Conversational AI uses chat-based interfaces to interpret questions, continue dialogue, and provide practical help with technical and workplace tasks. ChatGPT, Claude, Gemini, and Copilot are examples of language model chatbots, while APIs bring similar capabilities into applications. section 1.4.1 section 2.0
Use conversational AI to support work such as:
- Analyze logs, documents, datasets, or service desk tickets to identify patterns, summarize information, and suggest next steps. section 1.1 section 8.2.2
- Draft, summarize, clarify, and review workplace email, especially when the chatbot integrates with Gmail, Outlook, or another business application. section 7.1.1
- Generate customized SQL queries, translate syntax between database platforms, and explain unfamiliar database concepts. section 10.1 section 10.4
- Help service desk agents interpret user language, ask clarifying questions, and troubleshoot documented technical problems. section 8.1.1 section 8.2.2
Clarifying a fuzzy laptop screen
Prompt: “A caller says, ‘The screen is fuzzy.’ What problem might they mean? Provide 3–5 questions to clarify the issue.” The chatbot interprets “fuzzy” as a possible display-quality problem and suggests asking what appears on screen, when it happens, whether heat or applications affect it, which fixes were attempted, and whether recent changes preceded it. Treat the response as a starting point: confirm the caller’s meaning before troubleshooting. section 8.1.1
Conversational AI is useful but not automatically reliable. Token limits can shorten memory and responses, large inputs can cause missing details, and searches across many emails can produce incomplete results. Verify important answers, protect sensitive data, remove identifying details from prompts, and follow workplace policies. section 3.2.2 section 7.1.2 section 7.5
Table 1.1 Glossary of AI terms
|
Term
|
Definition
|
Example
|
|---|---|---|
|
AI
|
Software that mimics human thinking and learning
|
OpenAI ChatGPT, Apple Siri, Tesla Autopilot, Amazon Alexa, Google Assistant
|
|
AI apps
|
Programs that use AI to handle specific tasks
|
Google Maps for navigation, Netflix and Spotify for recommendations, Zoom for noise reduction
|
|
AI model
|
Software trained on data to spot patterns and make decisions
|
OpenAI GPT-4, Tesla Full Self-Driving (FSD) AI, Stable Diffusion
|
|
API
|
Rules that let software systems talk to each other; used to add AI features to programs
|
OpenAI API, Anthropic API, Google Cloud AI API, AWS Bedrock API, Azure AI Services API
|
|
Chatbot
|
AI program that talks with users to help or provide information
|
OpenAI ChatGPT, Anthropic Claude (Sonnet frontend)
|
|
Cloud-hosted AI
|
AI systems running on internet servers that are easy to use and scale
|
OpenAI API, Azure AI services, AWS AI services, Google AI services
|
|
Generative AI
|
AI that creates new content by learning from existing data
|
DALL-E, Stable Diffusion, Midjourney, ChatGPT, Google Imagen
|
|
LLMs
|
Big AI systems trained on lots of text that can understand and write like humans
|
OpenAI GPT-4, Google Gemini, Anthropic Sonnet, Meta Llama
|
|
Local LLM
|
AI that runs on your own computers for better privacy and control
|
Meta Llama, Mistral AI, Microsoft Phi, Black Forest Labs FLUX
|
|
SLMs (small language models)
|
Smaller AI models that work well with limited resources
|
OpenAI GPT-4o mini, Microsoft Phi, Mistral Small
|
Use the glossary to distinguish chatbots from AI models, APIs, generative AI, and cloud-hosted AI. section 1.4.1
