Lesson 33: Argument Evaluation & Logical Fallacies

Introduction

Every tool this course has built — evidence, causation, counterfactual reasoning, hidden assumptions — now converges on one question: how good is this argument, exactly? Today's lesson gives you both the evaluation framework and the vocabulary of fallacies — recurring, nameable patterns of weak reasoning. But the goal isn't to become a fallacy-hunter who rejects every argument on sight. It's to evaluate precisely: to say exactly how far the evidence actually allows a conclusion to travel.

Argument Evaluation: Three Questions

Every argument deserves these three separate checks

1Is the evidence true?

2Is the evidence relevant to the claim?

3Does the evidence support the conclusion strongly enough?

True evidence doesn't guarantee a strong conclusion

An argument's evidence can be entirely factual while its conclusion is still too broad or too certain for what that evidence actually shows — the third question is usually where an argument's real weakness hides.

Argument Strength: Three Tiers

Rate the whole argument, not just its individual sentences

Strongly supported — evidence is substantial, and major alternatives are reasonably addressed. Reasonably supported — evidence exists, but real uncertainty or limitation remains. Insufficiently supported — the evidence doesn't establish the conclusion. Reaching one of these three verdicts, explicitly, is the end goal of every argument evaluation in this lesson.

Validity vs. Soundness

Two related, but different, standards

Valid reasoning — if the premises are true, the conclusion logically follows. Sound argument — valid, and its premises are actually true or well-supported.

All ports in Country X use System A. Port B is in Country X. Therefore, Port B uses System A. If the first premise is true, the conclusion follows validly. But if that first premise is false, the argument is valid yet not sound.

What Is a Logical Fallacy?

Naming a fallacy is the start of the analysis, not the end

A logical fallacy is a recurring, nameable error or weak pattern in reasoning. Recognising the name of a fallacy doesn't automatically prove an argument false — your real job is to identify the reasoning problem and explain why it weakens the argument, not just to shout "fallacy!" and move on.

Fallacy 1 — Hasty Generalization

Too little evidence, too big a conclusion

Two employees made mistakes while using the new software. Therefore, the software is unsuitable for all employees.

Two cases can't support a universal conclusion. Better: two employees experienced difficulties with the software — a claim that stays inside the evidence's actual scope.

Fallacy 2 — Anecdotal Generalization

One story, treated as the whole population

One truck driver said the automated system made his work harder. Therefore, automation makes logistics work harder.

One person's experience — a useful anecdote — cannot by itself establish a general, population-level claim.

Fallacy 3 — Correlation → Causation

A familiar mistake, now seen as a named fallacy

Employees who use AI tend to perform better. Therefore, AI causes better performance.

Alternative: more skilled employees may simply be more likely to use AI, or managers may assign AI to already high-performing teams. Association alone does not establish causation.

Fallacy 4 — Post Hoc Reasoning

Post hoc ergo propter hoc — 'after this, therefore because of this'

The company introduced a new dashboard in March. Profits increased in April. Therefore, the dashboard increased profits.

Sales, prices, costs, and demand could each explain the same increase. This is Lesson 32's after vs. because trap, given its formal name.

Fallacy 5 — False Dilemma

Presenting only two options when more exist

Either the company adopts AI or it will become uncompetitive.

Real alternatives might include partial automation, process redesign, employee training, outsourcing, or better non-AI software. "Either A or B" is rarely a complete map of the real options.

Fallacy 6 — False Equivalence

One shared feature does not make two things the same

Both systems use computers, so they provide the same level of security.

Relevant dimensions like architecture, access control, encryption, and monitoring can differ enormously even when a surface feature is shared.

Fallacy 7 — Straw Man

Attacking a weaker, distorted version of the actual argument

Original: "AI should be introduced gradually with appropriate oversight." Response: "So you want companies to replace every human employee with AI."

The response exaggerates the original position into something easier to attack. Always ask: did this response actually address the argument that was made?

Fallacy 8 — Appeal to Authority

An expert's opinion is not automatic proof

A famous business leader says AI always improves productivity. Therefore, AI always improves productivity.

Expertise can be relevant supporting evidence, but it isn't automatic proof. Check: is the expertise relevant? Is there independent evidence? What does the wider research say?

