Reflection and self-critique pattern in Agentic AI - Model Metrics & Evaluation

Metrics & Evaluation - Reflection and self-critique pattern

Which metric matters for Reflection and self-critique pattern and WHY

The Reflection and self-critique pattern focuses on improving AI agents by evaluating their own outputs and decisions. Key metrics include accuracy to measure correctness, precision and recall to understand error types, and F1 score to balance these. These metrics help the agent identify where it makes mistakes and how to improve. Without these, self-critique would lack clear guidance.

Confusion matrix example

      Actual \ Predicted | Positive | Negative
      -------------------|----------|---------
      Positive           |    80    |   20    
      Negative           |    10    |   90    
    

This matrix shows the agent's decisions: 80 true positives (correct), 20 false negatives (missed), 10 false positives (wrongly flagged), and 90 true negatives (correctly ignored). The agent uses this to reflect on errors.

Precision vs Recall tradeoff with examples

Reflection helps balance precision and recall. For example, a medical AI must have high recall to catch all diseases (few misses), even if precision drops (some false alarms). A spam filter AI needs high precision to avoid marking good emails as spam, even if some spam slips through (lower recall). Self-critique guides the agent to adjust this balance based on goals.

What "good" vs "bad" metric values look like

Good: High accuracy (e.g., 90%+), balanced precision and recall (both above 80%), and F1 score close to 1. This means the agent correctly identifies most cases and makes few mistakes.

Bad: High accuracy but very low recall (e.g., 10%), meaning the agent misses many true cases. Or very low precision, causing many false alarms. These show poor self-critique and need improvement.

Common pitfalls in metrics for Reflection and self-critique

• Accuracy paradox: High accuracy can be misleading if data is imbalanced (e.g., 95% accuracy but misses all rare cases).
• Data leakage: If the agent learns from future or test data, metrics look better but are not real.
• Overfitting indicators: Very high training metrics but poor test metrics show the agent is not generalizing well.
• Ignoring recall or precision: Focusing on one metric alone can hide serious problems.

Self-check question

Your agent has 98% accuracy but only 12% recall on fraud detection. Is it good for production? Why or why not?

Answer: No, it is not good. The agent misses 88% of fraud cases (low recall), which is dangerous. High accuracy is misleading because fraud is rare. The agent needs better recall to catch fraud effectively.

Key Result

Reflection and self-critique rely on balanced precision, recall, and F1 to guide AI improvement effectively.

Practice

1. What is the main purpose of the Reflection and self-critique pattern in AI?

easy

A. To store large amounts of data

B. To speed up AI computations

C. To help AI review and improve its own answers

D. To create new AI models automatically

Solution

1. Step 1: Understand the pattern's goal

   The reflection and self-critique pattern is designed to let AI look back at its answers and find mistakes.

2. Step 2: Identify the main benefit

   By reviewing its own work, AI can fix errors and improve future responses.

3. Final Answer:

   To help AI review and improve its own answers -> Option C

4. Quick Check:

   Reflection and self-critique = improve answers [OK]

Hint: Focus on improvement through self-review [OK]

Common Mistakes:

• Confusing speed with accuracy
• Thinking it stores data
• Assuming it creates new models

2. Which of the following is the correct way to describe the reflection step in the pattern?

easy

A. AI reviews its previous answers to find mistakes

B. AI ignores previous answers and generates new ones

C. AI deletes all previous data to start fresh

D. AI copies answers from other models without checking

Solution

1. Step 1: Define reflection in AI context

   Reflection means looking back at past answers to check for errors or improvements.

2. Step 2: Match options to definition

   Only AI reviews its previous answers to find mistakes correctly states that AI reviews previous answers to find mistakes.

3. Final Answer:

   AI reviews its previous answers to find mistakes -> Option A

4. Quick Check:

   Reflection = review past answers [OK]

Hint: Reflection means reviewing past work carefully [OK]

Common Mistakes:

• Thinking reflection means ignoring past answers
• Confusing reflection with deleting data
• Assuming copying answers is reflection

3. Consider this simple AI pseudo-code using reflection and self-critique:

answer = AI.generate_answer(question)
errors = AI.reflect(answer)
if errors:
    answer = AI.fix_errors(answer, errors)
print(answer)

What will print(answer) show if the AI finds errors?

medium

A. The original answer without changes

B. The corrected answer after fixing errors

C. No output because the program stops

D. An error message instead of an answer

Solution

1. Step 1: Understand the code flow

   The AI first generates an answer, then reflects to find errors. If errors exist, it fixes them.

2. Step 2: Determine the final printed output

   Since errors are fixed before printing, the output is the corrected answer.

3. Final Answer:

   The corrected answer after fixing errors -> Option B

4. Quick Check:

   Errors fixed before print = corrected answer [OK]

Hint: Errors fixed before print means corrected output [OK]

Common Mistakes:

• Assuming original answer prints despite errors
• Thinking program stops on errors
• Confusing error message with fixed answer

4. You have this AI code snippet:

answer = AI.generate_answer(question)
errors = AI.reflect(answer)
if errors:
    AI.fix_errors(answer, errors)
print(answer)

Why might this code fail to print the corrected answer?

medium

A. Because fix_errors does not update answer variable

B. Because reflect never finds errors

C. Because print is called before generating answer

D. Because answer is not defined

Solution

1. Step 1: Analyze variable updates

   The fix_errors function is called but its result is not assigned back to answer.

2. Step 2: Understand impact on output

   Since answer is unchanged, print shows the original, not corrected, answer.

3. Final Answer:

   Because fix_errors does not update answer variable -> Option A

4. Quick Check:

   Fix must assign back to answer [OK]

Hint: Assign fixed answer back to variable before printing [OK]

Common Mistakes:

• Assuming reflect never finds errors
• Thinking print is called too early
• Ignoring variable assignment after fixing

5. You want to improve an AI assistant using the reflection and self-critique pattern. Which approach best applies this pattern to reduce repeated mistakes over time?

hard

A. AI copies answers from a fixed database without checking

B. AI generates answers randomly to explore new possibilities

C. AI deletes old answers to save memory without review

D. After each answer, AI reviews its response, identifies errors, fixes them, and updates its knowledge base

Solution

1. Step 1: Identify key steps in the pattern

   The pattern involves reviewing answers, finding errors, fixing them, and learning from mistakes.

2. Step 2: Match approach to pattern goals

   After each answer, AI reviews its response, identifies errors, fixes them, and updates its knowledge base describes reviewing, fixing, and updating knowledge, which fits the pattern perfectly.

3. Final Answer:

   After each answer, AI reviews its response, identifies errors, fixes them, and updates its knowledge base -> Option D

4. Quick Check:

   Review + fix + learn = improved AI [OK]

Hint: Choose option with review, fix, and learning steps [OK]

Common Mistakes:

• Ignoring learning from errors
• Choosing random or fixed answers
• Skipping error identification