Exercise Type 10: Free Energy Principle (FEP) Comprehension
What the exam asks: You are given statements about the Free Energy Principle and Active Inference, and must identify which ones are consistent with the theory.
Part 0: What Is the Free Energy Principle? (Plain English)
The Core Idea
The Free Energy Principle (FEP) is a theory about how living things (especially brains) work. It says:
Living things try to minimize "surprise" — the difference between what they expect to sense and what they actually sense.
They do this in TWO ways: 1. Perception: Update your beliefs about the world to better match your sensations 2. Action: Change your sensations by acting on the world
Key Vocabulary
| Term | Plain English Meaning |
|---|---|
| Generative model | Your internal model of how the world works — it predicts what you should sense |
| Sensory inputs | What you actually see, hear, feel, etc. |
| Free Energy | A measure of how wrong your predictions are (like "surprise") |
| Perception | Updating your beliefs to minimize Free Energy |
| Action (Active Inference) | Choosing actions that make your predictions come true |
| Expected Free Energy | How much surprise you EXPECT to have in the future if you take a certain action |
| Target priors | Your preferred/desired future states — what you WANT to happen |
The Key Principles
- Agents have a generative model — an internal model of how sensory data is generated
- Perception minimizes Free Energy — you update beliefs to match observations
- Actions minimize Expected Free Energy — you choose actions that reduce expected future surprise
- Goals are encoded as target priors — you specify what future states you want, then act to make predictions match those targets
Part 1: FULL Walkthrough of Real Exam Questions
EXAM QUESTION 1 (2021-Part-B, Question 1a)
The Free Energy Principle is about biological self-organization. Which statement is most consistent with FEP?
Options: - (a) Our actions aim to reduce the complexity of our model of the environment - (b) Learning maximizes variational free energy - (c) Perception aims to reduce the complexity of our model of the environment - (d) We act to fulfil our predictions about future sensory inputs
STEP-BY-STEP SOLUTION
(a) "Reduce complexity of model" — FEP is about minimizing surprise/free energy, not specifically about model complexity. ELIMINATE.
(b) "Maximizes variational free energy" — WRONG direction. We MINIMIZE free energy, not maximize. ELIMINATE.
(c) "Reduce complexity" — again, not the core idea. Perception minimizes free energy by updating beliefs, not specifically "complexity." ELIMINATE.
(d) "We act to fulfil our predictions" — YES! In Active Inference, we act to make our sensory predictions match what we want. Our predictions become self-fulfilling prophecies through action. ✓
Answer: (d) ✅
EXAM QUESTION 2 (2021-Part-B, Question 1d)
How to equip an agent with goal-driven behavior in FEP?
Options: - (a) Specify a cost function and minimize costs - (b) Extend the generative model with target priors for future observations. Then minimize Free Energy. - (c) Specify a cost function of actions and minimize - (d) Extend with a posterior distribution for actions and maximize posterior
STEP-BY-STEP SOLUTION
(a) "Cost function" — this is the traditional engineering approach, NOT the FEP way. FEP doesn't use external cost functions. ELIMINATE.
(b) "Target priors in the generative model + minimize Free Energy" — YES! In FEP, goals are encoded as preferred future states (target priors) built into your model. Then you act to minimize expected free energy, which naturally achieves those goals. ✓
(c) "Cost function of actions" — again, not the FEP approach. ELIMINATE.
(d) "Maximize posterior for actions" — this describes action selection but doesn't explain how GOALS are encoded. ELIMINATE.
Answer: (b) ✅
EXAM QUESTION 3 (2021-Part-B, Question 5b)
Which statements are consistent with FEP? - (a) An active inference agent holds a generative model for its sensory inputs - (b) Actions are inferred from differences between predicted and desired future observations - (c) Actions are inferred from differences between predicted and actual future observations - (d) An active inference agent focuses on explorative behavior only
Options: - (a) (a) and (b) - (b) (a) - (c) (b) and (d) - (d) (c) and (d)
STEP-BY-STEP SOLUTION
(a) True. Agents MUST have a generative model — that's the foundation of FEP.
(b) True. Actions bridge the gap between what you PREDICT and what you DESIRE (target prior). This is the essence of Active Inference.
(c) False. It's about "desired" future observations (target priors), not "actual" future observations. You can't know actual future observations.
(d) False. Agents balance exploration AND exploitation — they don't ONLY explore.
So (a) and (b) are true.
Answer: (a) — (a) and (b) ✅
EXAM QUESTION 4 (2023, Question 4a)
Active Inference agent:
Options: - (a) Perception minimizes the complexity of the states - (b) Agent infers actions by maximizing free energy in future states - (c) Agent infers actions by maximizing expected accuracy in future states - (d) Agent infers actions by minimizing expected free energy in future states
STEP-BY-STEP SOLUTION
(a) "Complexity of states" — not the core idea. Perception minimizes free energy, not "complexity." ELIMINATE.
(b) "Maximizing free energy" — WRONG direction. We MINIMIZE free energy. ELIMINATE.
(c) "Maximizing expected accuracy" — this sounds plausible but isn't the standard FEP formulation. ELIMINATE.
(d) "Minimizing expected free energy in future states" — YES! Actions minimize Expected Free Energy (EFE). ✓
Answer: (d) ✅
EXAM QUESTION 5 (2022, Question 4a)
Which statement most consistently describes FEP?
Options: - (a) Actions aim to minimize the free energy of future states of the world - (b) Actions aim to minimize the complexity of future states of the world - (c) Intelligent decision making requires minimization of a functional of beliefs about future states - (d) Intelligent decision making requires minimization of a cost function of future states
STEP-BY-STEP SOLUTION
(a) "Minimize free energy of future states of the world" — close, but FEP is about minimizing expected free energy of your BELIEFS, not directly of the world.
(b) "Minimize complexity" — not the core idea. ELIMINATE.
(c) "Minimization of a functional of beliefs about future states" — YES! Expected Free Energy is a functional (a function of functions) of your probability distributions (beliefs) about future states. This is the most accurate description. ✓
(d) "Cost function" — this is the traditional approach, not FEP. ELIMINATE.
Answer: (c) ✅
Part 2: Tricks & Shortcuts
TRICK 1: Always MINIMIZE, Never Maximize
FEP is about MINIMIZING free energy (or expected free energy).
If an option says "maximize free energy" → wrong.
TRICK 2: Beliefs, Not Cost Functions
FEP uses probability distributions (beliefs), not external cost functions.
If an option mentions "cost function" → wrong (usually).
TRICK 3: Desired, Not Actual
Actions bridge predicted vs. DESIRED future states (target priors).
If an option says "predicted vs. actual future" → wrong.
TRICK 4: Generative Model Is Fundamental
Agents MUST have a generative model.
If a statement says agents hold a generative model → true.
Part 3: Practice Exercises
Exercise 1
In FEP, how does an agent choose actions?
Options: - (a) Minimize a cost function - (b) Minimize expected free energy in future states - (c) Maximize free energy in future states - (d) Maximize expected accuracy
Exercise 2
Which statement is consistent with FEP?
Options: - (a) Agents hold a generative model for sensory inputs - (b) Actions maximize free energy - (c) Agents only explore - (d) Goals are specified as external cost functions
Exercise 3
How are goals encoded in FEP?
Options: - (a) As a cost function to minimize - (b) As target priors in the generative model - (c) As a posterior distribution to maximize - (d) As constraints on actions