Ai, Machine Learning and Models

What was Symbolic AI (1950s–1980s)?

A rules-based approach where researchers programmed computers with step-by-step instructions and logical rules to perform tasks.

What is Machine Learning (ML)?

A subfield of AI focused on algorithms that allow computers to learn and improve at tasks automatically through data and experience rather than explicit instructions.

What is a Machine Learning Model?

A function that takes an input, processes it, and produces an output—with the ability to learn and adjust how it processes inputs.

What is Tom Mitchell’s famous definition of Machine Learning?

A computer program learns from Experience (E) regarding Task (T) and Performance (P) if its performance at T (measured by P) improves with E. Simplified: Performs a task -> gets objective feedback -> uses feedback to get better

In Machine Learning, what are Features?

The input data/attributes given to a model to help it perform a task (e.g., length, width, or type of bread)

In Machine Learning, what are Parameters?

Adjustable internal numbers (like weights) that the model modifies during training to change how it processes inputs into outputs.

What is a Label in Machine Learning?

The correct answer or ground truth for a given set of inputs used as feedback during training

What is the role of a Performance Measure (Loss Function)?

Quantifies how well or accurately the model's output matches the target label.

What is the role of an Optimizer?

A tool/algorithm that uses the performance measure to update and fine-tune the model's parameters

What is a Neural Network?

A powerful category of ML models built from interconnected units called neurons organized in layers.

What are Weights?

The specific name for parameters within a neural network

How do layers in a neural network interact?

Each layer takes the outputs of the previous layer as its input and sends its outputs to the next layer.

What makes a neural network "Deep" (Deep Learning)?

Having multiple hidden layers of neurons stacked sequentially.

What is the Turing Test (originally the "Imitation Game")?

A test proposed by Alan Turing (1950) where a human judge chats via text; if they cannot distinguish the machine from a human, the machine exhibits intelligent behavior.

Who coined the term "Artificial Intelligence"?

John McCarthy in 1952 at Dartmouth College, defining it as "the science and engineering of making intelligent machines".

What is Artificial Intelligence (AI)?

AI is the science and engineering of making intelligent machines capable of performing tasks that normally require intelligence.

What question did Alan Turing ask in his famous 1950 paper?

"Can machines think?"

What are three key ideas behind AI?

1. Artificial/Machine: Performed by non-biological intelligence.

2. Task-Oriented: Takes action (output) to achieve a useful outcome.

3. Intelligence: Includes decision-making in ambiguous circumstances and learning from experience.

Why is converting a color image to grayscale considered a low-intelligence task?

It can be solved using a straightforward mathematical formula.

Why is cropping an image to a user-specified size not considered highly intelligent?

It follows explicit rules.

Intelligence includes what two important abilities?

Making decisions under ambiguity and learning from experience.

Why is removing an object from an image considered an intelligent task?

The computer must infer what should exist behind the removed object.

One way to think about intelligence is what?

The ability to perform tasks humans do well.

As task difficulty increases, what generally happens to the number of rules needed?

The number of required rules and exceptions increases dramatically.

Why did researchers move beyond rule-based AI?

Many real-world tasks became too complex to solve by writing explicit rules.

Machine learning is primarily based on what discipline?

Statistics.

Who first coined the term "Machine Learning"?

Arthur Samuel.

What game did Arthur Samuel's machine learn to play?

Checkers.

How did Arthur Samuel's machine improve?

By playing games against itself and learning from outcomes.

By playing games against itself and learning from outcomes.

Predict → Compare with correct answer → Adjust → Repeat.

Why is feedback important in machine learning?

It tells the model how well it performed so it can improve.

What is the correct answer for a training example called?

Why are many training examples necessary?

They help the model learn robust patterns instead of memorizing.

Give examples of features in the toaster analogy.

How are parameters initialized?

Usually with random values.

Why are parameters adjusted?

To improve model performance.

In the toaster analogy, what do parameters represent?

Is AI one single model?

No. AI consists of many specialized models.

What type of data is called structured data?

Numbers and categories.

What major category of machine learning powers many modern AI systems?

What is the fundamental unit of a neural network?

AI, Machine Learning, and Models relate how?

AI is the broad field; Machine Learning is a subfield; Models are the algorithms that perform tasks.

Complete the hierarchy: AI → ______ → Neural Networks → Deep Learning.

Machine Learning.

What are parameters specifically called inside a neural network?

What is a "layer" in the context of a neural network architecture?

What is the exact relationship between the terms "parameter" and "weight"?

What fundamentally occurs during the training phase of a machine learning model?

What fundamentally occurs during the inference phase of a machine learning model?

Are a model's internal parameters updated during the inference phase?