Fallacy Detection

Automated identification and categorization of logical fallacies in arguments with educational explanations

Overview

Fallacy Detection is one of the most powerful features of Argument Cartographer. Our AI actively scans source material for logical errors and rhetorical manipulation, educating users not just on what is being said, but how they might be misled.

The system identifies 15+ common fallacy types with confidence scoring, severity ratings, and suggested improvements.

How Fallacy Detection Works

The detection process is integrated directly into the argument analysis pipeline:

1. Context Analysis

   Gemini 1.5 Pro reads all scraped source material (up to 20,000 tokens)

2. Pattern Recognition

   AI identifies patterns matching known fallacy structures using trained examples

3. Quote Extraction

   System extracts the exact problematic text from sources

4. Confidence Scoring

   Each detection receives a 0-100% confidence score based on pattern clarity

5. Severity Classification

   Fallacies are rated Critical, Major, or Minor based on impact on argument validity

6. Educational Content Generation

   AI generates explanation, definition, and suggested improvements

Detected Fallacy Schema

Each detected fallacy follows this comprehensive structure:

DetectedFallacy

interface DetectedFallacy {
  id: string; // Unique identifier
  name: string; // E.g., "Ad Hominem", "Straw Man"
  severity: 'Critical' | 'Major' | 'Minor';
  category: string; // "Logical", "Rhetorical", "Statistical"
  confidence: number; // 0-1 (AI confidence)
  problematicText: string; // Exact quote
  explanation: string; // Why it's fallacious
  definition: string; // Formal definition
  avoidance: string; // How to avoid
  example: string; // Clean example
  suggestion: string; // Improved phrasing
  location?: string; // Where in argument (e.g., "Claim 3")
}

Common Fallacy Types

Logical Fallacies

Errors in reasoning structure that invalidate conclusions.

Ad Hominem - Attack the Person

Definition: Attacking the character, circumstances, or identity of an individual instead of addressing their argument.

Example:

"You can't trust Dr. Smith's climate research - 
she drives an SUV!"

Why Problematic: The validity of an argument stands independent of the person making it. A hypocrite can still make a valid point.

How to Avoid: Focus on the evidence and logic of the argument itself. Address claims, not the claimant.

Suggested Fix:

"Dr. Smith's climate research methodology has 
limitations because [specific technical reasons]."

Severity: Usually Major - Completely sidesteps the actual argument

Straw Man - Misrepresentation

Definition: Distorting, exaggerating, or oversimplifying an opponent’s position to make it easier to attack.

Example:

"Gun control advocates want to ban ALL guns and 
leave law-abiding citizens defenseless against criminals."

Why Problematic: Refuting a distorted version doesn’t address the actual argument. Most gun control proposals are nuanced, not absolute bans.

How to Avoid: Steelman instead - represent opposing views in their strongest, most charitable form.

Suggested Fix:

"Gun control advocates propose background checks 
and assault weapon restrictions, which raises concerns 
about [specific legitimate concerns]."

Severity: Critical - Fundamentally dishonest argumentation

False Dichotomy - Limited Options

Definition: Presenting only two options or sides when more alternatives exist.

Example:

"Either we cut all social programs or the national 
debt will destroy the economy."

Why Problematic: Oversimplifies complex issues and excludes middle-ground solutions like selective cuts, revenue increases, or program reforms.

How to Avoid: Acknowledge the full spectrum of options and trade-offs.

Suggested Fix:

"We have several options for addressing the national 
debt: targeted program cuts, revenue increases, 
economic growth strategies, or combinations thereof."

Severity: Major - Artificially constrains debate

Slippery Slope - Unwarranted Chain Reaction

Definition: Assuming that one action will inevitably lead to a chain of events without providing evidence for the causal links.

Example:

"If we legalize same-sex marriage, next people will 
want to marry their pets, then inanimate objects!"

Why Problematic: No evidence that legal recognition of one adult consensual relationship leads to fundamentally different scenarios.

How to Avoid: Provide empirical evidence for each step in the causal chain.

Suggested Fix:

"Marriage law changes may have unintended consequences. 
Evidence from countries that implemented similar changes 
shows [specific documented effects]."

Severity: Major - Fearmongering without evidence

Circular Reasoning - Begging the Question

Definition: The conclusion is assumed in one of the premises, creating a logical loop.

Example:

"The Bible is true because it's the word of God. 
We know it's God's word because the Bible says so."

Why Problematic: Provides no independent verification - assumes what it’s trying to prove.

How to Avoid: Provide independent evidence that doesn’t rely on the conclusion.

Suggested Fix:

"The Bible's historical claims can be evaluated 
through [archaeological evidence, textual analysis, 
historical corroboration]."

Severity: Critical - Logically invalid

Rhetorical Fallacies

Manipulative persuasion tactics that exploit emotions.

Appeal to Emotion - Pathos Over Logos

Definition: Manipulating emotions instead of providing logical arguments.

Example:

"Think of the children! We must ban violent video 
games to protect our innocent youth from corruption."

Why Problematic: Emotional appeals can be powerful but don’t constitute evidence. Research on media effects is complex and nuanced.

Severity: Minor to Major depending on context

Appeal to Authority - False Expertise

Definition: Citing an authority figure’s opinion as evidence when they lack relevant expertise.

Example:

"Einstein believed in God, so atheism must be wrong."

Why Problematic: Expertise in physics doesn’t transfer to theology. Authority must be relevant to the claim.

Severity: Major

Bandwagon - Appeal to Popularity

Definition: Arguing something is true or good because many people believe it.

Example:

"50 million people can't be wrong - this diet plan 
must work!"