Fallacy 9 — Appeal to Popularity

What many people believe is not automatically true

Most managers believe automation improves efficiency. Therefore, automation improves efficiency.

Popular belief isn't evidence on its own — though if public opinion is the actual topic under discussion, it can become genuinely relevant data. Context matters.

Fallacy 10 — Ad Hominem

Attacking the person instead of the argument

The analyst's proposal should be rejected because he has little management experience.

The proposal's own evidence is the real question. A person's background might be relevant context, but it doesn't automatically make their proposal false.

Fallacy 11 — Circular Reasoning

Using the conclusion as its own premise

The policy is effective because it works well. Why does it work well? Because it is effective. Structure: A is true because A is true — with no independent evidence anywhere in the loop.

Fallacy 12 — Slippery Slope

A small step, assumed to cascade into an extreme outcome

If the company allows employees to use AI for simple tasks, eventually employees will stop thinking independently, and the organization will lose all expertise.

Each link in this chain needs its own evidence. The real question: what evidence supports each individual causal step, not just the scary ending?

Fallacy 13 — Appeal to Ignorance

Absence of disproof is not proof, and absence of proof is not disproof

X hasn't been disproved. Therefore, X is true. Or: we don't have evidence that X is true. Therefore, X is false. Both moves are problematic. Remember Lesson 30's rule: absence of evidence ≠ evidence of absence — though how much evidence a claim needs can genuinely vary by the claim's nature.

Fallacy 14 — Cherry Picking

Choosing only the evidence that fits, and ignoring the rest

An author cites one positive study while ignoring five other studies with mixed results on the same question. The critical reader's question: is the cited evidence representative of the full body of evidence, or has it been hand-picked?

Fallacy 15 — Moving the Goalposts

Shifting the standard of proof after it's been met

"Show me one study." A study is given. "One study isn't enough." More studies are given. "They must all be from a particular country." The bar keeps moving. This is different from a genuine, evolving methodological concern — the difference is whether the new standard is actually reasonable, or just a way to never be satisfied.

Fallacy 16 — Appeal to Emotion

Feeling replaces evidence

We must adopt this system because our employees deserve the best technology.

Employees deserve good technology may be a legitimate value claim, but whether this specific system is actually effective is a separate factual question with its own evidence requirement.

Fallacy 17 — Red Herring

Changing the subject instead of answering it

Question: did the software improve billing accuracy? Response: the company has always been innovative. This doesn't answer the actual question. Always check: is this response relevant to the claim that was actually made?

Fallacy 18 — Overgeneralization From Scope

This one deserves special attention

Evidence: 100 workers in one port. Conclusion: AI improves productivity in all ports worldwide. Sample scope and conclusion scope don't match — this is precisely Lesson 30's scope-matching skill, looked at again as a formal reasoning error.

Not Every Weak Argument Is a Named Fallacy

Sometimes the honest answer is simply 'insufficient,' with no label needed

The company introduced a new dashboard, and employee satisfaction increased. Therefore, the dashboard caused the increase.

Rather than forcing a fallacy name onto this, the more precise statement is: "The argument makes a causal inference from a before-and-after association without adequately ruling out alternative explanations." This is more accurate, and more useful, than reaching for a fallacy label that doesn't quite fit.

The Problem With Fallacy-Hunting

Chasing fallacies everywhere is its own reading mistake

Treating every argument as a hunt for faults can turn a minor issue into a major one, cause you to miss the author's strongest interpretation, and make you unnecessarily skeptical of reasonable claims. Better approach: first understand → then reconstruct → then evaluate.

Principle of Charity + Fallacy Detection

Two questions, always in this order

First: "What is the strongest reasonable version of this argument?" Then: "Does that strongest version survive scrutiny?" This protects you from unfair criticism and keeps your evaluation genuinely honest, carrying Lesson 31's principle of charity directly into fallacy detection.

A Twelve-Point Argument Evaluation Framework

Working through any argumentative paragraph, end to end

1What is the claim?

2What evidence is offered?

3Is the evidence relevant to the claim?

4Is the evidence sufficient?

5How reliable is the evidence?

6What unstated assumptions does the argument need?

