Why understand AI Agents? Because investors are currently only focusing on AI Agent companies, and cutting-edge AI companies like OpenAI are researching it.
AI Agents are a direction for the practical application of AI.
There's a joke in the startup circle: I have a great idea, but I'm just missing a programmer. In the future, AI Agents will be the 'programmers' that help you realize your ideas.
After reading this article, you will understand what an AI Agent is and how it works.
1. What is an AI Agent?
An AI Agent refers to an Artificial Intelligence Agent, an intelligent entity capable of perceiving the environment, making decisions, and executing actions.
For example, an AI Agent is like a Xiao Ai, living in your phone or computer, with intelligence and observational abilities.
When you say to it: "Xiao Ai, I'm feeling a bit unwell."
It will, like magic, observe your condition, body temperature, and recent 24-hour activity trajectory, and combine it with data and information from the internet. After a dazzling series of analyses, it concludes within 1 second that you have tested positive.
Then it actively generates a leave request for you, and with a nod, the leave request is sent to your leader.
It also kindly tells you that your ibuprofen and mineral water at home are running low, and has already selected the products for you. Just give the word, and they will be delivered to your doorstep in 30 minutes.
It perceives that driving is not a good idea right now, so it conveniently calls a car for you to go home, which will arrive in 10 minutes, so hurry up.
This is the result of a series of Agents working together.
2. How does it work so excellently?
A diagram illustrates how an AI Agent works. This diagram describes how an intelligent agent processes, analyzes, and responds to external information.
A bit confusing, right? Let me break it down.
AI Agents are divided into 4 parts:
1. Perception
This is the first step of the process. AI establishes perception of the external world through sensors, cameras, microphones, etc.
Inputs: The perceived information is input into the system. In this example, the input is: "I'm feeling a bit unwell", my body temperature, mental state, sleep duration, etc.
External Environment: The environment or context in which the system exists. For example, the situation of "I'm feeling a bit unwell" may involve weather, environment (such as whether it's in a place with pollen allergies), etc.
2. Information processing (The brain of the Agent)
It can be described as a general large model plus numerous knowledge bases used to process information. It includes the following systems:
1) Information storage-related
Memory system: Includes Storage and Memory, used to store long-term and short-term data.
For example, long-term data includes my basic information, hobbies, underlying health conditions, etc.;
Short-term data, such as there being only one bottle of mineral water left at home, can be deleted after making a purchase.
Knowledge Base (Knowledge): Includes medical knowledge bases, product databases, etc., used to diagnose my current condition and manage subsequent treatment and daily needs.
2) The large model processes information.
Based on perceived information (input + Environment), memory, knowledge base, and other information, it processes and derives conclusions (Decision Making): "I'm sick, and I have tested positive for COVID."
3) Then formulate the next step plan (Planning).
Action/Reasoning is based on the specific actions of its decisions but has not yet been implemented.
To help me write a leave request, buy medicine, buy water, call a car, etc.
3. Execution
Based on a series of dazzling operations from the Brain, a conclusion is reached, and the next step plan is formulated, which then needs to be executed (Action).
The large model itself cannot complete these tasks and needs to call external tools.
At this point, third-party tools (Tools and Calling API) are invoked to interact with other apps through interfaces or applications to achieve the final effect.
4. Output
After execution, a feedback mechanism is needed to inform you of the results. For example, my Xiao Ai tells you: "You tested positive, and I've already prepared a leave request and called a car for you."
This is the working principle of AI Agents.
In summary, this system describes a simplified model that shows how an AI agent begins with perceived information, undergoes a series of internal processing and decision-making, and ultimately makes a response.
3. Summary
AI Agent is one of the directions for the future development of AI (the other direction will be discussed in the next article).
It can be a personal assistant or a helper at work, amplifying your capabilities. It fills in your weaknesses, making you a super individual.
