wgu D429 Intro to AI (MAIN SET)

Section 1 -plain, section 2 -underlined, section 3 -underlined and bolded

intelligence

the ability to understand, apply knowledge, and solve complex problems. The capability for the machines to exhibit behaviors that would be considered performed by humans.

rationality

the ability to make sound decisions based on the information provided. A rational agent selects actions that maximize its performance measure given its precepts sequence and built-in knowledge.

artificial intelligence

machine driven, human-like intelligence and problem solving skills. AI helps to create algorithms. It also helps machines learn, reason, and make decisions.

Turing test

A thought experiment test proposed by Alan Turing in 1950 to determine a machines/ computer’s ability to respond as a human.

If a person can not tell whether they are communicating with a person of machine, the machine is said to pass the test.

NLP (national language processing)

A computers ability to communicate in human language.

enables AI to interpret, comprehend, and generate human language.

NLP powers chatbots, virtual assistants, language translation, and text analysis.

knowledge representation

that ability to store information (from what it knows and hears).

The method of encoding information about the world in a form that AI systems can use to solve complex tasks through reasoning and inference.

automated reasoning

The computer’s ability to answer questions and draw new conclusions.

The ability of an AI system to use stored knowledge and logical rules to draw conclusions, answer questions, and derive new information automatically.

machine learning (ML)

Adapt to new circumstances and detect and extrapolate/(discover) patterns.

The field of study that gives computers the ability to learn from data without being explicitly programmed.

An subset of AI where algorithms learn from data and real life observations to identify patterns and make predictions, improving through experience without being explicitly programmed. Rooted in the empiricist tradition of learning from direct experience.

Total Turing test

An extension of the Turing test that also requires the machine to demonstrate computer vision and robotics capabilities, testing physical interaction as well as conversational intelligence.

Evaluates a machine’s intelligence by assessing its ability to not only communicate in natural language but also perceive and interact with the physical world and become indistinguishable from a human.

computer vision

a field of computer science that focuses on enabling computers to find and understand objects and people in images and videos, including tasks like image classification, object recognition and scene understanding.

Example plant identification app: the app uses the camera to identify the plant.

robotics

the ability to manipulate and move objects.

A branch of AI focused on building machines (robotics) that can interact with the physical world, perform complex tasks, collaborate with humans and adapt to different environments.

dualism

points to the existence of two distinct and independent realities; the body and the mind connect together.

Unlike AI (which builds smart machines), dualism examines the natures of consciousness and mind-body relationship.

empiricism (em-pier-ruh-cisim)

emphasizes knowledge acquisition through sensory experiences and observations.

A philosophical approach that values direct experience and observation as the primary sources of knowledge. In AI, empiricism underpins machines learning, where systems learn from data and real-world observations.

induction

involves drawing general conclusions from specific observations. A methods of reasoning that moves from specific observations to broader generalizations. For example, observing that every car driven has battery may lead to general conclusions that all cars have batteries.

Example: I can see that a car can not move without wheels, so therefore I can conclude that all cars must use wheels to move.

ai observation sentence

a statement generated by an ai system based on its analysis of data, reflecting the ais interpretation of observed patterns or trends

Describes how an AI system perceives and understands information from its environment or data inputs, enabling it to recognize pattern, gather information, and make predictions or decisions.

legal positivism (think of laws)

a theory of laws where rules are created by humans and valid based on legitimate authority.

AI must be used legally and follow the laws.

The validity of a law is based on its source (lawmaker or legal authority) rather then its moral content. AI can assist in analyzing legal texts, but final legal interpretation still requires human judgement.

Neural Networks

computational models inspired by the human brain, consisting of interconnected layers of nodes that process and learn from data to recognize patterns and make predictions.

think of deep learning… to find patterns and identify different things like NLP, etc.

