"Newspaper Archives Expose AI Tools' Factual Errors on Historical Events"
Hallucination (AI)
LLM generating confident but factually incorrect outputs due to statistical prediction rather than verified retrieval
Large Language Model (LLM)
AI system trained on vast text corpora to predict and generate human-like language
Artificial Intelligence (AI) language models — commonly called Large Language Models (LLMs) — are trained on vast corpora of text data scraped from the internet, digitised books, and other sources. Their ability to generate fluent, confident-sounding responses has made them widely adopted as information retrieval tools. However, a structural limitation of LLMs is their tendency to produce hallucinations — plausible-sounding but factually incorrect outputs — particularly when queried about historical events that are underrepresented in their training data.
In India, the governance of AI-generated content intersects with the Information Technology Act, 2000, the IT Rules, 2021, and evolving proposals under the Digital India Act. The IT Rules, 2021 were amended to require platforms to flag and remove AI-generated misinformation, and the government has issued advisories to AI companies to label outputs that could be false or misleading. Globally, the EU AI Act and the AI Action Summit (co-chaired by India and France) have placed AI accountability — including transparency about model limitations — at the centre of international governance discourse.
Legacy media institutions, by contrast, maintain structured, timestamped archives of primary reporting that predate the digital era. These archives represent an irreplaceable epistemic resource: they record events as they were understood at the time, without the retrospective distortions that can enter secondary sources. The tension between AI-generated convenience and archival accuracy is therefore not merely a technical question but a governance and epistemological one.
The core limitation exposed in this article is what AI researchers call hallucination or confabulation — the generation of confident, syntactically coherent responses that are factually wrong. This arises from several structural features of LLMs. First, their training data is heavily skewed toward content that is widely indexed and linked online; pre-digital newspaper reports, even if historically significant, are systematically underrepresented. Second, LLMs do not retrieve facts from a verified database — they predict the most statistically probable next token given a prompt, which means popular but incorrect narratives can outcompete accurate but obscure ones.
A second dimension is algorithmic bias in information retrieval. Search engines and AI tools are shaped by engagement signals, recency bias, and the volume of corroborating sources. A recent high-profile event (such as a contemporary leader's visit) generates far more indexed content than a 1958 presidential visit, making the recent event the statistically dominant answer even when it is historically incorrect. …
Consider the following statements regarding the AI Action Summit held in Grand Palais, Paris in February 2025:
I.Co-chaired with India, the event builds on the advances made at the Bletchley Park Summit held in 2023 and the Seoul Summit held in 2024.
II.Along with other countries, US and UK also signed the declaration on inclusive and sustainable AI.
Which of the statements given above is/are correct?
With the present state of development, Artificial Intelligence can effectively do which of the following?
1.Bring down electricity consumption in industrial units
2.Create meaningful short stories and songs
3.Disease diagnosis
4.Text-to-Speech Conversion
5.Wireless transmission of electrical energy
Select the correct answer using the code given below:
With reference to the limitations of AI language models, consider the following statements:
1.AI hallucination refers to the generation of factually incorrect but syntactically fluent outputs.
2.Training-data bias can cause AI tools to favour recent, widely-indexed events over historically accurate but less-represented ones.
3.Semantic search and AI language models use identical underlying mechanisms to retrieve information.
Which of the statements given above is/are correct?
Consider the following statements about the IT (Intermediary Guidelines and Digital Media Ethics Code) Rules, 2021:
1.Significant Social Media Intermediaries are defined as platforms with over 50 lakh registered users in India.
2.These rules require SSMIs to remove deepfake content within 36 hours of a government or court directive.
3.The rules apply equally to all digital intermediaries regardless of their user base size.
Which of the statements given above is/are correct?
Assertion (A): AI language models can produce confident, detailed responses about historical events that are factually incorrect.
Reason (R): AI language models generate outputs by predicting statistically probable text sequences based on training data, rather than retrieving verified facts from primary sources.
Which of the following is correct?
With reference to international AI governance, which of the following statements is/are correct?
1.The EU AI Act classifies AI systems by risk level and is the world's first comprehensive binding AI regulatory framework.
2.The AI Action Summit held in Paris in 2025 was co-chaired by India and France.
3.UNESCO's Recommendation on the Ethics of AI is a legally binding instrument adopted by its member states.
Want the full note, Prelims Facts & MCQ answers?
Create a free account — takes less than a minute.
Create free account