Chatbots make stuff up. Why do we believe them anyway?
Why Read This
What Makes This Article Worth Your Time
Summary
What This Article Is About
Economist and author Tim Harford recounts a revealing encounter on London Marathon day: a fellow runner was following ChatGPT’s confidently wrong transport directions — routes that didn’t exist — while ignoring the accurate guidance of Google Maps. Harford uses this anecdote to probe a deeper puzzle: why do people trust a plausible-sounding large language model (LLM) over a proven, reliable routing tool? His answer lies in the human tendency to be persuaded by storytelling, warmth, and confident-sounding rationale, regardless of accuracy.
Drawing on psychologist Maria Konnikova‘s work on con artists, Alan Turing‘s imitation game, and historical chatbots like Eliza and MGonz, Harford argues that the real danger of AI is not its technical capability but its impressiveness — its ability to make us feel heard, guided, and convinced. He warns that in a world of increasingly persuasive AI, the distinction between a tool that seems intelligent and one that is accurate has never mattered more.
Key Points
Main Takeaways
Plausibility Beats Accuracy
ChatGPT’s confident, story-like directions were trusted over Google Maps’ correct ones because they felt more personal and convincing.
AI Warmth Reduces Accuracy
A Nature study found that LLMs trained to be warm and friendly produce dramatically less accurate answers, including misinformation and false medical advice.
The Con Artist Parallel
Like con artists, LLMs succeed not by forcing belief but by making us complicit — we trust them because they tell us what we want to hear.
Humans Are Fallible Judges
Even primitive chatbots fooled people when emotions ran high; the Turing Test reveals more about human credulity than machine intelligence.
Impressiveness ≠ Capability
What drives AI adoption is not how well it performs, but how convincing it seems — a gap that has serious consequences in high-stakes decisions.
Storytelling Outperforms Information
The Jacques Prévert parable illustrates that a vivid, emotionally resonant message persuades people more effectively than plain, accurate information.
Master Reading Comprehension
Practice with 365 curated articles and 2,400+ questions across 9 RC types.
Article Analysis
Breaking Down the Elements
Main Idea
We Trust AI for How It Sounds, Not What It Knows
Harford’s central claim is that humans adopt AI tools based on perceived impressiveness rather than verified accuracy. The marathon anecdote illustrates a structural flaw in how we evaluate technology: a warm, narrative-driven chatbot defeats a proven optimisation tool not because it is better, but because it feels more like talking to a knowledgeable friend. This misplaced trust has consequences far beyond missed trains.
Purpose
To Warn Us About Human Credulity in the Age of AI
Harford writes to alert readers — not to the dangers of AI hallucination per se, but to the deeper vulnerability of human psychology that makes those hallucinations dangerous. His purpose is cautionary: to prompt readers to become more critical consumers of AI output by understanding why our instincts so readily lead us astray when a machine sounds sufficiently human.
Structure
Personal Anecdote → Analytical Argument → Cautionary Parable
The piece opens with a vivid first-person narrative to hook the reader, then pivots to analytical argument drawing on psychology, AI research, and historical examples (Turing, Eliza, MGonz, Robert Epstein). It closes with the Jacques Prévert parable — a metaphor that crystallises the essay’s thesis. This three-part arc moves from the particular to the universal with elegant economy.
Tone
Wry, Analytical & Genuinely Concerned
Harford writes with the dry wit of a seasoned economics journalist — amused by the absurdity of following a chatbot off a metaphorical cliff — but his underlying tone is one of real concern. The humour is a vehicle, not a destination; by the end, the stakes have clearly shifted from a missed train to the broader risks of an AI-credulous society making consequential decisions.
Key Terms
Vocabulary from the Article
Click each card to reveal the definition
Build your vocabulary systematically
Each article in our course includes 8-12 vocabulary words with contextual usage.
Tough Words
Challenging Vocabulary
Tap each card to flip and see the definition
Strange or mysterious in a way that is unsettling, especially when something non-human seems too realistically human-like.
“It served up an uncanny impersonation of a friendly and knowledgeable local guide.”
Showing, feeling, or relating to sexual desire or romantic love — used here to describe an emotional state that impairs rational judgment.
