AI Glossary of Terms

What is AI? The Basics

• Artificial Intelligence (AI): Computer systems that can think and learn like humans, performing tasks such as understanding conversations, recognizing patterns, solving problems, and making decisions

• Artificial Narrow Intelligence (ANI): AI that's good at one specific job (like playing chess or recommending films) but can't do other tasks outside its specialty

• Artificial General Intelligence (AGI): A future type of AI that could learn and perform any task a human can do - this doesn't exist yet and remains theoretical

• Agentic AI: AI that can work independently to complete goals without constant human guidance, making its own decisions along the way

• Algorithms: Step-by-step instructions that tell computers how to solve a problem or complete a task

• Cognitive Computing: Computer systems designed to think more like humans, using reasoning and understanding to help people make better decisions

  
  

Types of AI You'll Encounter Daily

• AI Assistant: Conversational tools like ChatGPT or Copilot that chat with you and help complete various work tasks

• Chatbots: Programs that have conversations with customers through text or voice, often used for customer service

• Generative AI: AI that creates new content - writing text, generating images, producing code, or making videos based on your instructions

• Computer Vision: Technology that allows computers to understand and interpret images and videos, like facial recognition or scanning documents

• Natural Language Processing (NLP): Technology that helps computers understand and respond to human language, both written and spoken

• Speech Recognition: Software that converts spoken words into written text, like voice dictation or voice commands

• Robotics: Combining AI with physical machines to automate tasks in factories, warehouses, or healthcare settings

• Predictive Analytics: Using data from the past to forecast what might happen in the future, like predicting customer behaviour or hospital readmissions

  
  

How AI Learns

• Machine Learning (ML): Teaching computers to learn from experience and improve over time without being explicitly programmed for every scenario

• Supervised Learning: Teaching AI by showing it examples with correct answers, like showing it labelled photos of cats and dogs so it learns to tell them apart

• Unsupervised Learning: Letting AI find patterns in data on its own without being told what to look for, like grouping customers with similar buying habits

• Reinforcement Learning: Teaching AI through trial and error with rewards for good decisions and penalties for bad ones, similar to training a pet

• Deep Learning (DL): A sophisticated type of machine learning inspired by how the human brain works, particularly good at recognizing complex patterns in images, speech, and text

• Transfer Learning: Using knowledge an AI learned from one task to help it learn a different but related task more quickly

• Training Data: The information and examples you feed into AI systems to teach them how to perform tasks correctly

  
  

AI Model Types & Architecture

• Neural Networks: Computer systems modeled on the human brain, using interconnected nodes that process information in layers

• Large Language Model (LLM): Powerful AI trained on massive amounts of text that can understand and generate human-like writing, like ChatGPT or Claude

• Transformers: A modern AI architecture that's particularly effective at understanding context and relationships in text, powering most advanced language AI today

• Generative Adversarial Network (GAN): Two AI systems that work against each other - one creates content while the other judges if it's real or fake, making both better over time

• Frontier Models: The most advanced AI systems available today that can handle many different types of tasks across multiple areas

• Multimodal AI: AI that can work with different types of information at once - text, images, audio, and video combined

  
  

Working with AI Models

• Fine-tuning: Customizing a pre-trained AI model with your specific data to make it better at your particular tasks, like adapting it to your industry's terminology

• Prompt Engineering: The skill of writing clear, effective instructions to get the best results from AI tools

• Token/Tokenization: How AI breaks down text into smaller chunks to process it - think of it as how AI "reads" information piece by piece

• Context Window: How much information an AI can remember and consider at one time during a conversation or task

• Parameters: The internal settings that determine how an AI behaves and makes decisions, learned during training

• Embedding: How AI converts words or concepts into a format it can understand and compare, placing similar ideas close together

• Feature Engineering: Selecting and preparing the most relevant information from your data to help AI make accurate predictions

  
  

Measuring AI Performance

• Model Validation: Testing an AI system on new data it hasn't seen before to check if it works properly in real situations

• Accuracy: What percentage of the AI's predictions or decisions are correct - the most basic measure of performance

• Precision: Of all the times AI says "yes," how often is it actually correct? Important when false alarms are costly

• Recall/Sensitivity: Of all the correct "yes" answers that exist, how many does the AI actually find? Critical when missing something is dangerous

• F1 Score: A balanced measure that considers both precision and recall together, giving you one overall quality score

• Overfitting: When AI memorizes training examples too exactly and struggles with new situations, like a student who only memorized past exam questions

• Underfitting: When AI is too simple to learn the patterns in your data, producing poor results across the board

• Model Drift: When an AI's performance gets worse over time because the real world has changed since it was trained

  
  

AI Problems & Challenges

• Hallucinations: When AI confidently presents false information as fact, making up details that sound plausible but aren't true

• Bias in AI: When AI makes unfair decisions based on prejudices hidden in its training data, potentially discriminating against certain groups

• Black Box AI: AI systems where you can't see or understand how they reached their decisions, making it hard to trust or verify their reasoning

• Anomaly Detection: Finding unusual patterns or outliers in data that might signal problems, fraud, or opportunities

  
  

Data & Information

• Big Data: Extremely large collections of information that are too big to analyze using traditional methods, requiring specialized tools

• Data Science: The practice of extracting useful insights and knowledge from data to inform business decisions

• Data Mining: Automatically searching through large amounts of data to discover patterns, trends, and relationships

• Data Analytics & Visualisation: Examining data to identify trends and presenting findings in charts, graphs, and dashboards for easy understanding

• Structured Data: Organized information stored in tables with clear categories, like spreadsheets or databases - easy for computers to process

• Synthetic Data: Computer-generated fake data that looks and behaves like real data, useful when real data is unavailable or sensitive

  
  

Technology & Infrastructure

• GPU: Specialized computer chips originally designed for graphics that excel at the type of calculations AI needs, making AI run much faster

• Computational Power: The amount of processing capability needed to run AI, often measured by hardware specifications

• Edge AI: Running AI directly on local devices (like phones or sensors) rather than sending data to distant servers, enabling faster responses

• IoT: Internet of Things - everyday devices connected to the internet that collect and share data, like smart thermostats or fitness trackers

• API: A way for different software programs to communicate and share information with each other automatically

  
  

Healthcare-Specific AI

• Clinical Decision Support System (CDSS): AI tools that provide doctors and nurses with evidence-based recommendations to support their clinical decisions

• Digital Twin: A computer simulation of a real patient that doctors can use to test different treatments virtually before applying them in real life

• Precision Health: Personalizing healthcare for each individual patient using AI to analyze their unique characteristics and predict what treatments will work best

• Population Health Analytics: Using AI to analyze health data from entire communities to identify trends, prevent disease, and improve care

• EHR (or EPR) Integration: Connecting AI tools directly into electronic health record systems so healthcare professionals get AI insights within their normal workflow

  
  

Ethics, Safety & Governance

• AI Ethics: The principles and values that guide responsible development and use of AI, ensuring it benefits society

• Responsible AI: Building and using AI in ways that are ethical, safe, transparent, and beneficial to people and society

• Fairness in AI: Ensuring AI treats all people equitably regardless of their background, characteristics, or circumstances

• Explainable AI (XAI): Designing AI so people can understand how it reaches its decisions, building trust and enabling oversight

• Transparency: Being open about how AI systems work, what data they use, and how they make decisions

• AI Governance: The policies, procedures, and oversight that ensure AI is developed and used appropriately within your organization

• Hyperparameters: The configuration choices made before training an AI that control how the learning process works