The TEAM Solutions AI Glossary
AI terms can be confusing. Here is a simple glossary written in plain language, with examples and quick FAQs. For decision support, use the TEAM Solutions Decision Assistant. For quick lookups across the site, tap the red magnifying glass Navigator on the right side or at the bottom of each page.
Chatbot
Definition: Like texting with a robot. You ask a question, it replies. Some are smart and flexible, others follow scripts.
Other Names: bot, virtual agent, service bot.
Example in Action: A website pop-up that asks “How can I help you today?” and answers basic questions.
For quick lookups on this site, use the red magnifying glass Navigator on the right side or at the bottom of each page.
FAQ: What is a chatbot? (click to expand)
A chatbot is a computer program that talks with you by text or voice to help you get information or complete simple tasks.
AI Assistant
Definition: A chatbot with extra brainpower. It does more than talk. It can summarize, draft, guide, and help you complete tasks.
Other Names: digital assistant, virtual assistant, smart helper.
Example in Action: Drafting an email, organizing ideas, or guiding a decision process.
Try the TEAM Solutions Decision Assistant for decisions using the McKenna 4AID Decision Model.
FAQ: How is an AI assistant different from a chatbot? (click to expand)
An AI assistant can hold context, perform useful tasks, and guide you through steps. A basic chatbot mainly answers short questions.
AI Co-Pilot
Definition: A helper that works alongside you. You stay in control. The AI suggests, checks, or fills in while you focus on decisions.
Other Names: co-pilot, coding co-pilot, writing co-pilot, design co-pilot.
Example in Action: Suggesting next steps in a plan, spotting gaps, or offering drafts you can accept or edit.
The Decision Assistant acts like a co-pilot when you apply the 4AID Decision Model.
FAQ: Why use a co-pilot instead of full automation? (click to expand)
A co-pilot boosts speed and quality while you keep judgment and control. Good for high-stakes work.
AI Automation
Definition: Setting rules so repetitive work happens automatically after a trigger. Think dominoes after the first push.
Other Names: workflow automation, process automation, RPA.
Example in Action: New customer purchases, then the system instantly delivers access and sends onboarding emails.
FAQ: What tasks are good for automation? (click to expand)
Repeatable, rules-based tasks like sorting messages, sending confirmations, generating routine reports, and routing tickets.
GPT
Definition: A powerful AI that learned from huge amounts of text so it can write and chat like a person.
Other Names: ChatGPT, GPT-4, GPT-5.
Example in Action: Answering questions, drafting plans, summarizing long docs.
The TEAM Solutions Decision Assistant uses GPT combined with the 4AID Decision Model.
FAQ: What does GPT stand for? (click to expand)
Generative Pre-trained Transformer. It generates text, was pre-trained on lots of data, and uses a transformer architecture.
Large Language Model (LLM)
Definition: A computer brain trained on tons of text that can read, write, and follow instructions in natural language.
Other Names: foundation model, transformer model.
Example in Action: ChatGPT is an LLM that powers conversational tools.
FAQ: Is GPT an LLM? (click to expand)
Yes. GPT is a family of LLMs.
Machine Learning (ML)
Definition: Computers learn patterns from examples instead of rules written by a programmer.
Other Names: ML, deep learning (a type of ML).
Example in Action: Recommendations based on your past behavior.
FAQ: How does ML learn? (click to expand)
It compares predictions with real outcomes, then adjusts until it gets better.
Neural Network
Definition: A math system inspired by brains. It has layers of simple units that pass signals to find patterns.
Other Names: deep neural net, CNN, RNN, transformer.
Example in Action: Face unlock on a phone uses neural networks.
FAQ: Why is it called a neural network? (click to expand)
Because its simple units are loosely inspired by neurons in the brain.
Training Data
Definition: The examples an AI studies to learn. More and better data usually means better results.
Other Names: datasets, corpora, labeled data.
Example in Action: Thousands of reports used to train a model to classify incidents.
FAQ: Why does data quality matter? (click to expand)
Poor or biased data leads to poor or biased results.
Prompt
Definition: The instruction or question you give to an AI.
Other Names: input, query, prompt engineering.
Example in Action: “Summarize this policy in 5 bullets.”
FAQ: What is prompt engineering? (click to expand)
Writing clearer instructions and examples so the AI gives better answers.
Hallucination
Definition: When AI makes something up and presents it like a fact.
Other Names: fabrication, model error.
Example in Action: Citing a source that does not exist.
FAQ: How do I reduce hallucinations? (click to expand)
Use retrieval from trusted sources, ask for citations, and verify important claims.
Bias
Definition: Unfair patterns in outputs caused by skewed data or processes.
Other Names: algorithmic bias, data bias.
Example in Action: An assistant that overlooks certain groups because of limited training data.
FAQ: How do I manage bias? (click to expand)
Improve data diversity, test outputs, add guardrails, and involve human review for sensitive use cases.
API
Definition: A way for apps to talk to each other. Like a waiter carrying your order to the kitchen and back.
Other Names: REST API, API call.
Example in Action: A website calling an AI model to generate a summary.
FAQ: Why use APIs with AI? (click to expand)
APIs make it easy to plug AI into websites, apps, and workflows.