Computing system inspired by biological neural network in the brain. They consist of layers of interconnected nodes that process information and learn patterns from data. First proposed McCulloch and Pitts in the 1940s.

ethical artificial intelligence

considers that the study and use of AI technologies should follow ethical rules to ensure fairness, transparency, and accountability (FAT).

think of FAT to remember Fairness, Accountability & Transparency

Ethics:

• transparency, accuracy, fairness & bias, privacy and data protection, social consequences, responsibility & accountability

The practice of developing and deploying AI system that are transparent, fair, accountable, and designed to minimize bias and social harm.

Deep Learning

A subset of machine learning using deep neural networks with multiple layers that achieves remarkable performance in tasks such as image recognition, speech recognition, NLP, and autonomous driving.

AI Winter

A period (notably in the 1970s-1980s) when AI research faced setbacks, progress stalled below exceptions, and funding declined significantly. The resurgence came in the 1990s with machines learning and big data.

Agents —- Simple Reflex Agents

Simple Reflex Agents - make choices based on what they see NOW.

Remember GUMSL to help you remember the agents.

Agent —- Model Based Reflex Agent

Model-Based Reflex Agents - use the current environment to make choices.

Remember GUMSL to help you remember the agents.

Agent —- Goal Based Reflex Agents

Goal-based Reflex Agents - work towards achieving set goals via planning.

Remember GUMSL to help you remember the agents.

Agents —- Utility-Based Agents

Utility-Based Agents - Think about goals AND (optimized) preferred outcomes. Look for the best-expected result.

Remember GUMSL to help you remember the agents.

Agents —- Learning Agents

Learning Agents - Learn from experience and get better with time.

Remember GUMSL to help you remember the agents.

agent

Anything that can be perceived or viewed through the environment through its sensors.

Example: using a camera to acknowledge a human by speaking, or slamming on the brakes to avoid hitting a car (also an example of an agent function).;

Agents can:

• sense, think, and decide; learn, plan, and understand language; store knowledge; communicate; and work independently.

An entity (human, robot, or software program) that perceives its environment through sensors and acts upon its through actuators/(movement). They key property is the ability to make choices on perception and goals.

environment

The surroundings in which an agent operates and performs tasks. Environments can vary in characteristics such as observability, determinism, and dynamism.

the part that affects what an agent perceives.

actuators

a device that causes motion/movement.

Devices that convert energy into motion, allowing an agent to take physical actions in its environment (motors in a robot, display outputs for a software agent).

precept

the content of an agents sensors.

The inputs an agent receives from its environment at any given moment through its sensors.

precept sequence

the history of everything an agent has perceived

example: its a timeline of the what that specific agents has seen; ring timeline.

The complete history of all precepts (inputs) an agent has received over time. An agent’s actions can depend on its entire precept sequence.

agent function

maps the precept sequence to actions.

A mathematical mapping from a precept sequence to an action. It describes what actions an agent takes for every possible sequence of precepts.

slamming on brakes to avoid an object.

Type of Agents: (FROM THE SLIDESHOW)

Simple Reflex Agents - make choices based on what they see NOW.

Model-Based Reflex Agents - use the current environment to make choices.

Goal-based Reflex Agents - work towards achieving set goals via planning.

Utility-Based Agents - Think about goals AND (optimized) preferred outcomes. Look for the best-expected result.

Learning Agents - Learn from experience and get better with time.

Remember GUMSL to help you remember the agents.

Common Agent Types Table 1.2

Table 1.2 Common Agent Types

Agent Type	Description
Simple Reflex Agents	These agents make choices based only on what they see now and set rules. They do not think about what happened before or what might happen later.
Model-Based Reflex Agents	These agents keep a picture of the environment inside and use it to make choices. They look at what they see now and what they saw before to decide what to do.
Goal-Based Agents	These agents have goals they want to achieve and work to reach them. They use planning and thinking to pick actions that bring them closer to their goals.
Utility-Based Agents	These agents think about their goals and what different outcomes they prefer. They make choices by looking for the best-expected result. optimized results.
Learning Agents	These agents learn from their experiences and get better over time. They can change what they do based on feedback and learn from past successes and mistakes.

