surpasses human ability.
  • Singularity: Hypothesized point in which an artificial intelligence exceeds that of the humans andcomputers start using their own intelligence to improve themselves, over and over. From this point on, it would be impossible for humans to regain control since machines would be the most intelligent thing on the planet.

Foundations of AI

·      Philosophy: It explores the limitations of AI, the basic functioning of intelligence, the ethical considerations.

·      Mathematics: It provides the tools to develop AI. 

o   Algorithm 

o   Define Incomputable functions and Intractable problems. 

o   Probability and logic.

·      Economics:

o   The combination of probability theory and utility theory has been a keystone to increase the richness of a few, but it is also a keystone in rational thinking.

o   “Models based on making decisions that are “good enough” rather than laboriously calculating an optimal decision usually gives a better description of actual human behavior.”

·      Neuroscience: If humans are intelligent due to their brains, the natural approach is to get inspiration on the process that take place in the brains.

Psychology: Psychologists adopted the idea that humans and animals can be considered information-processing machines. 

·      Computer Engineering: We hardware to execute our complicated programs in time.

·      Control theory and cybernetics: 

o   Control theory deals with designing devices that act optimally on the basis of feedback from the environment. 

o   Modern control theory, especially the branch known as stochastic optimal control, aims at designing systems that maximize an objective function over time. This roughly matches the idea of behaving rationally. 

·      Linguistics

o   It seems human language is more complex than it initially seemed. We need to include context, matter, and structure.