7If it's a causal claim, what alternative explanations exist?

8Does the conclusion stay inside the evidence's scope?

9Is there a counterargument or opposing evidence?

10Does the author acknowledge any uncertainty?

11Does the conclusion reasonably follow from the premises?

12Overall judgment: strong, moderate, weak, or insufficiently supported

Deep Example: AI Forecasting Tool

A full evaluation, worked from claim to final judgment

A logistics company introduced an AI-based forecasting tool and subsequently reduced inventory costs by 15%. Management concluded that the tool had substantially improved supply-chain efficiency. The conclusion is supported by the fact that the greatest cost reduction occurred in warehouses that used the tool most extensively. Critics, however, note that the company also reduced overall inventory levels and renegotiated several supplier contracts during the same period. Management responds that these changes cannot fully explain the improvement because the warehouses with the highest AI adoption experienced the largest reductions.

Claim: the AI forecasting tool substantially improved supply-chain efficiency. Evidence: costs decreased by 15%; warehouses with higher AI adoption saw larger reductions (useful dose-response evidence). Alternative explanations: reduced inventory levels; renegotiated supplier contracts. Potential logical problem: management still assumes AI adoption is the primary explanation for the warehouse-to-warehouse difference — but were the inventory reductions and contract renegotiations themselves warehouse-specific? That information is missing. Most reasonable conclusion: "The evidence is suggestive of a beneficial effect of the AI tool, but the causal contribution of the tool is not fully isolated from concurrent operational changes" — considerably more precise than "AI definitely caused the improvement."

Ready-Made Evaluation Language

Phrases that let you evaluate precisely instead of vaguely

Strong criticism: the evidence does not support the conclusion. Moderate criticism: the evidence provides limited support for the conclusion. Qualification: the evidence is consistent with the claim, but does not establish it conclusively. Alternative explanation: the observed effect may be attributable to other factors. Scope problem: the conclusion extends beyond the scope of the evidence. Causal problem: the analysis does not establish a causal relationship. Measurement problem: the measure used may not adequately capture the underlying concept. Evidence quality: the evidence is suggestive but not definitive.

"Wrong" vs. "Not Established"

One of the most important distinctions in academic English

If evidence is insufficient for a claim like "the evidence proves that AI caused the improvement," you should not automatically conclude "AI did not cause the improvement." The precise statement is: "the evidence does not establish that AI caused the improvement." Not established ≠ disproved — exactly the same distinction from Lessons 21, 30, and 32, now applied directly to evaluating an argument's final verdict.

The Master Logical Reading Map

Lessons 30 through 33, assembled into one framework

DATA → PATTERN → INTERPRETATION → CLAIM → EVIDENCE → ASSUMPTION → CAUSAL MECHANISM → COUNTERFACTUAL → ALTERNATIVE EXPLANATIONS → SCOPE → LOGICAL VALIDITY → COUNTERARGUMENT → QUALIFICATION → FINAL EVALUATION. This is now your working framework for any serious nonfiction paragraph.

Vocabulary in Context

evaluateverbformal

to judge the quality, strength, or validity of something carefully (মূল্যায়ন করা)

Often with evaluate the argument, critically evaluate, evaluate evidence

“The panel will evaluate each bid against the published criteria before making an award.”

unwarrantedadjectiveformal

not justified by the evidence or reasoning actually given (অযৌক্তিক/অসমর্থিত)

Often with unwarranted assumption, unwarranted conclusion, entirely unwarranted

“The claim that the strike caused the recession is an unwarranted leap from the evidence.”

misrepresentverbformal

to describe something inaccurately, often making it easier to criticize (ভুলভাবে উপস্থাপন করা)

Often with misrepresent the facts, deliberately misrepresent, misrepresent a position

“The article misrepresented the union's position by quoting only its most extreme demand.”

cherry-pickverbneutral

to select only the evidence that supports a desired conclusion, while ignoring contrary evidence (বেছে বেছে সুবিধাজনক তথ্য নেওয়া)

Often with cherry-pick data, accused of cherry-picking, cherry-pick the evidence

“The vendor cherry-picked its three best months and ignored the poor performance of the rest of the year.”