Why Problematic: Popularity doesn’t equal truth. Many popular beliefs have been false (flat earth, etc.).

Severity: Minor to Major

Statistical Fallacies

Misuse or misrepresentation of data and statistics.

Cherry Picking - Selective Evidence

Definition: Presenting only data that supports your position while ignoring contradictory evidence.

Example:

"Global warming is a hoax - it snowed in Texas 
last week!"

Why Problematic: Local weather ≠ global climate. Comprehensive data shows clear warming trends.

Severity: Critical - Deliberately misleading

Hasty Generalization - Insufficient Sample

Definition: Drawing broad conclusions from limited or unrepresentative samples.

Example:

"I met two rude French people - the French are rude!"

Why Problematic: Sample size of 2 can’t support universal claim about 67 million people.

Severity: Major

Correlation ≠ Causation

Definition: Assuming that because two things correlate, one must cause the other.

Example:

"Ice cream sales and drownings both increase in summer - 
ice cream causes drowning!"

Why Problematic: Both are caused by a third factor (hot weather). Correlation requires further investigation.

Severity: Major

Severity Classification

Fallacies are rated on a 3-tier system:

Critical

Impact: Completely invalidates the argument

Examples:

• Straw Man
• Circular Reasoning
• Cherry Picking
• Non Sequitur

Visual: Red badge, highest priority display

Meaning: The argument cannot be considered valid until this is addressed

Major

Impact: Significantly weakens the argument

Examples:

• Ad Hominem
• False Dichotomy
• Slippery Slope
• Appeal to Authority (irrelevant)

Visual: Orange badge, high priority

Meaning: The argument may have merit but needs substantial revision

Minor

Impact: Weakens persuasiveness but doesn’t necessarily invalidate logic

Examples:

• Appeal to Emotion (light)
• Bandwagon
• Anecdotal Evidence

Visual: Yellow badge, moderate priority

Meaning: Worth noting but may be acceptable rhetorical flourish in context

Confidence Scoring

Each detection includes an AI confidence score:

Confidence	Interpretation	Action
90-100%	Very High	Definitely a fallacy
70-89%	High	Likely fallacious, worth reviewing
50-69%	Moderate	Borderline case, context-dependent
30-49%	Low	Possibly fallacious, may be false positive
0-29%	Very Low	Likely false positive

Use the confidence filter in the UI to hide low-confidence detections and focus on clear-cut cases.

Fallacy Card UI

Fallacies are displayed in expandable cards:

Collapsed State

• Severity badge (colored)
• Category tag
• Fallacy name (bold)
• Brief definition (first sentence only)
• Location indicator (which claim)
• Confidence percentage

Expanded State

Click to reveal full educational content:

Fallacy Card Component

<FallacyCard fallacy={detectedFallacy}>
  <Section color="red">
    <Icon>Quote</Icon>
    <Title>Problematic Text</Title>
    <Quote>"{fallacy.problematicText}"</Quote>
  </Section>
  
  <Section color="orange">
    <Icon>Info</Icon>
    <Title>Why This Is Problematic</Title>
    <Text>{fallacy.explanation}</Text>
  </Section>
  
  <Section color="blue">
    <Icon>Book</Icon>
    <Title>Definition</Title>
    <Text>{fallacy.definition}</Text>
  </Section>
  
  <Section color="green">
    <Icon>Shield</Icon>
    <Title>How to Avoid</Title>
    <Text>{fallacy.avoidance}</Text>
  </Section>
  
  <Section color="teal">
    <Icon>Sparkles</Icon>
    <Title>Suggested Improvement</Title>
    <Text>{fallacy.suggestion}</Text>
  </Section>
</FallacyCard>

Educational Value

Fallacy detection serves a dual purpose:

Immediate Analysis

Identify weak points in current arguments being analyzed

Long-Term Learning

Build critical thinking skills by seeing examples in real-world context

Media Literacy

Recognize manipulation tactics in news, advertising, and social media

Better Arguments

Improve your own reasoning by learning what to avoid

Limitations & Considerations

AI Detection Isn’t Perfect: The system may:

• Flag rhetorical flourishes as fallacies (false positives)
• Miss subtle logical errors (false negatives)
• Misclassify fallacy types
• Struggle with sarcasm, irony, and complex context

When to Trust Detections

1. Check Confidence Score

   Focus on 70%+ confidence detections first

2. Read the Context

   Click through to source to verify the quote in context

3. Assess Severity

   Critical fallacies deserve immediate attention; Minor ones may be acceptable

4. Use Your Judgment

   The AI is a tool, not the final authority. Think critically!

Integration with Argument Mapping

Fallacies are linked to specific argument nodes:

Node with Fallacies

const claimNode = {
  id: "claim-3",
  content: "Socialism has never worked anywhere...",
  fallacies: ["hasty-generalization", "no-true-scotsman"],
  // ... other properties
};

Visual indicators:

• ⚠️ Warning icon on nodes with fallacies
• Fallacy count badge (e.g., “2 fallacies”)
• Click to expand inline fallacy details

API Access

Developers can use fallacy detection independently:

Standalone Fallacy Detection

import { identifyLogicalFallacies } from '@/ai/flows/identify-logical-fallacies';

const result = await identifyLogicalFallacies({
  argumentText: "Your argument text here..."
});

console.log(result.fallacies); // Array of fallacy names
console.log(result.explanation); // Detailed explanations

Next Steps

Argument Mapping

See how fallacies integrate with argument visualization

Credibility Scoring

Learn how fallacies impact credibility scores

Creating Analyses

Best practices for getting accurate fallacy detection

AI Orchestration

Technical details of fallacy detection AI