Why the Dictionary?

GenAI is a liminal space. It's changing rapidly. As a result, new terms are continuously being added.

The Problem

Understanding new AI "buzzwords" is not always an easy task. Definitions can be too abstract and fail to describe how these terms are actually being used.

Example

What is an AI agent?

When we use the term 'agent', we typically refer to apps like customer service chat-bots or coding agents.

IBM defines it as: "a system that autonomously performs tasks by designing workflows with available tools."

• The word "system" is too generic. Agents usually refer to an app.
• Agents are not always autonomous; human-in-the-loop and user-feedback are popular ways of ensuring agents are safely executing tasks.
• Agents don't always design workflows. Some apps use orchestration graphs to make LLM models perform single, specific tasks. Even when agents design workflows, it might be a sub-flow within a larger, developer-designed workflow.
• The phrase "with available tools" is too technical relative to the rest of the definition. Someone who doesn't know what an "agent" is will probably not understand that "tools" refers to programming functions that provide additional context or computations.

Overall, the definition doesn't provide you with a concrete picture of what an agent is.

A Solution

A dictionary that simply and concretely defines AI terms. To do so, the dictionary will do the following:

1. Provide a "term type" for each term. Similar to word types in traditional dictionaries, a "term type" will describe what the term refers to. Examples include: architecture, document format, protocol, and behavior.
2. Provide "related terms" for each term. This is also similar to conjugations/derivations in traditional dictionaries, and it helps route users to the root term. For example, some of the related terms to "agent" are agents and agentic.
3. A simple set of definitions that defines terms as they're commonly used. To do so, the dictionary will provide users with different definitions depending on their background.
4. A diagram! A picture is worth a thousand words, and this dictionary is planning on using diagrams to convey the various elements associated with a term.
5. Finally, there should be informative examples of the term currently in use. This helps ground the definition in further context and improve the dictionary's authority in defining AI buzzwords.

Laying out the Definitions

So, what are the different definitions that will be offered, and why?

• First, there should be a general definition that is understood by the average professional. It should be as independent as possible, without relying on further buzzwords. This definition is the core artifact of The Simple AI Dictionary, so it's meant to be very accessible.
• There should also be a software definition that is understood by the average technical professional. This definition can rely on other buzzwords, as it's meant for professionals with existing familiarity of the GenAI space. It should also follow the general definition as closely as possible, functioning as the "level 2" in terms of technical jargon.