Skip to content

Lesson 2: What is an agent?

By the end of this lesson:

You will:

  • Know what an agent (or AEA) is (and is not)
  • Be familiar with some of the primary types of components an agent can be made from

What is an Autonomous Economic Agent (AEA)?

An AEA is...

An intelligent agent that acts on its owner's behalf, with limited or no interference, whose focus is on generating economic value for its owner.

Let's break down the term Autonomous Economic Agent (AEA):

  • Agent: An AEA is first and foremost an agent, representing an individual, organisation or object (a.k.a. its "owner") in the digital world. An AEA looks after its owner's interests and has their preferences in mind when acting on their behalf.
  • Autonomous: AEAs act independently of constant input from their owner and autonomously execute actions to achieve their goals.
  • Economic: AEAs have a narrow and specific focus: creating economic value for their owner.
Some of the other characteristics AEAs have:
  • Proactive: AEAs are proactive; they take the initiative and perform actions that take them closer to their goals.
  • Reactive: AEAs are also reactive; they are aware of the environment they are in, perceive changes in the environment, and react to these changes in accordance to their goals.
  • Self-interested: An AEA primarily looks after its own interests (which is aligned with those of its owner) and not necessarily the interests of other agents or the larger system.

What is NOT an AEA?

  • Any agent: AEAs are NOT meant to address any needs their owners might have. They have a clear and well-defined focus, which is generating economic value for their owner and this is manifested in a variety of different ways in their design.
  • Digital twins: An AEA is NOT it's owner's twin in the digital world; i.e. mirroring their preferences, values, and priorities. An AEA can be given whatever preference, value, and priority its owner wants them to have in the digital world.
  • APIs or Sensors: These do NOT have any agency. They just "sit there" and respond to requests/changes in the environment.
  • Smart contracts: Similar to APIs and sensors, smart contracts do NOT display any proactiveness; they are purely reactive to external requests (in their case, contract calls and transactions).
  • Artificial General Intelligence (AGI): AEAs have a well-defined, narrow, and goal directed focus that involves some economic gain.

Agents and AEAs

In the rest of the course, unless specified, we will use the terms AEA and agent interchangeably to mean AEA as defined above.

Agent Components

AEAs can be made from various components, much like legos, and these components can be of differing types. Below are some of the more important types of components an agent can have.

Skill

A Skill is an isolated, self-contained, (and preferably atomic) functionality that AEAs can take on board to expand their capability. Skills contain the proactive and reactive behaviour that ultimately makes it possible for an AEA to deliver economic value to its owner.

Protocol

A Protocol defines the structure and nature of an interaction that can happen between agents, or between components of an agent. You can think of a protocol as the language that two agents speak and a skill for this protocol as a particular way of speaking this language. From a game-theoretic viewpoint, a protocol defines the rules of a game and a skill for this protocol defines a particular strategy for playing this game.

Connection

Connections act as interfaces between an agent and the outside world. As such, a connection allows the agent to communicate with some entity outside of it, for example, another agent, a traditional HTTP server, a database, a reinforcement learning training environment, a blockchain, etc.

This video summarises some of what was covered in this lesson: