Glossary of AI Terms
Must-Know Basics
1. Artificial Intelligence (AI)
The broad field of computer science focused on building systems that can perform tasks that typically require human intelligence, such as learning, reasoning and decision-making.
📌 Analogy: Think of AI as a digital assistant that learns over time—like an intern who starts with limited knowledge but improves with experience.
2. Machine Learning (ML)
A subset of AI where computers learn from data without being explicitly programmed for every possible scenario.
Machine learning is a way for computers to learn from data and improve over time without being explicitly programmed.
📌 Analogy: It's like teaching a child to recognize animals by showing pictures instead of listing every possible trait, or a chef getting better with experience instead of just following a fixed recipe.
⚙️ Practical Use: Fraud detection systems in banking that learn from past transaction data to spot unusual activity.
3. Deep Learning (DL)
A more advanced form of ML that uses neural networks with multiple layers to process complex patterns in data, like a brain recognizing faces or voices.
📌 Analogy: Imagine a detective solving a mystery by piecing together tiny clues spread across different crime scenes—deep learning does the same with vast amounts of data.
⚙️ Practical Use: Facial recognition on smartphones.
4. Neural Networks
Algorithms inspired by the human brain that process information in interconnected layers to identify patterns and relationships. These are smart algorithms that learn from data instead of following fixed rules, making them more like a brain than traditional programs.
📌 Analogy: It’s like a web of neurons in your brain firing up when you recognize a familiar face.
⚙️ Examples: Tesla Pilot Deep Neural Networks (DNNs) , Google Lens Convolutional Neural Networks (CNNs).
5. Natural Language Processing (NLP)
The branch of AI that enables machines to understand, interpret, and respond to human language like how Siri or chatbots can have conversations with you.
📌 Analogy: Imagine a multilingual translator who can instantly understand and respond in different languages and dialects.
⚙️ Practical Use: A customer support chatbot that understands and responds to user queries in natural language.
6. Generative AI
AI that can create new content, such as text, images, music, and code, by learning from existing data, similar to how an artist gets inspired to paint something new.
📌 Analogy: Think of it as an artist who has studied thousands of paintings and can now create original masterpieces.
⚙️ Practical Use: AI-powered marketing tools that draft social media posts or design logos.
7. Large Language Model (LLM)
An AI model trained on massive amounts of text to generate human-like responses and understand context, like a supercharged autocomplete that can write essays, answer questions, and even chat with you.
📌 Analogy: Think of a well-read scholar who has absorbed the world’s literature and can craft responses in various styles and tones.
⚙️ Examples: ChatGPT, Gemini, Claude, X (Grok), Mistral, and Llama.
8. Prompt Engineering
The art of designing inputs (prompts) to get the best responses from AI models.
📌 Analogy: It’s like knowing the exact words to say to a genie to get the wish you truly want.
9. Supervised Learning
A type of ML where the AI is trained using labeled data, meaning it learns from examples with correct answers.
📌 Analogy: Like a student learning with an answer key—getting immediate feedback on what’s right or wrong.
10. Unsupervised Learning
ML where AI finds patterns in data without labeled examples, discovering hidden relationships on its own.
📌 Analogy: Like a detective sorting out crime scenes without knowing which evidence belongs to which case.
11. Reinforcement Learning
A training method where AI learns by trial and error, receiving rewards for good decisions and penalties for bad ones.
📌 Analogy: Like training a dog—if it sits on command, it gets a treat.
12. Bias in AI
Unintended favoritism in AI outputs caused by skewed or unrepresentative training data.
📌 Analogy: Like teaching a child only one side of a story and expecting them to make fair judgments.
13. Hallucination (in AI)
When an AI generates incorrect or misleading information that sounds convincing.
📌 Analogy: Like a confident but misinformed friend making up "facts" during a conversation.
14. Explainability (XAI - Explainable AI)
The ability to understand and interpret how an AI model arrives at a decision.
📌 Analogy: Like asking a chef for the recipe behind a dish instead of just accepting the meal.
15. Model Training
The process of feeding data into an AI algorithm so it can learn patterns and make predictions.
📌 Analogy: Like practicing a sport—repeating drills until muscle memory kicks in.