Agent Type Example from textbook P.E.A.S

An example of Taxi driver AI system on page 42, Figure 2.4

task environment

specific setting or context in which an AI agent operates and performs its designated task

The complete specification of the problem an agent is designed to solve, including the performance measure, environment, actuators, and sensors (describes by PEAS).

P.E.A.S.

performance, environment, actuators, sensors

An acronym for performance measure, Environment, Actuator, sensors — used to fully specify a task environment for an AI agent.

software agent (Softbot, software robot)

a computer program that relates/acts for another user or a program in a relationship with an agent.

Commands within software and interpret the feedback.

An AI agent that operates in a software environment rather than physical one, interacting through code execution, data queries, or digital interfaces.

fully observable

occurs when sensors detect all aspects that are relevant to the choice of an action.

An environment where the agent’s sensors give it complete access to the full state of the environment at each point in time.

Playing an video game tutorial

partially observable

occurs when parts of a state are missing from sensor data

An environment where the agent cannot see the complete state due to noisy, inaccurate, or limited sensors.

example: a camera that use sensors at night, playing an solitaire game (unable to see all the cards)

unobservable

when an agent has no sensors.

An environment where the agent has no sensors at all and must rely entirely on prior knowledge to act.

single agent

one agent performing a task and operates independently.

multiagent

when two or more agents/(multiple agents) perform a task together cooperatively or competitively.

competitive

maximizes agent performance measures by avoiding the pitfalls of predictability.

where agents’ goals conflict — one agent’s success may come at another’s expense.

it competes with itself and learns from those experiences to become better.

cooperative

allows single space occupancy.

stays to itself; think of networks like a firewall everything stays in its own lane.

deterministic

when the state of an environment is completely decided by the current state and action executed by an agent.

the next state is completely determined by the current state and the agent’s action, with no randomness involved.

everything is right in front of you and its already decided.

nondeterministic

when certain behaviors are unpredictable or unexpected

An environment where the outcome of an action is not fully predictable, through not necessarily due to probability — outcomes may be uncertain without explicit probabilities.

think of weather; its a chance of snow but it is still unpredictable and it relates to stochastic.

stochastic (stuh-ka-astic)

when the model of an environment explicitly deals with probabilities and introducing quantified randomness into state transitions.

relates to nondeterministic and relates to probability; there is an 9% chance of snow.

episodic

a process where agents do not think ahead but base decisions on current issues.

An environment where the agent’s experience is divided into independent episodes; each action depends only on the current episode, not on previous ones.

Examples: episode connects with the show but it is still independent b/c you can watch that episode without watching the whole show.

An agent notices an error in the assembly line and the agent has to make an decision to fix regardless of past or future possible defectives.

sequential

a concept where decisions affect future decisions.

An environment where the current decision can affect all future decisions; requiring the agent to plan ahead.

static

an environment that remains unchanged while an agent is deliberating or acting.

An environment that does not change while the agent is deliberating about what action to take.

dynamic

environments that consistently questions agents to make decisions and do nothing until agents make decisions.

An environment that can change while the agent is deciding what to do, requiring the agent to act quickly or account for changes.

constantly changing based on the environment until an agent makes a decision.

semi-dynamic

an environment that does not change with time despite an agents changing performance score

keeps going and does not change with time.

discrete

a system that models problems that are too large to be continuous.

can not be counted; its too large.

environment with a finite number of distinct, clearly defined states, precepts, and actions.

digital technology

continuous

an environment where performed actions cannot be numbered.

potentially infinite range of states, precepts, or actions (real-valued inputs or time)

waveform/analog (antennas) technology

known

an environment where the outcomes (probabilities of outcomes) of all actions is provided.

unknown

a situation where the AI agent has little or prior knowledge about the environment it is operating in.

• where the agents does not know the outcomes of its actions and must learn through explorations.

environment class

a category or grouping in programming that encapsulates environmental variables and settings, allowing for the management and configuration of application behavior based on the specific context in which it operates.

