AI FLUENCY · PHASE 2: APPRECIATION
Module 9 — Automation, Augmentation, and Agency
What Is AI Agency, and How Can AI Systems Act Toward Goals?
Discover how AI systems can plan, use tools, and take connected actions toward assigned goals.
What changes when AI can act instead of only answering?
Introduction: From Assistance to Action
Artificial Intelligence can support human activities in different ways.
In Lesson 033, we explored AI automation: technology performing specified tasks with reduced direct human involvement.
In Lesson 034, we explored AI augmentation: AI helping people understand, think, create, analyze, and perform activities more effectively.
Now consider a different situation.
Instead of asking AI to explain how to organize a workshop, imagine assigning an AI system the objective of preparing a workshop plan.
An appropriately equipped system might check information, compare options, prepare a schedule, and draft invitations.
What happens when an AI system can pursue an assigned goal through several connected actions?
This introduces the concept of AI agency, often discussed through AI agents and agentic AI systems.
1. What Is AI Agency?
AI agency describes an artificial system's ability to select or execute actions toward a defined objective, within its capabilities and constraints.
An AI agent may identify intermediate steps, use available tools, observe the results, and determine what to do next.
Not every AI agent has the same capabilities.
Some agents handle narrowly defined activities. Others can coordinate more complex sequences of permitted actions.
What should be accomplished?
↓
What actions might be needed?
↓
Use available tools within assigned permissions
AI agency enables systems to undertake goal-directed actions within defined capabilities, permissions, and constraints.
AI agency does not automatically imply independent authority, consciousness, or moral responsibility.
2. Automation, Augmentation, and Agency
These three concepts describe different ways AI can contribute to human activities.
Automation
Execute a Task
Technology performs specified activities with reduced direct human involvement.
Example: Send a reminder at a scheduled time.
Augmentation
Assist a Person
AI helps people perform activities while they remain actively engaged.
Example: Suggest possible dates for a meeting.
AI Agency
Pursue an Assigned Goal
AI carries out connected steps toward an objective within assigned boundaries.
Example: Check calendars, propose times, and prepare invitations.
The Important Difference
Automation emphasizes performing specified tasks.
Augmentation emphasizes supporting a person's activity.
AI agency emphasizes goal-directed sequences of actions.
These categories can overlap. An AI agent may use automated workflows and also augment a person's capabilities.
Automation: Execute a Task.
Augmentation: Assist a Person.
AI Agency: Act Toward a Goal.
3. Five Elements of AI Agency
Many agentic systems can be understood through five important elements.
🎯
1. Goal
The assigned objective provides direction for the system.
🗺️
2. Planning
The system selects or identifies steps toward the goal.
🛠️
3. Tools
Authorized software functions allow information retrieval or permitted actions.
🔄
4. Feedback
Results from actions can influence subsequent steps.
🛡️
5. Boundaries
Permissions, safeguards, and oversight constrain what the system may do.
These are conceptual elements, not a requirement that every AI agent must implement an identical architecture.
Some systems use explicit step-by-step plans. Others select their next action dynamically based on available information.
Human-defined boundaries should apply throughout the process rather than only at the end.
4. Everyday Example: Organizing a College Workshop
Imagine a college coordinator organizing an AI awareness workshop for students.
Several activities must be coordinated: choosing a date, checking venue availability, preparing a schedule, and inviting participants.
An appropriately equipped AI agent could help perform these connected activities.
The Assigned Goal
"Prepare a proposed plan for an AI awareness workshop, including suitable dates, a venue, a schedule, and draft invitations. Present it for my approval."
How the Agent Might Proceed
Prepare the workshop plan
↓
Use permitted booking information
↓
Check authorized calendars
↓
Organize proposed sessions
↓
Prepare communications without sending them
↓
Coordinator reviews and authorizes next actions
The system is not merely answering a single question.
It is potentially performing a sequence of connected activities toward the assigned objective.
What Remains Human?
The coordinator defines the workshop's purpose, determines priorities, sets permissions, reviews the proposal, and approves important decisions.
The agent must not be assumed to have access to calendars, venues, or email systems unless those tools have actually been connected and authorized.
AI can coordinate selected actions. Humans retain authority over consequential decisions.
5. The AI Agency Framework
A simple framework helps explain how goal-directed AI activity may work.
What outcome is required?
↓
What steps could help?
↓
What tools may the system use?
↓
What happened, and what next?
↺ Planning and action may repeat
Monitor, approve, correct, or stop actions
The framework highlights that agentic activity may involve repeated planning and action.
Oversight is not necessarily a single final step.
Appropriate safeguards and limits should operate throughout execution, with approval checkpoints before important actions.
The Central Idea
Goal → Planning → Authorized Actions → Feedback → Oversight
6. Does AI Agency Mean Human Agency?
No. The words are related, but they describe different concepts.
In Lesson 031, we discussed human agency: the ability of people to pursue goals, make meaningful choices, exercise judgment, and retain responsibility.
In Lesson 035, AI agency refers to goal-directed behavior by an artificial system.
| Dimension | Human Agency | AI Agency |
|---|---|---|
| Goals | People establish meaningful purposes | Systems pursue assigned objectives |
| Decisions | People exercise judgment | Systems select actions using computational processes |
| Authority | Human and institutional roles | Limited by assigned permissions |
| Accountability | People and institutions remain accountable | Requires human or organizational accountability arrangements |
A system can exhibit goal-directed behavior without demonstrating human-like intentions or conscious experience.
