How AI Agents Are Changing Lives?

AI agent is becoming the new AI buzzword. As the development direction of AI applications that many people are optimistic about, AI agents have attracted much attention.

Here comes the AI agent

Not long ago, OpenAI laid out a five-level roadmap to achieve artificial general intelligence: L1 is a chatbot; L2 is the reasoner, i.e., the AI that can solve problems like a human; L3 is an agent, an AI system that can not only think, but also act. L4 is an innovator; L5 is the organizer.

What exactly is an AI agent? "Generally speaking, an AI agent is like a 'little assistant' with intelligence, emotional intelligence, understanding, and help." Chen Hao, deputy director of the Advanced Technology Center of the Beijing Institute of General Artificial Intelligence, said.

What is the difference between an AI agent and a large model?

Pu Zhiqiang, a researcher at the Institute of Automation of the Chinese Academy of Sciences, said that AI agents can be understood as more three-dimensional, "human-like" intelligent systems. In addition to providing language communication that is widely used by large models, AI agents can also complete more complex tasks.

For example, if you give the task command "Buy coffee", the large model will tell you "I can't buy coffee directly for you" and give other suggestions; The AI agent will first disassemble how to buy coffee, and draw up steps such as placing an order and paying on behalf of an app, and then call the app to select takeout according to these steps, and then call the payment program to place an order and pay, without the need for the person who gave the instruction to specify each step of the operation.

At present, AI agents have been applied in many scenarios, such as customer service, programming, content creation, knowledge acquisition, finance, investment research, mobile phone assistants, industrial manufacturing, etc.

On the Baidu Wenxin Yiyan app, everyone can create and display their own AI agents, conduct "video conversations", memorize words, correct spoken language, and can also simulate interviews and dialogue with AI ancients through AI agents.

Resume can be scored and optimized in seconds; Generate multiple styles of copywriting; After teaching the work content to the AI agent, you can help yourself complete part of the work, which is equivalent to having your own intern......

In January this year, the personalized agent customization function of Zhipu was launched, and users can create their own AI agents with simple prompt word commands. According to the relevant person in charge of Zhipu, there are hundreds of thousands of users in the intelligent body center, and the average number of questions per day exceeds 500,000.

What changes will be brought about

With the rapid development of AI technology, many people have used it more as a "tool" rather than a "toy" to improve work efficiency in their daily lives.

More and more AI agents are being applied to all walks of life, which will bring about an improvement in productivity and a quiet transformation in the way enterprises operate and manage.

Pu Zhiqiang noticed that the enthusiasm of the organization to deploy "digital employees" is high, the application scenarios are rich, and the tasks are becoming more and more complex.

Some market segments have seen disruptive changes. For example, in the content creation industry, AI agents can generate videos in one sentence, or use pipelines to generate short videos with more complex content through the collaboration of multiple large models. In literary creation, the generation of scripts and ideas through AI agents has been accepted by more practitioners.

"The wide application of AI agents can reduce labor costs. For ordinary people, it is equivalent to a lot of helpers; For enterprises, it is also an opportunity for intelligent transformation. Zhou Ming, CEO of Lanzhou Technology, said.

Chen Hao said that the general AI agent can actively participate in each link, rather than passively performing preset tasks. This not only helps to increase production efficiency, but also to better respond to unexpected situations and increase flexibility in production and service.

In addition, AI agents also play a more active role in solving social problems such as aging. According to Chen Hao, the "embodied intelligent robot" and "high-density tactile dexterous hand" of the Beijing Institute of General Artificial Intelligence can recognize the environment and adjust the action strategy according to the environment, without the need for additional instructions from people.

"AI intelligent physical fitness can assist the daily life of the elderly, which will be a good helper for home care. In addition, in places that cannot be reached by people, such as at the site of a major natural disaster, AI agents can be used to carry out dangerous operations or rescue tasks, greatly reducing casualties. Chen Hao said.

At the same time, the development of AI agents can also open up space for more enterprises to innovate and form a more complete AI industry chain.

As of August this year, the number of developers on Baidu's Wenxin intelligent twins platform has reached 600,000, and the number of enterprises has risen to 100,000. He Junjie, general manager of Baidu's mobile ecosystem business group, said that AI agents will reshape the way people interact with technology, bringing new application ecosystems, traffic patterns and business models.

How to balance "wisdom" and "governance"?

Many people in the industry believe that AI agents are the future trend.

According to the "2024 Digital Technology Frontier Application Trends" report, AI agents are expected to become the next generation of platforms. Robin Li, founder of Baidu, also said that he is most optimistic about the development direction of AI applications in the future.

There is a view that in the foreseeable future, there will be an AI agent on the smartphone, which will collect all kinds of data to meet the needs of users anytime, anywhere; It will be the "AI programmer" of the enterprise, assisting developers with a series of tedious tasks; Or a "data analyst", who automatically collects and analyzes enterprise data and mines the value of data gold.

Robin Li said that as the basic model becomes more and more powerful, it is becoming easier to develop applications. As long as the workflow is explained clearly in "human words", coupled with a proprietary knowledge base, you can make a valuable AI agent, which is easier than making a web page in the Internet era.

Many interviewed experts believe that the AI agents that have emerged can only complete relatively simple and fixed work, and are still in the initial stage of development.

At the same time, AI agents are also facing technical risks, ethical and privacy issues.

Industry insiders said that AI agents will collect a large amount of data in the process of providing services, and it is necessary to ensure data security and prevent privacy leakage. AI agents can provide advice, but not all of them are authoritative. In addition, the more autonomous an AI agent is, the more likely it is to make unpredictable or inappropriate decisions in complex or unforeseen situations.

Zhou Ming believes that security barriers should be set up from different dimensions: in the R&D process, security should be taken as the core design principle, and corresponding training should be provided for R&D personnel to prevent bias in model algorithms; In the use process, enhance the public's awareness of the boundaries of technology application, risks and challenges, and increase feedback channels.

The relevant person in charge of Zhipu said that enterprises should formulate strict internal specifications and operating procedures to ensure that development, testing, deployment and use are all within a controllable range. Promote the formulation of industry standards and provide guidance for the healthy development of AI agents. At the same time, it is necessary to establish a responsibility traceability mechanism.

Zhang Haifeng, an associate researcher at the Institute of Automation, Chinese Academy of Sciences, also believes that policy formulation should incorporate the opinions of multiple stakeholders, establish an appropriate AI ethics review mechanism, and conduct independent reviews of high-risk projects to ensure that they meet ethical standards and are beneficial to society.