It is influenced by its environment and context it operates in.

think of business ordering inventory. the amount of inventory depends on the customers and the size of the store (environment).

smart agent

An agent that can sense its surroundings, reason about them, and act to reach specific goals, making choices based on perceptions, knowledge, and objectives.

a software program capable of performing tasks autonomously, learning from its environment, and making decisions or taking actions to achieve specific goals

AI agent

a software entity that perceives its environment, processes information, and takes action autonomously to achieve a specific objective.

learning agents

An Agent that handles making improvements.

• Works in new environments and improves its performance over time through experience. Composed of a performance element, critic, learning element, and problem generator.

For example, a self-driving car that looks at the traffic patterns and learns what to do (in other traffic patterns) and makes a decision in those moments.

performance element

selects external actions

The component of a learning agent responsible for selecting actions based on the agent’s precepts and knowledge. It is what the agent uses to decide what to do.

Common Tasks of AI Agents Table 1.1

Table 1.1 Common Tasks of AI agents

Task	Description
Sensing	They understand and gather information about their surroundings using sensors or input channels. They can figure out data from cameras, microphones, or other sensors to know what is happening around them.
Thinking & Deciding	They analyze and think about the data they sense. They can use logic, probability theory, or machine learning to make smart choices. This means thinking about different options and choosing the best action to reach their goals.
Learning	They learn from their experiences to get better. They can learn from training data, feedback, or their surroundings. This learning helps them adapt and make better choices in similar situations later.
Planning	They make plans to achieve specific goals. They can use algorithms to find the best sequence of actions that leads to the outcome they want. This is useful for complex tasks with many steps.
Understand & Language	They understand and use human language. They can understand and process human language, which helps them talk to people through speech or text. This is important for virtual assistants, chatbots, or language translation systems.
Storing Knowledge	They keep knowledge in a structured way. They can store facts or rules that help them think and make choices. This knowledge can help them to solve problems or answer questions.
Communicating	They talk and interact with people or other agents. They can understand what people ask, respond, and have conversations. This helps them to assist, give advice, or offer help in different areas.
Working Independently	They work on their own without human help. They can watch their surroundings, make choices, and act without human control. This independence helps them do tasks well on their own.

critic

decides performance element modifications.

The components of a learning agent that evaluates how well the agent is doing by providing feedback on its actions relative to a fixed performance standard.

It’s a measure of accuracy for a model.

problem generator

suggests actions that lead to new and informative experiences.

The components of a learning agent that suggests exploratory actions to discover new and potentially better strategies, rather than always exploiting know approaches.

example: regenerated AI; generates new data.

reward

provides direct feedback on the quality of agent behavior

positive feedback given to an agent when it performs a desired action, reinforcing that behavior.

penalty

provides critical/negative feedback on agent behavior.

Negative feedback given to an agent when it performs an undesired action, discouraging that behavior.

training data

a dataset used to teach a machine learning model, allowing it to learn patterns and make predictions or decisions.

Used to teach a machine learning model by providing input-output examples for the algorithm to learn from.

operational data (AKA test data)

Information generated during the regular functioning of a system. It is used to monitor, manage, and improve ongoing operations.

Real-world data that the trained model processes during deployment to generate predictions or decisions.

Human level AI

commonly referred to as artificial general intelligence (AGI); it’s AI’s ability to understand, learn, and perform humanlike tasks.

AI that matches human cognitive abilities across a wide range of tasks. And it remains a theoretical goal not yet achieved.