Token
Definition: A small chunk of text the model reads or writes. Models count tokens, not characters.
Other Names: tokenization, input tokens, output tokens.
Example in Action: Long prompts cost more tokens and may hit context limits.
FAQ: Why do tokens matter? (click to expand)
They affect speed, cost, and how much context the model can handle at once.
AI Model
Definition: The trained brain that generates answers.
Other Names: model, foundation model, fine-tuned model.
Example in Action: Choosing between a general model and a smaller fine-tuned model for a specific task.
FAQ: When should I use a smaller model? (click to expand)
When speed, cost, or privacy are critical and the task is narrow.
Fine-Tuning
Definition: Teaching a model your special cases with extra lessons so it performs better for your needs.
Other Names: custom training, tuning.
Example in Action: Training on your SOPs so the assistant writes in your style.
FAQ: How is fine-tuning different from retrieval? (click to expand)
Fine-tuning changes the model’s internal behavior. Retrieval feeds the model fresh facts at answer time without changing the model.
Inference
Definition: The act of the model generating an answer from your prompt and context.
Other Names: generation, prediction.
Example in Action: You ask, it answers. That moment is inference.
FAQ: How do I speed up inference? (click to expand)
Use smaller models, shorter prompts, and efficient settings. Cache what you can.
Embeddings
Definition: Turning text into numbers so the system can measure similarity between ideas.
Other Names: vector embeddings, text embeddings.
Example in Action: Finding related documents even when they do not share exact words.
FAQ: Why are embeddings useful? (click to expand)
They power semantic search, recommendations, clustering, and RAG.
RAG (Retrieval-Augmented Generation)
Definition: The model looks up trusted sources first, then writes the answer using what it found.
Other Names: retrieval-guided generation, hybrid search.
Example in Action: An assistant that searches your policies and then answers with citations.
FAQ: Why use RAG? (click to expand)
RAG reduces hallucinations and keeps answers current without retraining the model.
Agent
Definition: An AI that can not only answer but also take actions with tools or apps.
Other Names: AI agent, autonomous agent, tool-using agent.
Example in Action: Searching, filling forms, or scheduling after your approval.
FAQ: Are agents safe to use? (click to expand)
Use guardrails, human approvals, and clear scopes. Start simple and expand carefully.
Conversational AI
Definition: Any AI that communicates in natural language through chat or voice.
Other Names: dialog system, chat AI.
Example in Action: Back-and-forth Q&A with context over multiple turns.
FAQ: What makes conversational AI feel natural? (click to expand)
Good memory of context, concise answers, and the ability to ask clarifying questions.
OpenAI
Definition: The company behind GPT and ChatGPT.
Example in Action: Many assistants use OpenAI models under the hood.
FAQ: Is ChatGPT the same as OpenAI? (click to expand)
ChatGPT is a product made by OpenAI. OpenAI is the company.
Anthropic
Definition: The company that makes Claude, another leading AI assistant model.
Example in Action: Some tools choose Claude for its reasoning style or safety profile.
FAQ: How is Claude different from GPT? (click to expand)
They are both strong LLMs. Differences show up in tone, safety choices, and specific capabilities.
Google DeepMind
Definition: Google’s advanced AI research group known for AlphaGo and other breakthroughs.
Example in Action: DeepMind models power parts of Google products.
FAQ: Is DeepMind separate from Google? (click to expand)
DeepMind is part of Google and operates as Google DeepMind.
Turing Test
Definition: A thought experiment that asks if a machine can converse well enough to be mistaken for a human.
Other Names: imitation game.
Example in Action: If a judge cannot tell which chatter is a machine, the machine “passes.”
FAQ: Does passing the Turing Test mean human-level AI? (click to expand)
No. It measures conversational mimicry, not full human intelligence.
AI Ethics
Definition: Principles for building and using AI fairly, safely, and transparently.
Other Names: responsible AI, trustworthy AI.
Example in Action: Explaining limitations, protecting privacy, and reducing bias.
FAQ: Who owns AI ethics in an organization? (click to expand)
Leaders share responsibility. Governance teams coordinate policy, testing, and oversight.
AI Governance
Definition: Policies and controls that guide how AI is selected, used, audited, and improved.
Other Names: AI oversight, governance frameworks.
Example in Action: Standard reviews, red-teaming, usage logs, and human approval steps.
FAQ: Why do I need AI governance? (click to expand)
To manage risk, meet regulations, protect users, and keep quality high as usage scales.
Explainable AI (XAI)
Definition: AI that can show why it gave a result so people can trust and verify it.
Other Names: interpretable AI.
Example in Action: A model that cites sources and highlights the parts it used.
FAQ: Do all AI systems need to be explainable? (click to expand)
High-impact decisions benefit from explanation. For low-risk tasks, speed may matter more.
AI Hallmarks for Business
Definition: Common ways AI shows up at work: assistants, co-pilots, automations, navigators, and agents.
Other Names: business AI patterns, enterprise AI use cases.
Example in Action: A decision assistant for leaders, a navigator for FAQs, automations for handoffs, and agents that take actions with approval.
FAQ: Where should I start with AI in my org? (click to expand)
Start with one valuable problem. Pilot an assistant or use the red Navigator icon for quick answers. Add automation and agents after the value is proven.