burden of proofnoun phraseformal

the obligation to provide sufficient evidence for a claim before it is accepted (প্রমাণের দায়িত্ব)

Often with shift the burden of proof, bear the burden of proof, burden of proof lies

“The burden of proof lies with the claimant, who must show that the goods were damaged in transit.”

undermineverbformal

to weaken the strength or credibility of an argument or conclusion (দুর্বল/ক্ষুণ্ণ করা)

Often with undermine confidence, seriously undermine, undermine the argument

“The discovery of duplicate invoices undermined the credibility of the entire audit.”

substantiateverbformal

to provide evidence that supports or confirms a claim (প্রমাণ দ্বারা সমর্থন করা)

Often with substantiate a claim, fully substantiate, unable to substantiate

“The company could not substantiate its claim that the terminal was the fastest in the region.”

reconstructverbformal

to rebuild an argument in its clearest and strongest form before evaluating it (পুনর্গঠন করা)

Often with reconstruct the sequence, reconstruct an argument, reconstruct events

“Investigators reconstructed the sequence of events from the crane's control log.”

dose-responsenoun phrasetechnical

a pattern in which a larger amount of a proposed cause is associated with a larger effect (মাত্রা-প্রতিক্রিয়া সম্পর্ক)

Often with dose-response relationship, dose-response pattern, dose-response curve

“A dose-response pattern appeared: the more overtime staff worked, the more errors they made.”

calibratedadjectivetechnical

carefully adjusted to match the actual strength of the available evidence (যথাযথভাবে সমন্বিত)

Often with well calibrated, carefully calibrated, calibrated response

“The ministry gave a calibrated response, stating what the data showed and no more.”

Guided Reading Practice

Read this passage once for its overall claim, then identify any fallacy pattern at work before checking the analysis underneath it.

A regional port authority recently claimed that its new automated crane system had eliminated the need for further safety investment, stating: "Since the system was introduced, we haven't had a single reported incident, and no one has proven that it isn't completely safe." Critics who raised concerns about the system's long-term reliability were dismissed by port management as "people who simply don't understand modern technology."

First claim's fallacy: "no one has proven that it isn't completely safe" is an appeal to ignorance — absence of disproof is being treated as proof of safety, when in fact a short incident-free record could simply mean too little time has passed to observe a rare failure. Second move's fallacy: dismissing critics as "people who simply don't understand modern technology" is an ad hominem — it attacks the critics rather than engaging with their actual concern about long-term reliability. A precise evaluation: the passage does not establish that the system is safe in the long term; it only reports an absence of incidents so far, and the dismissal of critics does not address their underlying concern.

Golden Rule

Good critical reading is not about finding reasons to reject an argument — it is about determining exactly how far the evidence allows you to go.

Lesson Summary

Today's lesson gave you both a framework for evaluating any argument's overall strength and a vocabulary of eighteen named reasoning errors — from hasty generalization and post hoc reasoning to straw man, cherry picking, and moving the goalposts. You practised the discipline of naming a fallacy only when it genuinely fits, and of saying simply "insufficiently supported" when no specific label applies cleanly. Throughout, the principle of charity kept your evaluation fair: reconstruct the argument's strongest form first, then test it — and close with a precisely calibrated verdict, distinguishing "not established" from "disproved" every time.

The sequence worth automating

What is the claim? What is the evidence? Is it relevant? Is it sufficient? What assumptions does it need? Could there be another explanation? Does the conclusion exceed the evidence? Is there a named fallacy, or just insufficient support? How strong is the argument overall? Running any argument through this sequence is today's entire lesson, compressed into a habit.

Practice: Test What You've Learned

Work through every question yourself before checking anything.

Before you start

Only reach for a fallacy's name when the pattern genuinely fits — a vague sense that "something's wrong" deserves its own precise description just as much as a hasty generalization does.

Part A — Name the Fallacy

Name the fallacy pattern in each example.