Table 1.3 AI Limitations 

Aspect	Description
Lack of Common Sense	AI systems lack the common sense that people do. can have a hard time understanding and reacting to unclear or new situations.
Limited Creativity and Originality	AI is good at tasks that follow specific rules and patterns, but it struggles with tasks that need creativity and imagination. AI can create things like art, music, or text, its work often lacks the depth and uniqueness of human creativity offers and can feel repetitive.
Ethical Decision-Making	AI systems follow set rules and do not have the ability to make moral choices. decisions are based on data patterns, and can lead to biased outcomes and/or decisions that are not considered; right or wrong. To be use ethically, you need human oversight.
Bias and Fairness	Bias in AI happens when the technology treats some people unfairly and/or inaccurately because of their background or personal traits. Fairness in AI means it is needs to designed to treat everyone equally and without discrimination so it works for all groups of people.
Lack of Emotional Intelligence	AI does not have feelings or the ability to understand human emotions. it can recognize emotions through facial expressions or text, it does not understand their emotional meaning. AI can not to have meaningful emotional conversations with people.
Data Dependency	AI relies heavily on high-quality and diverse data to learn and make decisions. If the data it uses is biased or incomplete, the results can also be biased or wrong. Collecting and organizing a lot of data is expensive/timely so, the quality data is highly important for functionality of AI.
Limited Transfer Learning	AI models are usually trained for specific tasks. These models find it hard to apply what they have learned to new tasks. it hard for AI to adapt quickly to new situations or handle multiple tasks at the same time.
Computing Power and Energy Consumption	AI and deep learning models often need a lot of computing power and energy to train and work. Hard to use AI in places with limited resources and it is a environmental issue with energy consumption.

Artificial Superintelligence (ASI)

refers to the future of AI, which could “potentially” surpass all levels of human intelligence.

A “hypothetical” AI that surpasses human intelligence across all domains, including creativity, problem-solving, and social intelligence.

Gorilla Problem

refers to apprehension about creating superintelligent machines.

The concern that creating a superintelligent AI could render humans unable to control or compete with it, similar to how gorillas cannot control the more intelligent humans who affect their fate.

From the course textbook, this is why it is called the gorilla problem…

“We might call this the gorilla problem: about seven million years ago, a now-extinct primate evolved, with one branch leading to gorillas and one to humans. Today, the gorillas are not too happy about the human branch; they have essentially no control over their future…”

King Midas problem

describes a hypothetical scenario or challenge that arises when designing intelligent systems that optimize for a specific objective without considering the broader consequences or trade-offs

The risk that an AI system may achieve its specified objective literally but in unintended harmful ways, analogous to King Midas's wish that everything he touched turn to gold.

From the course textbook, this is why it is called the King Midas problem,

“Many cultures have myths of humans who ask gods, genies, magicians, or devils for something. Invariably, in these stories, they get what they literally ask for and then regret it. The third wish, if there is one, is to undo the first two. We will call this the King Midas problem: Midas, a legendary King in Greek mythology, asked that everything he touched should turn to gold, but then regretted it after touching his food, drink, and family members.“

assistance game

occurs when a machine tries to achieve a human objective.

A framework where the AI's goal is to maximize human satisfaction rather than a fixed objective, requiring the AI to learn human preferences through interaction.

inverse reinforcement learning

Occurs when machines learn about human preferences by observing human choices.

A technique where an AI system learns what humans value by observing their behavior, rather than being explicitly told the reward function.

AI ethics

the examination of moral issues related to the development and use of artificial intelligence, including fairness, transparency, and societal impact.

The study and practice of ensuring AI systems are developed and used responsibly, addressing issues like bias, fairness, transparency, accountability, and social impact.

ChatGPT ensures that its user’s privacy and data security are maintained while using the tool.

False

• Privacy and data security are crucial ethical considerations when using ChatGPT. The tool requires access to large amounts of data, including potentially personal and sensitive information, which must be collected, stored, and used in a secure and ethical manner.

ChatGPT has built-in safeguards to ensure that its responses are fair and unbiased.

False

• Although ChatGPT has automatic guardrails, their effectiveness is not guaranteed. Additional measures, such as carefully selecting and diversifying training data, are needed to ensure its responses are fair and unbiased.

negative side effects

The harmful effects that technologies have on the world.

Unintended harmful consequences of an AI system's actions that were not anticipated during design or training.

surveillance cameras

recording devices that capture movement.

cybersecurity

The practice of protecting computer systems, networks, and digital data from unauthorized access, attacks, damage, or theft.