16. Inference
When an AI model makes a decision or prediction based on what it has learned.
📌 Analogy: Like recognizing a song after hearing just a few notes.
17. API (Application Programming Interface)
A set of rules that allows different software programs to communicate with each other.
📌 Analogy: Like a waiter in a restaurant—taking your order (request) to the kitchen (backend system) and bringing your meal (response) back.
🆕 Update (MCP – Model Context Protocol):
In modern AI systems, APIs enable structured interaction between models and external tools or data. The Model Context Protocol (MCP) builds on this concept—standardizing how AI models securely access, request, and use contextual information across different systems.
📌 Analogy (MCP): If an API is the waiter, MCP is the restaurant manager who ensures every waiter follows the same service rules, knows where the kitchen is, and can serve any customer consistently—no matter which restaurant (system) they’re in.
18. Chatbot
A conversational AI that interacts with users via text or voice.
📌 Analogy: Like an automated customer service rep that can answer FAQs without human intervention.
19. Computer Vision
AI that enables machines to interpret and process visual data from images or videos.
📌 Analogy: Like giving a robot "eyes" that can recognize objects and people.
20. Edge AI
AI that processes data locally on a device instead of relying on the cloud.
📌 Analogy: Like a smartwatch that analyzes your heart rate in real-time without needing an internet connection.
21. Fine-Tuning
Adjusting a pre-trained AI model with new, specific data to improve performance for a particular use case.
📌 Analogy: Like training a general athlete to specialize in one sport.
22. Tokenization
Breaking text into smaller units (tokens) so an AI model can process language efficiently.
📌 Analogy: Like breaking a long novel into chapters, paragraphs, and words for better comprehension.
23. Data Pipeline
A system that automates the collection, processing, and flow of data into AI models.
📌 Analogy: Like an assembly line in a factory, moving raw materials (data) through different stages until a finished product (AI insights) is ready.
24. AI Ethics
Guidelines for ensuring AI is used responsibly, fairly, and without causing harm.
📌 Analogy: Like setting rules for self-driving cars to ensure they follow traffic laws and don’t cause accidents.
25. Zero-Shot Learning
An AI’s ability to handle tasks it hasn’t seen before without prior training examples.
📌 Analogy: Like a child guessing the meaning of a new word based on surrounding words in a sentence.
Advanced AI Concepts
1. Retrieval-Augmented Generation (RAG)
A hybrid AI approach that combines retrieving relevant documents from a knowledge source with a generative AI model to provide more factual, contextually accurate responses.
📌 Analogy: Like an open-book exam where an AI not only generates answers but also pulls supporting references from trusted materials.
2. Agentic AI
AI systems that operate with a degree of autonomy, making decisions and performing actions based on goals rather than just responding to prompts.
📌 Analogy: Like a digital assistant that proactively books meetings, suggests improvements, and executes tasks without constant human input.
3. Multi-Modal AI
AI that can process and understand multiple types of inputs (text, images, audio, video) in a single model.
📌 Analogy: Like a detective who can read, listen to interviews, analyze photos, and watch security footage to solve a case.
4. Embeddings
Numerical representations of words, images, or other data that allow AI models to understand similarity and relationships.
📌 Analogy: Like mapping books in a library based on themes rather than alphabetical order, so similar ideas are placed closer together.
5. Vector Database
A specialized database designed to store and search embeddings efficiently, used in RAG systems for fast retrieval.
📌 Analogy: Like a highly organized digital filing cabinet that can instantly find the most relevant documents based on meaning, not just keywords.
6. Grounding (in AI)
The process of ensuring an AI system's responses are based on real-world data or trusted sources rather than purely generating answers from its training data.
📌 Analogy: Like fact-checking an article before publishing it, ensuring accuracy.
7. Long Context Window
The ability of AI models to remember and process large amounts of previous conversation history or documents at once.
📌 Analogy: Like a lawyer who can recall every detail from a case file instead of just summarizing the highlights.
8. Latency (AI Response Time)
The time delay between a user input and an AI's response, influenced by model size, retrieval speed, and computation power.
📌 Analogy: Like waiting for a webpage to load—faster response times lead to better user experience.
