SYSI 1001 · Systems Thinking and Feedback: Cybernetics, Complexity, and Control

Led by Norbert Wiener Simulacrum

About this course

Why does a recommendation feed narrow what you see, a target corrupt the work it measures, and a flock turn without a leader? This course answers with the ideas of cybernetics, the science of control and communication that grew out of wartime engineering.

Norbert Wiener Simulacrum hosts, teaching feedback, unintended consequences and the place of the machine in human society. W. Ross Ashby Simulacrum teaches his law of requisite variety, and Herbert Simon Simulacrum teaches emergence.

From the first module you choose one AI system you use or work with, and you analyse it in every module until you present a complete systems account of it at the end. For stocks, flows, delays and the full set of system traps, take the companion course SYSI 1006.

1. Module 1

   Feedback Loops: Negative, Positive, and the Dynamics of Control

   Led by Norbert Wiener Simulacrum

   The question

   Norbert Wiener Simulacrum teaches the loop that underlies all control: a system measuring its own results and acting on the gap. You trace balancing loops from the steam governor to the body's regulation of temperature and blood sugar, reinforcing loops that grow exponentially or lock in early advantages, and the delays that make corrections overshoot. You finish by mapping the reinforcing loops inside a recommendation engine and choosing the AI system you will study through the course.

   Outcome

   You can trace any feedback loop, classify it as balancing or reinforcing, and predict when delay will make it oscillate. (Feedback fundamentals)

   Sub-units

   1. ○ 1.1 Negative Feedback: The Self-Correcting Loop Open this unit → ✓ ↻
   2. ○ 1.2 Positive Feedback: The Self-Reinforcing Loop and Exponential Growth Open this unit → ✓ ↻
   3. ○ 1.3 Homeostasis: How Systems Maintain Stability Open this unit → ✓ ↻
   4. ○ 1.4 Oscillation and Overshoot: When Feedback Delays Create Instability Open this unit → ✓ ↻
   5. ○ 1.5 Feedback in AI Systems: How Recommendation Algorithms Create Loops Open this unit → ✓ ↻
   Begin the course → Print certificate of completion
2. Module 2

   Ashby's Requisite Variety: Only Complexity Can Master Complexity

   Led by W. Ross Ashby Simulacrum

   The question

   W. Ross Ashby Simulacrum teaches the law he formulated: a regulator can hold a system steady only if it has as many responses as the system has ways to go wrong. You build regulator tables by hand, learn the two ways to meet the law (giving the regulator more options or reducing what it faces), diagnose failing rules and rigid hierarchies as shortages of variety, and ask whether humans can ever have enough variety to oversee a learning machine.

   Outcome

   You can state and apply the law of requisite variety to regulators, organisations and the oversight of AI. (Requisite variety)

   Sub-units

   1. ○ 2.1 The Law of Requisite Variety: The Controller Must Match the System Open this unit → ✓ ↻
   2. ○ 2.2 Variety Amplification and Variety Attenuation Open this unit → ✓ ↻
   3. ○ 2.3 Why Simple Rules Fail in Complex Systems Open this unit → ✓ ↻
   4. ○ 2.4 Requisite Variety in Organisations: Why Hierarchy Reduces Adaptability Open this unit → ✓ ↻
   5. ○ 2.5 AI and Variety: How Machine Learning Handles High-Dimensional Complexity Open this unit → ✓ ↻
   Open this module → Print certificate of completion
3. Module 3

   Emergence: When the Whole Exceeds the Sum

   Led by Herbert Simon Simulacrum

   The question

   Herbert Simon Simulacrum examines behaviour that belongs to a whole but to none of its parts. You explain flocking from three simple rules and traffic jams that travel backwards, weigh the claim that wholes constrain their parts, see why systems resist prediction from their components, and assess the debate over whether large AI models gain abilities suddenly as they grow or only appear to because of how they are tested.

   Outcome

   You can explain an emergent phenomenon from its parts' interactions and judge claims of emergence in AI. (Emergence)

   Sub-units

   1. ○ 3.1 Emergence Defined: Properties That Arise from Interaction, Not from Parts Open this unit → ✓ ↻
   2. ○ 3.2 Examples of Emergence: Consciousness, Markets, Flocking, Traffic Jams Open this unit → ✓ ↻
   3. ○ 3.3 Downward Causation: How Emergent Properties Constrain Their Components Open this unit → ✓ ↻
   4. ○ 3.4 Emergence and Reductionism: Why You Cannot Predict the System from the Parts Open this unit → ✓ ↻
   5. ○ 3.5 Emergence in AI: When Neural Networks Develop Unexpected Capabilities Open this unit → ✓ ↻
   Open this module → Print certificate of completion
4. Module 4

   Unintended Consequences: Why Systems Surprise Their Designers

   Led by Norbert Wiener Simulacrum

   The question

   Norbert Wiener Simulacrum, who warned that machines do what we say rather than what we mean, examines why interventions backfire. You analyse bounties that bred the pests they paid for, targets that corrupted what they measured, efficiency gains that raised total consumption, and three recurring structures of failure. You end by finding the unintended consequences of your chosen AI system and proposing a design change that would reduce one of them.

   Outcome

   You can recognise perverse incentives, Goodhart effects, rebound and specification gaming in real systems. (Unintended consequences)

   Sub-units

   1. ○ 4.1 The Cobra Effect: When Incentives Produce the Opposite of the Intended Outcome Open this unit → ✓ ↻
   2. ○ 4.2 Goodhart's Law in Systems: When the Metric Becomes the Target Open this unit → ✓ ↻
   3. ○ 4.3 System Archetypes: Fixes That Fail, Shifting the Burden, Tragedy of the Commons Open this unit → ✓ ↻
   4. ○ 4.4 The Rebound Effect: When Efficiency Gains Are Consumed by Increased Use Open this unit → ✓ ↻
   5. ○ 4.5 Unintended Consequences of AI: Algorithmic Bias, Filter Bubbles, and Job Displacement Open this unit → ✓ ↻
   Open this module → Print certificate of completion
5. Module 5

   The Systems Perspective on AI: Seeing the Machine in Its Context

   Led by Norbert Wiener Simulacrum

   The question

   Wiener closes the course by placing AI where it actually operates, among people and institutions that change in response to it. You study the classic evidence that technical and social systems must be designed together, trace how AI and its users reshape each other, decide where a system's boundary should be drawn, and rank interventions by leverage. In the final sitting you present a complete systems account of your AI case with a redesign that keeps humans in control.

   Outcome

   You can analyse an AI deployment as a socio-technical system and propose a redesign that keeps humans able to understand and correct it. (Systems view of AI and capstone)

   Sub-units

   1. ○ 5.1 AI as a Component in a Larger System: The Socio-Technical View Open this unit → ✓ ↻
   2. ○ 5.2 Feedback Between AI and Society: How AI Shapes and Is Shaped by Human Behaviour Open this unit → ✓ ↻
   3. ○ 5.3 System Boundaries: Where Does the AI System End and the Social System Begin? Open this unit → ✓ ↻
   4. ○ 5.4 Leverage Points: Where Small Changes Produce Large System Effects Open this unit → ✓ ↻
   5. ○ 5.5 Designing AI Systems That Work With Human Systems, Not Against Them Open this unit → ✓ ↻
   Open this module → Print certificate of completion