How AI mistook Chhattisgarh truce move as religious leaders' appeal for Israel-Palestine peace!

By Rajiv Shah 

Today, I realized why one shouldn't fully depend on AI, which can, at times, be extremely misleading. What happened was, I uploaded a PDF on one of the AI apps that claims to be the best among those publicly available. The PDF had been emailed to me by Kavita Shrivastava, a senior activist associated with the People's Union for Civil Liberties and the National Alliance of People's Movements, both well-known human rights organizations.

I wanted the AI app, ChatGPT, to prepare a detailed summary of the PDF so that I could understand, in a nutshell, what it said. The email provided a very brief overview of the two-page PDF, which wasn’t sufficient for me to give a complete picture.

The email stated: "More than two hundred people’s organizations, activists, lawyers, academics, journalists, and concerned citizens from across the country have issued the attached public appeal, calling for an immediate ceasefire and dialogue between the CPI (Maoist) and the government."

It further mentioned: "The Adivasi-dominated districts of the Bastar division in Chhattisgarh, West Singhbhum in Jharkhand, and Gadchiroli in Maharashtra are currently at the epicenter of this conflict. The lives and well-being of the residents must be given top priority in any talks."

Noting that "the signatories have welcomed the CPI (Maoist) offer of peace talks and the Government of Chhattisgarh’s response, which has kept the door open for talks," it insisted, "the government needs to demonstrate its intent by stopping the war on the ground immediately."

The email concluded by stating: "The signatories have called upon both parties to keep the best interests of the Adivasis and other villagers in mind and to engage in peace talks while respecting the constitutional, democratic, and human rights of citizens within the broader framework of the Constitution of India."

The PDF, which ran over 1,200 words and included a long list of individual and civil society signatories, was uploaded to the app so I could receive a detailed summary. And what did I get in return? It first offered a small summary -- imagine! -- under the headline "Religious Leaders' Peace Appeal" in the context of Israeli-Palestinian conflict! 

When I said this was totally misleading, the AI modified it, strangely stating: "The document you uploaded is a public appeal dated April 4, 2025, calling for a ceasefire and dialogue in the Israeli-Palestinian conflict, signed by over 200 scholars and public figures from around the world". Except the line "signed by over 200 scholars and public figures", it was all rong. Even "from around the world" wasn't true.

Finally, I decided to open the PDF, copied its text—which was full of line breaks—and asked the AI to format the text without the line breaks.

This part, it did perfectly well, enabling me to upload a summary of the civil society statement on a site I handle, along with the full text and a link to the list of all the signatories.