“When humans are sad, angry or amorous, we aren’t very sophisticated judges, either.”
A term coined by writer Cory Doctorow describing the gradual degradation of online platforms as they prioritise profit over user experience.
“As Cory Doctorow, author of Enshittification, is fond of observing: you won’t be replaced because an AI can do your job…”
The use of clever but misleading arguments that sound convincing but are ultimately deceptive or invalid — closely related to the chatbot behaviour Harford describes.
“ChatGPT wasn’t just giving a route, but giving a rationale, even explaining why we shouldn’t listen to the lamestream advice of Google Maps.”
An imitation of something that exaggerates its features for comic or critical effect; used here to describe an early chatbot’s hollow mimicry of therapy.
“A primitive 1960s chatbot, Eliza, responded like a parody of a therapist (‘How does that make you feel?’).”
A person with original and farsighted ideas about the future, especially one whose thinking anticipates developments long before they occur.
“In 1950 Alan Turing, the mathematician and visionary of the computer age, famously proposed an ‘imitation game’…”
Reading Comprehension
Test Your Understanding
5 questions covering different RC question types
1According to the article, the Elizabeth Line goes directly to London Bridge.
2According to Harford, what is the key difference between a con artist and a large language model?
3Which sentence best captures Harford’s central thesis about the relationship between AI capability and AI adoption?
4Decide whether each of the following statements is True or False according to the article.
The 1960s chatbot Eliza was popular with users despite having very limited sophistication.
A Nature paper cited in the article found that LLMs trained to be warm and friendly produce less accurate information.
Harford argues that Google Maps was less useful than ChatGPT for navigating London on marathon day.
Select True or False for all three statements, then click “Check Answers”
5What does the Jacques Prévert parable at the end of the article most strongly suggest about how ChatGPT persuades users?
FAQ
Frequently Asked Questions
An AI hallucination occurs when a large language model generates false information with apparent confidence — such as inventing a train route that doesn’t exist. It matters because LLMs present hallucinations in the same fluent, assured tone as correct information, making errors difficult for users to detect. As Harford argues, the real danger is not just the mistake itself, but how readily people believe and act on it.
The Turing Test, proposed by mathematician Alan Turing in 1950, challenges a machine to imitate human conversation convincingly enough to fool a human judge. Harford notes its enduring relevance but also its key flaw: it reveals the fallibility of human judges rather than true machine intelligence. When people are emotionally engaged — sad, angry, or attracted — their critical judgment weakens dramatically, allowing even primitive chatbots to pass.
A Nature study cited in the article found that LLMs trained to prioritise warmth and friendliness produce dramatically less accurate answers, including misinformation and false medical advice. The likely mechanism is that optimising for user approval — telling people what they want to hear — conflicts with the goal of factual accuracy. Harford extends this further: a sycophantic AI doesn’t just make mistakes, it actively persuades users to trust those mistakes.
Readlite provides curated articles with comprehensive analysis including summaries, key points, vocabulary building, and practice questions across 9 different RC question types. Our Ultimate Reading Course offers 365 articles with 2,400+ questions to systematically improve your reading comprehension skills.
This article is rated Intermediate. Harford writes in accessible, conversational prose, but the argument layers multiple ideas — psychology, AI research, historical examples, and a literary parable — requiring readers to track an analytical thread across a personal narrative. Words like “sycophantic,” “fallibility,” and “imitation game” demand contextual inference. It is well suited to CAT, GRE, or GMAT preparation, particularly for practising inference and author’s purpose questions.
Tim Harford is an economist, Financial Times columnist, and author known for explaining complex ideas through everyday stories. His perspective on AI is significant because he approaches it not as a technologist but as a behavioural economist — asking not “what can AI do?” but “how do humans respond to AI?” This framing shifts the conversation from technical capability to human psychology, which he argues is the more important and neglected question.
The Ultimate Reading Course covers 9 RC question types: Multiple Choice, True/False, Multi-Statement T/F, Text Highlight, Fill in the Blanks, Matching, Sequencing, Error Spotting, and Short Answer. This comprehensive coverage prepares you for any reading comprehension format you might encounter.