The practice of protecting computer systems, networks, and data from digital attacks. AI can both enhance cybersecurity defenses and introduce new security vulnerabilities.

de-identification

The process of removing identifying information.

The process of removing personally identifiable information from data to protect individual privacy while still allowing the data to be used for analysis.

generalizing fields

a form of minimizing risks by minimizing information.

This refers to protecting individual data online, such as asking for a general age range (e.g., 20-40) instead of an exact age in surveys.

k-anonymity

indistinguishable database; databases with each data different from the k-1 data.

A privacy-preserving technique ensuring that each record in a dataset is indistinguishable from at least k-1 other records, making it harder to identify specific individuals.

passage from textbook; “…if every record in the database is indistinguishable from at least k-1 other records. If there are records that are more unique than this, they would have to be further generalized...”

aggregate querying

Processes the data from multiple indexed entities to return a single summary value.

A data privacy technique that returns summarized or aggregated results rather than individual records, protecting personal information in query responses.

Think of SQL dataset tables.

CREATE TABLE Employee (

id int PRIMARY ID , Name varchar(100),Salary int, int EM_ID),

(Sally, 225,000,001) ;

differential privacy

a mathematical framework that ensures the privacy of individuals in a dataset by providing guarantees that the inclusion or exclusion of any single individual's data does not significantly affect the outcome of any analysis, thereby protecting personal information from being revealed.

A mathematical framework that adds controlled noise to data or query results, providing a formal guarantee that individual records cannot be identified.

Example: joining a new person in the dataset will not affect the outcome, and their personal data will be protected.

federated learning

Often referred to as collaborative learning, this is a decentralized approach to training machine learning models that does not require an exchange of data from client devices, thereby ensuring privacy.

A machine learning approach where models are trained across multiple decentralized devices or servers holding local data, without exchanging the raw data itself.

Machines work among themselves and do not exchange data from client devices, keeping the data private and maintained.

secure aggregation

A privacy-preserving technique that allows multiple parties to collaboratively compute an aggregate value of their individual data without revealing their raw data to each other.

A cryptographic protocol that allows multiple parties to compute a joint result (like a model update) without revealing their individual inputs to each other.

The AI Machines work together but DO NOT share their individual private data with each other.

Which characteristic is crucial for autonomous vehicles to navigate and make real-time decisions?

Robotics

• involves the design and operation of robots, enabling autonomous vehicles to perceive their environment and make real-time decisions.

Which AI functionality involves analyzing historical data to predict future trends?

Machine Learning

• analyzes historical data to identify patterns and predict future trends, providing insights for decision-making.

Which AI functionality assists in diagnosing diseases by analyzing patient data and suggesting possible conditions?

Machine learning

• uses algorithms to identify patterns in complex data sets, making it a powerful tool for applications like medical diagnosis, where it analyzes patient data to suggest possible conditions, among many other uses.

Ontological Engineering

The study of creating and designing ontologies — formal representations of knowledge that define concepts, relationships, and properties within a specific domain.

Ontology

A formal representation of knowledge that defines the concepts, relationships, and properties within a particular domain. It provides a shared vocabulary and structure for organizing information.

Upper Ontology

A high-level, general ontology that provides broad categories and relationships applicable across multiple domains, serving as a foundation for more specific ontologies.

Concepts

The fundamental ideas or categories represented in an ontology that describe the types of things that exist in a domain.

Classes

Categories in an ontology that represent groups of similar concepts. Classes describe qualities and connections and provide a way to display how different concepts are related.

Instance

A specific, real-world example or object that belongs to a particular class in an ontology.

Knowledge Representation Language

A formal system used to describe ontologies, showing the structure, constraints, and meanings of concepts, relationships, and properties within a domain.

Ontology Alignment

The process of connecting and combining different ontologies so that concepts, relationships, and properties are consistent and interoperable across systems and domains.