Simplifying the World of Artificial Intelligence
Welcome to the AI Glossary
We've created this resource to help you understand the complex world of artificial intelligence.
We update this guide every month to keep you in the loop of the latest developments in AI. Our experts add new terms, polish existing ones, and provide fresh examples. This way, you'll always have the latest and greatest AI info at hand.
Foundational AI Concepts
These terms reference the building blocks and core ideas that form the foundation of artificial intelligence.
Artificial Intelligence (AI)
Computer systems that can do tasks that usually need human smarts, like seeing, talking, and making decisions.
Ex. When your phone recognizes your face to unlock it, that’s AI in action.
Deep Learning
A more complex form of machine learning that uses layers of artificial "neurons" to process information, kind of like a human brain.
Ex. The technology behind those AI-generated art pieces you might have seen online uses deep learning.
Machine Learning (ML)
A type of AI where computers get better at tasks by looking at lots of examples, instead of being told exactly what to do.
Ex. Netflix uses machine learning to figure out what shows you might like based on what you've watched before.
Natural Language Understanding
A branch of AI focused on a machine's ability to comprehend and interpret human language. It encompasses:
- Natural Language Processing (NLP): The technical ability to analyze human language, including parsing sentences and recognizing entities.
- Natural Language Generation (NLG): The ability to produce human-like text based on data and context.
NLU aims to grasp the intent, context, and nuances of human communication, going beyond mere word recognition to understand meaning.
Ex. When asked "How's the weather? Should I grab an umbrella?", an NLU system would understand you're inquiring about rain likelihood and the need for rain protection, potentially responding: "It's partly cloudy with a 30% chance of showers this afternoon. You might want to take an umbrella just in case."
Want to Use AI in Your Business?
We offer software that puts AI to work for you. Our tools use natural language processing and machine learning to boost innovation and efficiency. We can tailor our AI solutions to fit your specific needs.
Interested in how AI could improve your operations? Get in touch, and we'll show you how it can give your business an edge.
Generative AI
This section explores the key components and techniques that power today's most advanced Generative AI systems.
Attention
Attention is a technique that helps AI focus on what matters most in data, much like how we zero in on key details when listening or reading. It's also called "self-attention" and is crucial to transformer models, which have transformed how AI understands and creates language.
Ex. When an AI summarizes a long article, it uses attention to pick out the main ideas, just as you might underline key points while reading.
Context Window
The maximum amount of text an AI model can process at once. While larger context windows (up to 128,000 tokens or about 100,000 words) allow models to handle more input, their effectiveness for complex reasoning tasks remains debated.
Ex. While an AI with a large context window could theoretically process entire research papers, it may still struggle with comprehensive analysis. Practitioners often use RAG and carefully designed chunking methods to effectively handle large documents and complex reasoning tasks.
Large Language Models (LLM)
An advanced AI system capable of understanding and generating human-like text across a wide range of topics. It processes and produces language at a scale that enables complex interactions.
Ex. Platforms like ChatGPT use LLMs to engage in nuanced conversations and help with various tasks.
Model Architecture
The overall design of an AI system. It shows how all the AI’s parts are set up and work together, like a blueprint for a building.
Ex. Just as houses and office buildings have different layouts, AI models have different setups depending on what they’re meant to do.
Neural Network Layer
A part of an AI that handles a specific job in processing information. Different layers work on different aspects of a task, kind of like an assembly line.
Ex. In an AI that recognizes faces, one layer might find the edges of features, while another puts those features into a face.
Retrieval-Augmented Generation (RAG)
A technique that helps AI language models provide better responses. It works by first finding relevant information from a knowledge base and then using that information to give more accurate and contextually appropriate answers.
Ex. A legal AI assistant using RAG could first find relevant case law before giving advice on a specific legal question.
Transformer
A clever AI design that’s really good at understanding language in context. It’s the brains behind many of today’s best language AIs.
Ex. When you use a translation app that seems to understand whole sentences, not just word-for-word, it’s probably using a transformer.
For an even deeper dive, check out the National Institute of Standards and Technology (NIST) AI Glossary. It's a comprehensive, authoritative resource that we reference to ensure our definitions align with industry standards.
AI Development and Operations
This section focuses on the operational aspects of creating and managing AI systems.
Evaluation
The process of assessing a machine learning model's performance using various metrics to understand its effectiveness, strengths, and weaknesses. This crucial step helps determine how well the model generalizes to new, unseen data and guides further improvements.
Ex. When developing an AI system to detect skin cancer from images, evaluation might involve using metrics like accuracy, sensitivity, and specificity on a separate test dataset. The evaluation could reveal that while the model has high overall accuracy, it has lower sensitivity for detecting early-stage melanomas. This insight would guide researchers to focus on improving the model's ability to identify subtle signs of early-stage skin cancer, potentially by adjusting the training data or model architecture.
Fine-Tuning
Fine-tuning adjusts a pre-trained AI model for specific tasks. It's like giving an expert a crash course in a new field. It works best for tweaking style and tone, not for teaching new facts. Fine-tuning is often used for customizing chatbots. For new information, Retrieval-Augmented Generation (RAG) is usually better.
Ex. A customer service AI could be fine-tuned for a company's tone, while using RAG to access current product info.
Loss Function
In machine learning, a loss function is a measure that guides an AI model's learning process by quantifying the difference between its predictions and the actual desired outcomes. It operates at a fundamental level during training, providing continuous feedback that allows the model to adjust its internal parameters.
Ex. Think of it like an audio engineer adjusting levels during a recording session. The loss function makes minute, ongoing adjustments to optimize the model's performance, much like an engineer tweaks various elements to achieve the best sound mix.
Machine Learning Ops (MLOps)
The way companies manage their AI projects from start to finish. It’s about making sure AI systems run smoothly in the real world.
Ex. MLOps helps teams update their AI chatbot quickly without breaking it or taking it offline.
Weight
How much importance an AI gives to different pieces of information. It’s like the AI deciding which clues are more useful for solving a puzzle.
Ex. In an AI that predicts house prices, the weight for location might be higher than the weight for paint color, because location usually matters more.
Applied AI Concepts
Here, we outline key considerations when leveraging AI systems
Bias in AI
When an AI system consistently makes unfair or inaccurate decisions, often favoring certain groups or ideas over others.
Ex. If an AI used in hiring consistently recommends men over women for tech jobs, even when they have similar qualifications, that AI likely has a gender bias. This could happen if the AI was trained on data from a time when the tech industry was mostly male.
Ethical & Responsible AI
The practice of developing and using AI systems in ways that respect human rights, ensure fairness, maintain accountability, and promote the well-being of individuals and society.
Ex. An ethical AI system used in criminal justice would be designed to avoid racial profiling, provide consistent judgments across similar cases, and allow for human oversight and appeals. It would also be regularly audited to ensure it's not perpetuating or amplifying existing biases in the legal system.
Explainability
The ability to understand and clearly explain how an AI system makes decisions or predictions. It involves making the AI's reasoning process transparent and interpretable to humans.
Ex. When a bank uses AI to approve or deny loan applications, explainable AI can provide clear reasons for each decision, such as credit score, income level, or debt-to-income ratio. This transparency helps both customers and regulators understand the basis for the AI's choices.
Governance
The framework of policies, practices, and processes used to direct, manage, and monitor AI systems throughout their lifecycle, ensuring they align with organizational values, legal requirements, and ethical standards.
Ex. A company's AI governance might include policies for data privacy, regular audits of AI models for bias, clear guidelines for human oversight of AI decisions, and processes for addressing ethical concerns raised by employees or customers about AI applications.
Let's Discuss How AI Can Help Your Business