9. Chunking (in AI Retrieval)
Breaking large documents into smaller, manageable sections to improve search accuracy and reduce processing time in RAG pipelines.
📌 Analogy: Like splitting a novel into chapters so it’s easier to find relevant passages.
10. Knowledge Distillation
A technique where a smaller AI model is trained to replicate the knowledge of a larger, more complex model while maintaining efficiency.
📌 Analogy: Like summarizing a 500-page book into a 5-page executive summary without losing key insights.
11. Hallucination Guardrails
Techniques and systems designed to minimize AI hallucinations by enforcing accuracy checks and content filtering.
📌 Analogy: Like a GPS that cross-checks multiple routes before suggesting directions to avoid sending you down a dead-end road.
12. Data Provenance
Tracking the origin and history of data used in AI models to ensure transparency and credibility.
📌 Analogy: Like verifying the authenticity of a painting by tracing its ownership history back to the original artist.
13. Prompt Injection Attack
A security vulnerability where malicious inputs are designed to manipulate AI models into executing unintended actions.
📌 Analogy: Like tricking a gullible friend into saying something they didn't intend to.
14. Adversarial Attacks (on AI models)
Deliberate manipulations of input data to deceive AI models into making incorrect predictions.
📌 Analogy: Like wearing camouflage to avoid detection by security cameras.
15. Fine-Grained Access Control
A security framework that restricts which users or AI systems can access specific data based on roles and permissions.
📌 Analogy: Like a VIP event where only guests with the right credentials can enter specific areas.
16. Differential Privacy
A technique to ensure AI models do not leak sensitive user data while still learning from vast datasets.
📌 Analogy: Like reporting average employee salaries without revealing any one person’s pay.
17. Federated Learning
A decentralized approach where AI models are trained across multiple devices or locations without sharing raw data.
📌 Analogy: Like training a global sales team using regional case studies without exposing confidential client information.
18. Data Poisoning
A security attack where adversaries manipulate training data to degrade an AI model’s performance or bias its responses.
📌 Analogy: Like tampering with a recipe by replacing sugar with salt to ruin the final dish.
19. Model Drift
A phenomenon where an AI model’s accuracy decreases over time as real-world data evolves away from its training data.
📌 Analogy: Like a weather prediction system becoming unreliable if it only learned from past climate trends and ignored new patterns.
20. Shadow AI
Unauthorized or unmonitored AI tools and models deployed within an organization without proper governance.
📌 Analogy: Like employees using unsanctioned software tools without IT’s knowledge, leading to security risks.
21. Synthetic Data
Artificially generated data used to train AI models when real-world data is scarce or privacy-sensitive.
📌 Analogy: Like a flight simulator creating virtual crash scenarios to train pilots without real-world risks.
22. Content Moderation AI
AI systems designed to detect and filter harmful, biased, or inappropriate content in chatbot interactions.
📌 Analogy: Like a bouncer at a nightclub ensuring only appropriate conversations happen inside.
23. Zero-Trust AI Security
A security model that assumes no entity (user, AI system, or data source) is inherently trustworthy and requires continuous verification.
📌 Analogy: Like airport security where every traveler must pass through checks, even frequent flyers.
24. Self-Healing AI Systems
AI architectures that automatically detect and correct their own errors or adapt to changing environments without human intervention.
📌 Analogy: Like a car that fixes minor mechanical issues on its own while driving.
25. Autonomous Agents
AI-driven entities capable of making decisions and taking actions independently within a defined scope, without requiring constant human supervision. These agents can process information, adapt to new inputs, and execute tasks to achieve a goal.
📌 Analogy: Like a well-trained service dog—it listens for specific cues, makes independent decisions based on the situation (like guiding its owner away from danger), and knows when to seek human assistance. It operates autonomously within its training but still adheres to clear boundaries.
Use-case: An inbound call voice bot for customer service—when a customer calls, it listens, understands the request, retrieves relevant information, and provides assistance or escalates the call when necessary. It operates autonomously within its defined scope, ensuring efficiency while respecting its limits.
Further Reading:
© 2025 • Florante Pascual • All Rights Reserved