FOUNDATIONS OF AI — TERMINOLOGY & CONCEPTS

Defining Artificial Intelligence

Definition: The simulation of human cognitive processes by machines, specifically pattern recognition and decision-making.
Distinction: Differentiate between "Narrow AI" (task-specific) and "General AI" (broad reasoning).
It is a toolset, not a sentient agent.

Intent: Define AI as a field of computer science. Avoid anthropomorphizing the software (e.g., "it wants," "it thinks").

FOUNDATIONS OF AI — TERMINOLOGY & CONCEPTS

Deterministic vs. Probabilistic Systems

Traditional Programming (Deterministic): Explicit, rule-based logic. (If X, then Y. Always.)
Machine Learning (Probabilistic): Inference-based predictions. (Given X, Y is 90% likely.)
Shift from "coding the solution" to "coding the learner."

Intent: Contrast traditional software engineering with machine learning paradigms.

FOUNDATIONS OF AI — TERMINOLOGY & CONCEPTS

Machine Learning (ML)

Algorithms that improve systematically through experience (data).
The system identifies correlations and patterns within datasets to make predictions.
The "rules" are inferred from the data, not explicitly programmed by a human.

Intent: Define the core mechanism driving modern AI.

FOUNDATIONS OF AI — TERMINOLOGY & CONCEPTS

The Model: A Mathematical Representation

A "Model" is a static file containing the weighted mathematical representation of patterns learned during training.
It is not a database of facts; it is a statistical structure of relationships.
Analogy: A map (the model) vs. the territory (the data).

Intent: Demystify the "brain" of the AI. It's a high-dimensional statistical map.

FOUNDATIONS OF AI — TERMINOLOGY & CONCEPTS

Training Data & Corpus

Corpus: The massive body of text/images used for training.
The model's output quality is strictly bound by the quality and diversity of this data.
GIGO: Garbage In, Garbage Out.
Ethical Considerations: Copyright, bias, and representation.

Intent: Explain the source of the model's capabilities and biases.

FOUNDATIONS OF AI — TERMINOLOGY & CONCEPTS

The Training Process: Iterative Optimization

The process involves feeding data, measuring error, and adjusting internal values to reduce that error.
Gradient Descent: The computational process of minimizing prediction error step-by-step.
It is a mathematical optimization process, not "reading" in the human sense.

Intent: Explain "how" it learns without getting lost in calculus.

FOUNDATIONS OF AI — TERMINOLOGY & CONCEPTS

Parameters: Weights & Biases

Parameters: The internal numerical values that the model adjusts during training.
They represent the strength of connections between data points.
GPT-4 has trillions; local models have billions. This count loosely correlates to complexity and nuance.

Intent: Introduce the technical term for "knowledge" in a neural network.

FOUNDATIONS OF AI — TERMINOLOGY & CONCEPTS

Model Architecture & Scale

Parameter Count: Often used as a proxy for "intelligence" or capability.
Trade-off: Larger models capture more nuance but require exponential amounts of compute (FLOPS) and energy.
Balancing Efficiency vs. Capability.

Intent: Discuss the trade-offs of computational scale.

FOUNDATIONS OF AI — TERMINOLOGY & CONCEPTS

Post-Training Immutability (The "Freeze")

Once training is complete, the weights are "frozen."
The model does not learn from user interactions during deployment.
It is a static snapshot of the world at the time of training.

Intent: Correct the misconception that models learn in real-time.

FOUNDATIONS OF AI — TERMINOLOGY & CONCEPTS

Inference: The Forward Pass

Inference: The process of feeding live data into a static model to get a prediction.
This is the "usage" phase (what happens when you type a prompt).
It is computationally cheaper than training but still resource-intensive.

Intent: Define the runtime operation of AI.

FOUNDATIONS OF AI — TERMINOLOGY & CONCEPTS

Predictive Modeling vs. Descriptive Statistics

Descriptive Statistics: Analyzing past data to understand what happened.
Predictive AI: Using patterns to generate likely future outcomes or new data.
Generative AI: Creating new artifacts that statistically resemble the training data.

Intent: Distinguish AI from standard data analysis.

FOUNDATIONS OF AI — TERMINOLOGY & CONCEPTS

Deployment: Cloud (API) vs. Local (Edge)

Cloud/API: Inference runs on remote data centers. (Issues: Latency, Privacy, Cost).
Local/Edge: Inference runs on local hardware. (Benefits: Privacy, Control, Zero Latency).
Data Sovereignty: Who owns the input and output?

Intent: Introduce the architectural choice of where inference happens.

FOUNDATIONS OF AI — TERMINOLOGY & CONCEPTS

Hallucination: Probabilistic Fabrication

Hallucination: When a model generates grammatically correct but factually invalid information.
Cause: The model is designed to predict the next likely token, not to verify truth.
It prioritizes plausibility over accuracy.

Intent: Address the reliability problem with technical accuracy.

FOUNDATIONS OF AI — TERMINOLOGY & CONCEPTS

The Role of the Human: Architect & Auditor

Because models are probabilistic, human oversight is non-negotiable.
The human provides the Specification (Logic) and Validation (Auditing).
AI accelerates production; humans ensure correctness.

Intent: Reframe the student's role from "user" to "expert overseer."

FOUNDATIONS OF AI — TERMINOLOGY & CONCEPTS

Transition to Architecture

"We have defined the software structures. Now, what defines the physical limits of these systems?"

Intent: Bridge the gap to the hardware deck.