课程: Understanding Artificial Intelligence Concepts and Terminology with ISO/IEC 22989:2022

AI agents (Clause 5.3)

课程: Understanding Artificial Intelligence Concepts and Terminology with ISO/IEC 22989:2022

开始一个月试用 购买团队方案

AI agents (Clause 5.3)

“

- The core idea behind artificial intelligence is simulating human intelligence and behavior. This is accomplished by programming artificial intelligent agents with human prescribed goals which they can independently choose the best course of action to meet. Imagine you're a player in a video game and the game world is your environment. You, as the player, are the agent and everything around you like the landscapes, obstacles, or other players represent the environment you must navigate. AI works similarly. It's the agent operating within its environment that makes choices based on what's happening around it, and it's trying to achieve specific goals that have been predetermined for a successful result. The programmers goals ultimately determine the agent's behavior and action. The agent deconstructs the goals into actionable task. The agent acquires information through sensing in an environment, and then with sufficient data or information, the agent implements task through actions. This is accomplished through various types of sensors and actuators. Agent types are deployed and developed based on business objectives. Here are some agent types. Reflex agents are confined to specific predefined rules in a closed environment. As an example, an agent in a voice activated environment allows a user to set up a meeting with someone in their contact list when they speak their name. Model agents can evaluate probable outcomes and consequences before deciding what action to take. They can build an internal model of the environment they perceive and make enhanced decision making. Imagine the voice activated AI meeting invite agent can review previous meetings with the selected contact, create an invite based on statistical data of open projects, and create an agenda from that data to discuss in the coming meeting. Goal-based agents, also known as rule-based agents, are more robust in reasoning. In addition to evaluating the data, they compare different approaches to achieve desired outcomes by selecting the most efficient path. Now, imagine the AI meeting invite agent considering the optimal meeting schedule. It could be based on the best day of the week, month, or year, or the frequency in accordance with participant availability and productivity to accomplish the business objectives. Utility-based agents utilize complex reasoning to analyze different scenarios and their corresponding usefulness and select the path with the most rewards or awards for the user. Now, our AI meeting invite agent can produce agenda items that focused on under explored or unexplored revenue generating opportunities with the selected contact. Learning agents maintain a continuous learning path based on previous experiences using sensory inputs, they modify future actions to improve results. Our AI meeting invite agent not only takes notes on the meeting, but also has input from job trackers to seek what future accomplishments need to happen. Finally, hierarchical agents are arranged interconnectedly in tiers. Higher tiered agents behave as supervisors breaking complex larger tasks into smaller ones and assigning them to lower level agents who report on progress back to the supervisor agents. Higher level agents collect the results and coordinate additional tasks to ensure goal achievement. Imagine taking the aforementioned use cases of an AI meeting scheduling agent and breaking the different abilities into coordinated tasks, supervised by higher level agents. Interactions with agents to produce these results are called a agentic AI. It is characterized by perceiving data from different multiple sources, reasoning with, for instance, a large language learning model, acting upon the reasoning and learning to provide feedback for future action. The AI agent takes action based on data, information, and knowledge, but this knowledge is not the type we are familiar with outside the arena of computer science. Let's get to that next.

内容