The question of whether AI can have subjective experience remains a subject of scientific and philosophical debate.
AI agency should therefore not be confused with human moral agency.
An AI system acting toward a goal does not mean it has become a human decision-maker.
7. Why AI Agency Requires Human Oversight
Agentic AI introduces important possibilities because systems can undertake several actions toward an assigned objective.
But connected actions also introduce risks.
An AI agent may misunderstand its goal, rely on incorrect information, select an unsuitable action, or repeat mistakes across multiple steps.
Six Important Safeguards
1. Clear Goals
Define the intended outcome and relevant constraints.
2. Limited Permissions
Provide access only to necessary tools and information.
3. Approval Checkpoints
Require human authorization before consequential actions.
4. Monitoring
Track important activities and unexpected behavior.
5. Verification
Check whether completed actions actually satisfy the goal.
6. Ability to Intervene
Allow authorized people to correct, pause, or stop execution.
These safeguards should be proportionate to the consequences of the activity.
For example, drafting a workshop invitation has different risks from actually sending messages, spending money, or changing official records.
Delegating actions to AI does not mean delegating away human responsibility.
Key Takeaway
AI agency enables systems to pursue assigned goals through connected actions.
Such actions depend on available tools, permissions, constraints, and system capabilities.
Delegate Actions. Preserve Human Control.
Quick Knowledge Check
Test your understanding before continuing to the final lesson of Module 9.
Question 1
What best describes AI agency?
A. Systems pursuing assigned goals through connected actions
B. Every AI system possessing human consciousness
C. Technology following only one fixed instruction
Reveal Answer
A. AI agency involves goal-directed action within system capabilities and constraints.
Question 2
Which example best illustrates agentic AI?
A. Sending a receipt after payment
B. Checking calendars, comparing dates, and preparing a workshop plan toward an assigned goal
C. Displaying a static webpage
Reveal Answer
B. This involves several coordinated steps toward a goal, assuming the system has suitable tools and authorization.
Question 3
Which of the following is not a core element of the framework introduced in this lesson?
A. Goal
B. Feedback
C. Human-like consciousness
Reveal Answer
C. AI agency does not require demonstrating human-like consciousness.
Question 4
Why do AI agents need permissions and safeguards?
A. To define and limit authorized actions
B. Because every AI action is automatically correct
C. To eliminate human accountability
Reveal Answer
A. Permissions and safeguards help control what a system can access and do.
Question 5
Does delegating actions to an AI agent remove human responsibility?
A. Yes, always
B. No, people and organizations remain accountable
C. Only when the agent uses multiple tools
Reveal Answer
B. AI systems can undertake authorized actions, but accountability must remain appropriately assigned to people and institutions.
Think About It
Imagine delegating the preparation of a college workshop to an AI agent.
Which actions could the agent perform independently, and which should require human approval?
Your AI Agency Reflection
- Describe the goal you would assign.
- Identify the steps the agent might undertake.
- List the tools or information it would require.
- Specify which actions should require your approval.
- Describe what could go wrong.
- Explain how you would verify successful completion.
Prepare a short AI Agency Blueprint: Goal, Possible Actions, Tools, Permissions, Human Approval, and Verification.
At this Appreciation stage, the objective is to understand the opportunity and responsibilities of agentic AI. Building agents belongs to later stages of the AI Fluency journey.
What Have You Learned?
After completing Lesson 035, you should be able to:
- Explain AI agency in simple terms.
- Distinguish AI agency from automation and augmentation.
- Identify goals, planning, tools, feedback, and boundaries as useful elements of agentic systems.
- Recognize examples of goal-directed AI activity.
- Explain how human agency differs from AI agency.
- Recognize why agentic AI needs permissions, monitoring, and safeguards.
- Explain why human accountability remains important when AI takes actions.
PHASE 2 — APPRECIATION
Module 9 — Automation, Augmentation, and Agency
You have now explored all three foundational concepts.
✓ Lesson 033 — AI Automation
How can AI perform selected tasks with reduced human involvement?
✓ Lesson 034 — AI Augmentation
How can AI enhance human capabilities?
● Lesson 035 — AI Agency
How can AI systems pursue assigned goals through connected actions?
Lesson 036 — Coming Next
How do automation, augmentation, and agency work together?
COMING NEXT · LESSON 036
How Do Automation, Augmentation, and Agency Work Together?
We have explored three different possibilities.
Automation can reduce direct human effort in selected tasks.
Augmentation can support people in performing activities.
AI Agency can enable systems to undertake connected actions toward assigned goals.
But these capabilities do not always operate in isolation.
In Lesson 036, we will explore how they can contribute differently within the same human-directed activity.
How can we recognize the right role for AI in different situations?
Your AI Fluency Journey
Understand AI
COMPLETE
Discover Possibilities
CURRENT PHASE
Use AI Effectively
Achieve Meaningful Outcomes
Continue Your Learning
Previous Lesson — AI Augmentation
Previous Lesson — AI Automation
Phase 1 — Awareness
New to AI Fluency? Explore six foundational modules and 24 introductory lessons.
AI Fluency Master Roadmap
Further Learning Resources
NIST — AI Risk Management Framework
Stanford Institute for Human-Centered Artificial Intelligence
Delegate Actions. Preserve Human Control.
Agentic AI can create new possibilities for coordinating activities toward meaningful goals.
Its value depends on appropriately defined objectives, authorized tools, reliable operation, and effective safeguards.
Human Purpose. AI Action. Human Oversight.
No comments:
Post a Comment