Technology Sharing

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

The nature of the mind: its power seems to arise from the complex interweaving of relationships between those agents

-- Marvin Minsky        

I recently read the book "Mind Society" written by Marvin Minsky, the father of artificial intelligence and Turing Award winner, and I suddenly felt enlightened. He gave a very elegant interpretation of concepts such as human thinking, intelligence, and intelligent agents.

Personally, I think he used the idea of ​​system engineering to build a model of thinking and intelligent bodies. This system engineering idea provides us with a strong systematic ideological foundation for building AI application systems.

thinking

Human problem-solving ability.

intelligent

Our minds contain programs that enable us to solve problems we consider difficult, and intelligence is the name we give to these as-yet-ununderstood programs.

Agent

Thinking can be decomposed, and each piece of thinking is composed of smaller programs. We call these small programs intelligent agents. Each thinking intelligent agent can only do some low-level intelligent things, and we can gather these intelligent agents into a system in a special way. The ability that emerges from this system is real intelligence. This is just like a straight flush in playing poker, which is valuable only when combined together.

Let's study this with a simple example, pick up a cup!

• The grasping agent wants to hold the cup
• The balancing agent wants to prevent the tea from spilling
• The thirst agent wants you to drink tea
• The mobile agent brings the cup to your mouth

These agents collaborate with each other, each completing their own small task, and together they are able to complete a big project: drinking tea.

        

We don’t know if this model of thinking is the true state of human thinking. But based on experience and common sense, our intuition tells us that this model of thinking is reasonable. Similar to any artificially constructed model,The models are all wrong, but they kind of work.

  In this article, we follow the thought of Marvin Minsky and take the idea of ​​system modeling as a guide to try to build an information model of thinking, and finally form the system architecture of AI application.

Agent Model

An intelligent agent is an entity with thinking ability, which has the following characteristics:

    input signal

Signals sent to an agent by other agents.

    output signal

The intelligent agent thought about it, made a judgment, and then sent a signal.

   Enter information

Information received by the agent.

Output information

Information provided by the agent to the outside world.

Mental Model

A mental model consists of a set of interconnected agents and environment information.

Both the external environment and the internal agent may change the environmental information.

Types of Agents

Agents can be divided into:

    Thinking Agent

Thinking agents need to think to achieve their goals. The thinking process is unclear and the results may be uncertain. For example, an agent that realizes reasoning thinking using the Tongguo language model is a thinking agent.

  Control Agent

The procedure for controlling intelligent agents to solve problems is already very clear, and the goals of the intelligent agents can be achieved directly through computer programs. Compared with thinking intelligent agents, controlling intelligent agents are intelligent agents that can achieve their goals without thinking.

Composite Agent

Multiple intelligent agents that are interconnected in a certain form constitute a composite intelligent agent.

Limitations of the model

It should be acknowledged that the thinking model constructed using deterministic connected agents has certain limitations. If the thinking of the human brain is a group of interconnected agents, then the connections between agents are mysterious and unpredictable. So far, people have not figured out what the routing algorithm for information transmission between agents is.

However, this model is useful in solving specific AI applications. In fact, for many problems, humans have developed mature mental processes. We can use mature mental processes to build intelligent agent networks.

summary

Human thinking is accomplished through a series of intelligent agents.

Agents can be decomposed into simpler agents.

Although we currently do not know how agents in the mind connect and transmit information, we can build a certain connection between agents based on human experience and common sense when solving specific tasks.

Intelligent agents can be divided into thinking agents and controlling agents.

The intelligent network haslimitationmental model, but it is useful.