ad hominem

person X makes claim Y
person X has negative characteristic Z
therefore claim Y is false

political example:
candidate a: "we should increase funding for education"
candidate b: "don't listen to him - he's a corrupt liar who can't be trusted"

analysis: even if candidate a has character flaws, this doesn't address whether education funding should increase

healthcare debate:
doctor: "patients should quit smoking for their health"
response: "of course you say that - you make money treating sick people!"

analysis: financial interest doesn't make the health claim false

environmental discussion:
activist: "we should reduce carbon emissions"
response: "you fly on airplanes all the time - you're a hypocrite!"

analysis: personal inconsistency doesn't invalidate the environmental argument

policy debate:
economist: "this tax policy could stimulate growth"
response: "that's the same argument used by [unpopular political group]"

analysis: the source of an idea doesn't determine its validity

fallacious: "senator x's healthcare proposal is wrong because she's never worked in healthcare"
better approach: "senator x's proposal has problems because [specific policy analysis]"

fallacious: "don't trust anything from that news source - they're biased"
better approach: "this specific claim is unsupported because [evidence analysis]"

fallacious: "climate scientist y's research is invalid because he once made an error"
better approach: "this specific study has methodological problems: [technical critique]"

fallacious: "pharmaceutical companies fund this research, so it must be wrong"
better approach: "we should examine the study design and data independently"

fallacious: "the witness is unreliable because of his criminal record"
appropriate: "the witness's testimony conflicts with physical evidence in these ways..."

note: in legal contexts, credibility can be relevant, but it must relate to the specific testimony

fallacious: "my neighbor says property taxes are too high, but he's always complaining"
better approach: "let's look at the actual tax rates and compare them to services provided"

fallacious: "sarah's parenting advice is worthless because she's divorced"
better approach: "this specific advice conflicts with child development research because..."

relevant: "as a cardiologist with 20 years experience, dr. smith's opinion on heart disease carries weight"
irrelevant: "dr. smith says the economy is failing - he's a heart doctor, what does he know?"

relevant: "the witness has been caught lying under oath in previous cases"
irrelevant: "the witness is ugly and poorly dressed"

relevant: "the study author owns stock in the company whose product he's testing"
irrelevant: "the author is from a different country, so his research is suspect"

relevant: "candidate x claims to be fiscally responsible but has declared bankruptcy twice"
irrelevant: "candidate x claims to support education but never went to college"

original claim: "we should invest in renewable energy"
ad hominem shift: "but you drive a gas-guzzling suv!"

detection: the response doesn't address renewable energy investment

claim: "smoking causes lung cancer"
personal attack: "the researcher is funded by anti-tobacco groups"
relevance test: does funding source change the medical evidence? usually no

exercise: read arguments without knowing who made them
then evaluate based purely on logic and evidence
finally, consider whether knowing the source changes the validity

def detect_ad_hominem_patterns(text):
    ad_hominem_indicators = [
        # direct attacks
        r"don't trust \w+ because (he|she|they) (is|are) \w+",
        r"\w+ says \w+ but (he|she|they) (is|are) \w+",

        # circumstantial
        r"of course \w+ says that.*(money|benefit|interest)",
        r"\w+ only says that because \w+",

        # tu quoque
        r"you're a hypocrite",
        r"you do the same thing",

        # guilt by association
        r"that sounds like something \w+ would say",
        r"(nazis|communists|extremists) also believed that"
    ]

    for pattern in ad_hominem_indicators:
        if re.search(pattern, text, re.IGNORECASE):
            return "potential ad hominem detected"

    return "no ad hominem patterns found"

# training data for ad hominem detection
training_examples = [
    ("john's economic theory is wrong because he's never run a business", "ad_hominem"),
    ("sarah's climate research has methodological flaws in the data collection", "legitimate_criticism"),
    ("don't trust anything from that biased source", "ad_hominem"),
    ("this study's sample size is too small to draw conclusions", "legitimate_criticism")
]

# challenges:
# - context sensitivity
# - distinguishing relevant from irrelevant personal information
# - cultural and domain variations

def analyze_argument_structure(argument):
    premises = extract_premises(argument)
    conclusion = extract_conclusion(argument)

    personal_references = find_personal_references(premises)
    claim_references = find_claim_references(premises)

    if len(personal_references) > len(claim_references):
        if not assess_relevance(personal_references, conclusion):
            return "likely ad hominem"

    return "argument addresses claims directly"

attack: "you can't trust sarah's proposal because she's inexperienced"
response: "let's focus on the proposal itself - what specific problems do you see with it?"

attack: "you're biased because you work in that industry"
response: "i do work in this field, which gives me relevant experience. now, about the evidence..."

attack: "your personal life is a mess, so your business advice is worthless"
response: "my personal relationships don't affect whether this business strategy will work. let's examine the market data..."

attack: "you're too young to understand this issue"
response: "you're right, my brain won't fully develop until i agree with you. meanwhile, what do you think about [specific evidence]?"

weak: "that article is from a biased website, so it's wrong"
stronger: "that article's data doesn't account for [specific factors]"

weak: "since x is unreliable, we can ignore everything they say"
stronger: "x has been wrong about y and z, so let's verify their claims about w independently"

weak: "everything my opponent says is wrong because they're incompetent"
stronger: "my opponent makes a good point about x, but i disagree about y because [reasons]"

weak: "critics of this policy are just partisan hacks"
stronger: "the strongest criticism of this policy is [x], which we can address by [y]"

hierarchical context: "elder x says y" may be more persuasive
egalitarian context: arguments evaluated more independently of source

academia: personal credibility matters for expertise claims
journalism: source credibility is central to news evaluation
online forums: arguments often evaluated independently of usernames

adversarial systems: attorney credibility affects persuasion but not legal validity
inquisitorial systems: judge focuses more on facts than advocate characteristics

exercise 1: relevance sorting
given personal information about an arguer, determine if it's relevant to their claim:
- climate scientist with oil company funding discussing global warming
- divorced person giving marriage advice
- wealthy person arguing for tax increases
- doctor recommending medical treatment

exercise 2: rewriting
take ad hominem attacks and rewrite them as substantive critiques:
- "politician x is corrupt" → "policy x has these specific problems..."
- "that source is biased" → "this claim lacks supporting evidence because..."

exercise 3: recognition in context
identify ad hominem in political speeches, news articles, social media debates

def moderate_ad_hominem(comment):
    if contains_personal_attack(comment):
        if not assess_relevance_to_topic(comment):
            return "remove: irrelevant personal attack"

    if contains_harassment_patterns(comment):
        return "remove: harassment"

    return "allow: substantive criticism"

def evaluate_debate_quality(transcript):
    arguments = extract_arguments(transcript)

    ad_hominem_count = sum(1 for arg in arguments if is_ad_hominem(arg))
    substantive_count = sum(1 for arg in arguments if addresses_claims_directly(arg))

    quality_score = substantive_count / (substantive_count + ad_hominem_count)
    return quality_score

def summarize_debate_removing_fallacies(debate_text):
    arguments = extract_arguments(debate_text)

    # filter out ad hominem attacks
    substantive_arguments = [arg for arg in arguments if not is_ad_hominem(arg)]

    # summarize the actual positions and evidence
    return generate_summary(substantive_arguments)