close_game
close_game

Best book recommendations; and why humans score over ChatGPT

Feb 09, 2024 05:49 PM IST

Ram Chandra Guha’s India after Gandhi made it to all the lists along with P Sainath’s Everybody Loves a Good Drought

Over the years, Five Books has earned a reputation among bibliophiles the world over for a reason. The team asks domain experts on various themes to recommend five books. Last month, two of the editors there ‘interviewed’ ChatGPT and asked it to recommend the best books on Artificial Intelligence. It did. More interesting still, ChatGPT reasoned with the editors on why it chose those books. It is tempting to conclude this is scary.

Author Sukety Mehta’s acclaimed book, This Land is Our Land: An Immigrant’s Manifesto. (Amazon)
Author Sukety Mehta’s acclaimed book, This Land is Our Land: An Immigrant’s Manifesto. (Amazon)

But Is it really so? To figure out, I interviewed ChatGPT and other Large Language Models (LLM) such as Google’s Bard and Copilot from Microsoft for books on India. The answers were most satisfactory. So much so that it was tempting to put it out in the public domain. There were many overlaps as well. Ram Chandra Guha’s India after Gandhi made it to all the lists along with P Sainath’s Everybody Loves a Good Drought and Amartya Sen’s “The Argumentative Indian”. There was a surprise as well—Suketu Mehta’s “This Land Is Our Land: An Immigrant’s Manifesto” that ChatGPT put out.

When asked on how does the book make it to the list, ChatGPT argued that “Though it’s a global narrative, Mehta’s Indian heritage and insights into the immigrant experience are profound. The book is a powerful commentary on immigration, colonialism, and global justice, that when looked at, offers a perspective that resonates with contemporary India.” This was difficult to argue against.

Some tinkering around later, it occurred: Why not do away with human reviewers and get these tools to curate books, films and music lists? The idea met with resistance from multiple quarters.

Also Read: Inside job: Charles Assisi offers tips on how to bid on a better you

First off was my colleague NS Ramnath who asked: What’s the value add here? His question was interesting. Anyone who knows to create the right prompts can extract decent output from LLMs. So, while Suketu Mehta’s book appears on ChatGPT, it does not appear on Bard or Copilot. It is an outlier. But the same prompts contained an element of sameness as well to the books it recommended. To place that in perspective, