Ntombizakhona Mabaso

for AWS Community Builders

Posted on Jan 12

Explain Basic AI Concepts And Terminologies

#aws #ai #aipractitioner #cloud

🤖 Exam Guide: AI Practitioner
Domain 1: Fundamentals of AI and ML
📘Task Statement 1.1

🎯 Objectives

This task is about building a shared vocabulary: what core AI terms mean, how major categories relate (AI vs ML vs Deep Learning vs GenAI), and how learning/inference/data types differ.

1) Basic AI Terms

Artificial Intelligence (AI)
Artificial Intelligence is the broad field focused on creating systems that perform tasks typically requiring human intelligence (perception, reasoning, language, decision-making).

Machine Learning (ML)
Machine Learning is a subset of AI where systems learn patterns from data to make predictions/decisions, rather than being explicitly programmed with rules.

Deep Learning (DL)
Deep Learning is a subset of ML that uses neural networks with many layers to learn complex patterns (especially effective for images, audio, and text).

Neural Network
A Neural Network is a model inspired by brain-like structures. It learns by adjusting weights to minimize errors. Deep learning uses large multi-layer neural networks.

Computer Vision
Computer Vision systems are AI methods that interpret images/video (e.g., object detection, classification, OCR _(Optical Character Recognition) , segmentation)._

Natural Language Processing (NLP)
Natural Language Processing systems are AI methods that understand/generate human language such as classification, translation, summarization and chat

Model
The Model is the trained artifact that maps inputs → outputs, for example “given features → predict churn”). In deep learning, the model includes network architecture + learned weights.

Algorithm
An Algorithm is the procedure used to learn or compute outcomes. In Machine Learning, it often refers to the learning approach such as logistic regression, decision trees, and the optimization method.

Training
Training is the process of learning model parameters from data like adjusting weights/parameters to reduce loss.

Inference (Inferencing)
Inferencing is using a trained model to produce outputs for new inputs: predictions, classifications, generated text, etc.

Fit
Fit is how well a model’s learned pattern matches the true relationship in the data.

Underfitting: too simple. The model performs poorly on training and test.
Overfitting: memorizes training. The model performs well on training but poorly on new data.

Bias
Bias in the context of Responsible AI is systematic unfair differences in outcomes across groups due to data, modeling choices, or deployment context.
Note: the ML term bias is different, as in bias-variance tradeoff). In the AI exam, a bias related question usually means fairness-related bias unless, of-course stated otherwise.

Fairness
Fairness is the goal of reducing unjustified performance gaps or unequal outcomes across groups demographic parity and equalized odds. Demographic parity and equalized odds are just fairness metrics.

Large Language Model (LLM)
A Large Language Model is a foundation model trained on large corpora of text to understand and or generate language.
Capabilities of LLMs:

summarization,
Q&A,
drafting,
extraction,
classification,
"reasoning" (with limitations).

2) AI vs ML vs GenAI vs Deep Learning

Similarities and Differences

I am sure you have seen the circle or variations of it that explain the relationship between AI, ML, DL, and GenAI: So I prompted an Image Generation LLM Model to create my variation of it.

Quick hierarchy

AI (broadest)
- ML (learns from data)
- Deep Learning (ML using multi-layer neural networks)

Generative AI (GenAI)
Generative AI is a category of AI (often deep-learning-based) that generates new content such as text, images, code, audio, video.
GenAI is frequently implemented using foundation models such as LLMss

Comparison table

Term	What it means	Typical outputs	Example tasks
AI	Broad umbrella for intelligent behavior	Decisions/actions/content	Planning, perception, language
ML	Learns patterns from data	Predictions or classifications	Fraud detection, churn prediction
Deep Learning	ML using deep neural networks	Predictions + representations	Vision, speech, NLP
GenAI	Generates new content	Text/images/code/audio	Summarization, chat, image generation

Similarities: All aim to solve problems by using data and computation to achieve intelligent behavior.

Differences: ML/DL often focus on predicting labels/values. GenAI focuses on creating new content, often probabilistically.

3) Types of Inferencing

Real-time Inference: Low-latency predictions per request (milliseconds to seconds), used for interactive apps such as chatbots, recommendations, fraud checks at checkout.

Key Concerns: latency, concurrency, autoscaling, cost per request.

Batch inference: Run predictions on many records at once (minutes to hours), used for nightly scoring jobs, large dataset enrichment, periodic reporting.
_
Key Concerns: _throughput, scheduling, cost efficiency.

If a user is waiting for the response → real-time.
If results are needed on a schedule or for a large dataset → batch.

4) Types of Data in AI models

By Labeling

Labeled Data: has input + correct output/label (used in supervised learning)
- Example: emails labeled “spam”/“not spam”
Unlabeled Data: inputs only (commonly used in unsupervised learning or pretraining)
- Example: a folder of images with no tags

By Structure

Structured Data: organized in rows/columns with a schema
- Example: SQL tables, CSV with fixed columns
Unstructured data: free-form content
- Example: documents, images, audio, raw text
Semi-structured Data (often mentioned): has some structure but flexible schema
- Example: JSON, logs

By Modality

Tabular: rows/columns (customer attributes, transactions)
Time-series: ordered by time (IoT sensor readings, stock prices)
Image: photos, medical scans
Text: emails, tickets, documents, chat logs

5) Supervised Learning vs Unsupervised Learning vs Reinforcement Learning

Supervised Learning: Learns from labeled examples (X → y).

Common tasks:
- Classification: predict a category (spam vs not spam)
- Regression: predict a number (house price)
Typical success metrics: accuracy/F1 (classification), RMSE/MAE (regression)

Unsupervised Learning: Learns patterns from unlabeled data (X only).

Common tasks:
- Clustering: group similar items (customer segmentation)
- Dimensionality reduction: compress features for visualization or efficiency

Reinforcement Learning (RL): An agent learns by interacting with an environment to maximize a reward.
Key Reinforcement Learning Concepts: agent, environment, state, action, reward, policy.
Common Reinforcement Learning Tasks: robotics, game playing, dynamic resource allocation.

Quick Selection Guide

If you have…	You likely use…	Example
Inputs + correct labels	Supervised	Predict churn from customer features
Inputs only, want structure/groups	Unsupervised	Segment customers by behavior
A sequence of decisions with rewards	Reinforcement learning	Optimize bidding/robot navigation

💡 Quick Questions

Is Deep Learning a subset of ML or separate from it?
What’s the difference between Training and Inference?
When would Batch Inference be preferable to Real-Time Inference?
Give one example each of Structured and Unstructured data.
Which learning type uses rewards and trial-and-error?

Additional Resources

✅ Answers to Quick Questions

Deep learning is a subset of ML (and ML is a subset of AI). Deep learning specifically uses multi-layer neural networks.
Training vs inference
- Training: the model learns from data by adjusting parameters/weights to minimize error (loss).
- Inference: the trained model is used to make a prediction/generate an output for new input data.
When batch inference is preferable
- When you need to score large volumes of data at once and latency isn’t critical (e.g., nightly churn scoring, weekly risk scoring, backfilling predictions, dataset enrichment for analytics).
Examples
- Structured data: a SQL table of customer records (columns like customer_id, age, plan, spend).
- Unstructured data: a collection of support ticket text, PDFs, images, or audio recordings.
Learning type that uses rewards and trial-and-error:
- Reinforcement learning (RL).

DEV Community