Two employees made mistakes with the new software; therefore, the software is unsuitable for all employees.
Show answer
Hasty generalization. Two isolated cases cannot support a universal conclusion about every employee; the honest claim would stay inside the evidence's actual scope, such as "two employees experienced difficulties with the software."
The company introduced a new dashboard in March, and profits increased in April, so the dashboard increased profits.
Show answer
Post hoc reasoning (post hoc ergo propter hoc). The dashboard simply came before the profit increase in time; sales, prices, costs, or demand could each explain the same rise, and nothing here rules them out.
Either the company adopts AI, or it will become uncompetitive.
Show answer
False dilemma. It presents only two options when real alternatives exist, such as partial automation, process redesign, employee training, outsourcing, or better non-AI software.
Original position: "AI should be introduced gradually with oversight." Response: "So you want to replace every worker with AI."
Show answer
Straw man. The response exaggerates a cautious, gradual position into an extreme one that's easier to attack, rather than engaging with the actual claim that was made.
A famous business leader says AI always improves productivity; therefore, it does.
Show answer
Appeal to authority. The leader's fame or position is treated as automatic proof, without checking whether their expertise is actually relevant or whether independent evidence supports the claim.

Part B — Argument Strength

Classify each argument as strongly supported, reasonably supported, or insufficiently supported, and justify your answer.

A single, small case study reports a positive result, with no mention of any contrary evidence.
Show answer
Insufficiently supported. One small case study, with no discussion of contrary evidence or alternative explanations, is far too narrow a base for a general conclusion.
Five independent, replicated studies report a consistent positive result, and the author also names a plausible mechanism.
Show answer
Strongly supported. Multiple independent, replicated studies reaching the same result, combined with a plausible named mechanism, is close to the strongest evidentiary base this lesson describes.
A company reports one before-and-after number with no comparison group and no discussion of what else changed.
Show answer
Insufficiently supported. A single before-and-after figure, with no comparison group and no accounting for other concurrent changes, cannot rule out alternative explanations for the result.

Part C — Validity and Soundness

All ports in Country X use System A. Port B is in Country X. Therefore, Port B uses System A. Is this argument valid? Explain.
Show answer
Yes, it is valid. If both premises are true, the conclusion follows necessarily from them by logical structure alone, regardless of whether the premises are actually true.
Under what condition would this same argument fail to be sound, even though it remains valid?
Show answer
It would fail to be sound if the first premise were actually false — that is, if not every port in Country X uses System A. The reasoning would still be structurally valid, but an untrue premise makes the argument unsound.

Part D — Deep Analysis

Return to this lesson's AI forecasting tool passage.

State the claim and the evidence offered for it.
Show answer
Claim: the AI forecasting tool substantially improved supply-chain efficiency. Evidence: inventory costs fell by 15%, and the warehouses that used the tool most extensively showed the largest cost reductions.
What alternative explanations does the passage raise?
Show answer
The passage notes that the company also reduced overall inventory levels and renegotiated several supplier contracts during the same period — either of which could have contributed to the cost reduction independently of the AI tool.
What unstated assumption does management's response still depend on?
Show answer
It assumes that AI adoption, rather than the inventory reductions or contract renegotiations, is the primary explanation for the warehouse-to-warehouse difference in savings — without establishing whether those other changes were themselves warehouse-specific.
Is there a genuine logical gap here, or a specifically named fallacy? Explain your choice.
Show answer
A genuine logical gap, not a single named fallacy. It doesn't cleanly fit hasty generalization, post hoc, or any other label on this lesson's list; the more precise description is that the causal inference hasn't adequately ruled out alternative, concurrent explanations.
Write the most reasonable, calibrated conclusion this evidence actually supports.
Show answer
The evidence is suggestive of a beneficial effect from the AI forecasting tool, but the tool's causal contribution has not been fully isolated from the concurrent inventory reductions and contract renegotiations happening in the same period.

Part E — Wrong vs. Not Established

Rewrite this overreaching claim precisely: "The evidence proves that the automated system was the sole cause of the improvement." Your rewrite should say exactly what the evidence does and does not establish, without claiming the opposite is true either.
Show answer
The evidence suggests that the automated system contributed to the improvement, but it does not establish that the system was the sole cause; other concurrent factors may also have played a role, and the available evidence cannot rule them out.

Lesson 34 — Rhetorical Structure, Author Purpose & Persuasive Language: moving from whether an argument's logic holds up to why the author chose to present it this particular way — purpose, framing, and the persuasive structures beneath the reasoning.