Multi-Agent Communication: Negotiation and Trading

Multi-Agent Communication: Negotiation and Trading

In multi-agent systems, agents communicate with each other to reach agreements, solve problems, and make optimal decisions. These processes are carried out through mechanisms such as negotiations and auctions.

​

Types of negotiation:�

• Cooperative Negotiation:�Agents collaborate to reach mutually beneficial agreements.�Example: Robots working together to complete a cargo transport task.
• Competitive Negotiation:�Each agent acts to protect only its own interests.�Example: Multiple companies competing for limited resources.

​

Cooperative Negotiation

Cooperative Negotiation in multi-agent systems is a process in which agents collaborate to reach mutually beneficial agreements.

In this approach, agents:

• Strive to find solutions that benefit all parties involved.
• Use flexible strategies to reach an agreement.
• Exchange information and maintain transparency throughout the negotiation.

Key Characteristics of Cooperative Negotiations:

• Mutual Benefit: Agents aim to achieve agreements that are advantageous for all participants.
• Transparency: Information exchange is open and trustworthy.
• Compromise: Each agent seeks to preserve its own goals as much as possible while making concessions for the common good.
• Adaptive Strategies: Agents can modify their strategies during the negotiation process.

​

Competitive Negotiation

Competitive Negotiation is a process in which agents compete with each other to acquire resources, complete tasks, or gain other advantages. In competitive negotiations, each agent makes decisions based on its own interests, aiming to achieve its goals while still contributing, indirectly, to the overall objective.

Key Characteristics of Competitive Negotiations:

1. Competition:�Agents compete with each other over resources, tasks, or other items.�Example: On an e-commerce platform, multiple seller agents check prices for the same product and conduct independent sales.
2. Self-Interest Presentation:�Each agent presents its own capabilities and constraints, but may not be fully truthful.�Example: An agent might understate its available resources to secure better terms.
3. Strategic Actions:�Agents perform strategic moves to achieve an optimal outcome.�Example: In an auction, agents may raise or lower bids to outmaneuver other agents.
4. Results:�The agent that secures the best terms benefits exclusively. Other agents, having not obtained the desired advantage, must wait for future opportunities.

​

Main stages of negotiation:

1. Offer:�One agent makes an offer to another agent. The offer can involve resources, tasks, or other items.�Example: “I offer you 10 resources, but you must also fulfill my requests.”
2. Counteroffer:�The other agent responds to the offer and may propose a different one.�Example: “I can accept your offer, but I need to receive 5 resources in return.”
3. Terms Confirmation:�The overall terms are confirmed, and the transaction is executed.�If one party does not accept the terms, the negotiation continues or ends.
4. Outcome:�In competitive negotiations, the agent that secures the best terms benefits from the agreement.

​

Negotiation Strategies:

• Negotiation Strategies:
• Real-Time Evaluation:�Agents adjust their decisions over time based on the evolving situation.
• Iterative Concessions:�Each party gradually relaxes its demands to move toward an agreement.
• Rule-Based Negotiation:�Agents negotiate according to predefined rules and guidelines.
• Examples:
• E-Commerce Platforms: On platforms like Amazon or eBay, sellers and buyers negotiate prices, delivery terms, and other conditions.
• Energy Market: Agreements are made between electricity producers and consumers on prices and quantities of energy.
• Autonomous Transport Systems: Multiple autonomous vehicles communicate to coordinate traffic rules and reduce congestion.

Difference Between Competitive and Cooperative Negotiations:

Languages and Protocols for Message Exchange

The ability of an agent to communicate (exchange messages) with other agents is essential for its effective operation within a multi-agent community. Agents interact through the following types of messages (performative acts):

• Informational Messages: Provide data to other agents, for example, notifying that a script has completed to synchronize with another agent’s task.
• Directive Messages: Instruct another agent to perform a specific action, for example, an "air traffic controller" agent directing a "Pilot" agent to modify certain flight parameters.
• Commitment Messages: Represent obligations, for example, a "flying" agent committing to follow instructions from other similar agents or complying with the constraints given by an "air traffic control" agent.
• Declarative Messages: Communicate facts intended to motivate the receiving agent to perform certain actions.

​

The foundation of agent communication is based on the speech act theory, originally developed to analyze human interactions. Human speech acts are viewed as actions that perform functions such as requests, offers, confirmations, and responses.

KQML (Knowledge Query and Manipulation Language) is a formal language for describing message formats (“conversational acts”), developed in 1990 under DARPA’s guidance. It includes:

• A language to define message formats, and
• A set of protocols that support communication processes using this language.

Agent messages can also be described in other formal languages. To resolve ambiguities in understanding agent messages, KQML introduces special descriptors called performatives, which indicate the type of message.

• Examples of performatives:
• Promise
• Request
• Tell
• Demand

These performatives define the intended action or purpose of the message, enabling clear and unambiguous communication between agents.

What is KQML?

KQML (Knowledge Query and Manipulation Language) is a communication language and protocol designed for agents in multi-agent systems to exchange messages and conduct negotiations. It is used to send and receive messages between agents in a standardized format.

Structure of KQML Messages:

1. Performative (Speech Act): Indicates the purpose of the message, e.g., ask-if, tell, subscribe.
2. Content: The main information or query being sent.
3. Additional Parameters: Fields providing detailed information about the message, such as sender, receiver, ontology, and others.

Example Format of a KQML Message:

(ask-if

:sender agent1

:receiver agent2

:ontology weather

:language Prolog

:content (temperature London ?t)

)

​

KQML - communication protocol

In multi-agent communication, message protocols are described at several levels:

• Lower (Transport) Level: Specifies the mechanism for sending messages, the message type, and the names of the sender and receiver.
• Intermediate Level: Defines the syntax of the message (see below).
• Higher Level: Describes the semantics (meaning) of the message.

A protocol is described using a data structure that includes the following fields:

• Sender: The agent sending the message.
• Receiver: The intended recipient agent.
• Protocol Language: The language used to structure the communication.
• Encoding and Decoding Functions: Methods for encoding outgoing messages and decoding incoming ones.
• Messages: The actual messages being exchanged.
• Actions to be Performed by the Receiver: The operations or responses the receiving agent is expected to carry out.

In the KQML protocol, there are several performatives that indicate the way agents communicate with each other.

Trends and Prospects

1. Integration with Artificial Intelligence:�MAS are increasingly combined with AI techniques such as machine learning, deep learning, and natural language processing to create smarter, more adaptive agents.
2. IoT and Smart Environments:�Multi-agent systems are being applied in Internet of Things (IoT), smart homes, and smart cities for autonomous coordination of devices and services.
3. Autonomous Vehicles and Robotics:�MAS play a key role in autonomous transport systems, drone coordination, and collaborative robotics for efficient task allocation and safety.
4. Cybersecurity Applications:�Multi-agent systems are used to detect intrusions, manage threats, and maintain secure communications in dynamic network environments.

​

5. Blockchain and Decentralized Systems:�Integration of MAS with blockchain enables secure, transparent, and decentralized coordination among agents.

6. Cloud and Edge Computing:�MAS can manage distributed computing resources in cloud and edge environments for optimized performance and resource allocation.

7. Future Prospects:

• Enhanced agent collaboration and negotiation mechanisms.
• More sophisticated reasoning and learning capabilities.
• Wider adoption in industrial automation, healthcare, finance, and e-commerce.

​