Gandhipedia digital portal doesn’t live up to expectations. AI can’t be trusted with history

The errors carried in Gandipedia resource materials raise concerns about the credibility of the digitised portal.

Credit: Gandhipedia portal

Gandhipedia, a digital tribute to MK Gandhi, was first proposed by Finance Minister Nirmala Sitharaman in her 2019 Budget speech. After over four years, the portal was officially launched by Union Minister Arjun Ram Meghwal on 27 December 2023. The platform is described as an AI-enabled “digitised resource for browsing Gandhian literature, his life events and social networks”.

There are at least two other portals dedicated to Gandhian resources. The gandhiheritageportal.org of Ahmedabad-based Sabarmati Ashram Preservation and Memorial Trust, and mkgandhi.org of Jalgaon-based Bombay Sarvodaya Mandal and Gandhi Research Foundation. Both portals offer a vast collection of Gandhi-related content, with new material being continually added. As of January 2024, Gandhi Heritage Portal hosts over 25 lakh scanned pages, of which more than five lakh have been OCR-enabled to facilitate search within text.

In contrast, Gandhipedia primarily relies on the 100 volumes of the Collected Works of Mahatma Gandhi in three languages, and seven books are identified as ‘key texts’. The portal’s homepage states that it was developed by the Ministry of Culture, in collaboration with IIT Kharagpur, IIT Gandhinagar, and the National Council of Science Museums. However, it remains unclear whether historians or Gandhi experts contributed to the development of the content or the application of technology to the content. The target audience of this engineer-driven project is also unknown.

Gandhipedia aims to enhance the user-friendliness of the Collected Works of Mahatma Gandhi (CWMG) volumes through the use of technology. However, the outcome does not live up to expectations or the standards of a credible source.

Gandhipedia—errors and AI

Here are the primary issues with Gandhipedia:

1. Limited scope: Despite the significance of the CWMG, which forms the main content of Gandhipedia, Gandhian literature is extensive, encompassing everything from Mahadev Desai’s diaries to memoirs, biographies, and numerous other books and journals. The CWMG section on the Gandhi Heritage Portal has 3,18,336 pages, and it constitutes a small fraction of the Gandhian literature available on the site.
2. Omissions in key texts: The ‘Key Texts’ of Mahatma Gandhi section features seven books in three languages—English, Hindi, and Gujarati. This includes The Story of My Experiments with Truth, Hind Swaraj, and Satyagraha in South Africa. However, all these books start directly from the first chapter, omitting the cover page, back cover, print line page, introduction, preface, table of contents, and any illustrations within the chapters. This deprives readers of understanding the context and purpose behind Gandhi’s writings. In the case of his autobiography, without the mention of the first edition’s year of publication, the readers are left wondering why his experiments with truth abruptly ended in the 1920s.
3. Secondary sources: Gandhipedia’s ‘Key Texts’content is sourced from mkgandhi.org and is presented in a plain text format, rather than scanned pages of the original books. This has led to errors in text reproduction. For instance, the original Hindi title of Gandhi’s autobiography is Satya Ke Prayog Athava Atmakatha, but on Gandhipedia, it is listed as Satya ke Saath Mere Prayog Athava Atmakatha. This discrepancy in the title of such a renowned work raises concerns about the accuracy of the rest of the text. The Gujarati version of the autobiography contains merged paragraphs and numerous proofreading errors, suggesting that the content from mkgandhi.org may not have been cross-verified with the original text before being included on Gandhipedia.

4. Inadequate introduction: Upon clicking on a book’s name in the Key Texts section, readers are presented with a brief and insufficient introduction. It lacks crucial details such as the year and context of the writings. For example, the 93-word introduction for Gandhi’s autobiography fails to mention South Africa, only stating that it covers his time in England, among other things.

5. Unfriendly presentation: The books on Gandhipedia are segmented into chapters, requiring the reader to click on each chapter tab to proceed, which disrupts the flow and continuity while reading the book. Additionally, the text cannot be zoomed in or viewed in full screen. The Gujarati text uses fonts that are not conducive to extended reading and contains spelling errors, particularly in some conjunct letters. Moreover, the justified spacing makes the text difficult to read, and the reader needs to scroll vertically instead of turning the page.
6. Inaccurate results: The ‘Key Persons’ sub-section under ‘Key Mention’ features a brief list of nine names, including Jawaharlal Nehru and Sardar Vallabhbhai Patel. Upon clicking a name, the reader is shown a list of CWMG volumes that mention the particular individual. For example, the list of Patel’s mentions is long, spanning from volume 14 to 97. But Gandhipedia doesn’t mention volumes 16, 30, 31, 39, 54, 78, 79, and 92. This may lead an unsuspecting reader to believe that Patel is not mentioned in these volumes. However, in the persons’ index volume of CWMG (Volume 99) available in a library or on the Gandhi Heritage Portal, mentions of Patel’s name are found in all these volumes. The reasons for such omissions on Gandhipedia are unclear.

The collected works have been divided into various useful sections on the portal, such as letters, speeches, telegrams, interviews, etc., to narrow down and specify the search. But a sample search for Gandhi’s interviews in CWMG yields a total of 84 volumes—without the mention of volume 20 in the search results. Upon closer examination, it is revealed that this volume does include interviews such as the one with the Hindi publication ‘AAJ’ (CWMG 20, page 483).

7. Lack of editorial intelligence: In the ‘Search CWMG’ section, ‘Vallabhbhai Patel’ yields 75 volumes, while ‘Sardar Patel’ only brings up 27 volumes. The AI doesn’t seem to have been programmed to recognise ‘Sardar Patel’ and ‘Vallabhbhai Patel’ as the same individual. When a user searches for the ‘Gandhi-Irwin Pact’, the first result is in volume 55 (April 1933-September 1933), which is two years after the pact was signed. This is because it was referred to as ‘the settlement’ or ‘provisional settlement’ at the time, not the ‘Gandhi-Irwin Pact’. Searches for ‘Salt March’, ‘Salt Satyagraha’, and ‘Dandi March’ return 5, 7, and 18 volumes, respectively.

While Gandhipedia attempts to introduce innovative methods to handle a substantial amount of content, it appears to be a hastily completed project with minimal oversight of content accuracy and an overemphasis on technology. This warns us of the potential dangers of using AI injudiciously and underscores the wisdom that we should never rely solely on technology for the handling of historical material.

Urvish Kothari is a senior columnist and writer based in Ahmedabad. He tweets @urvish2020. Views are personal.

(Edited by Ratan Priya)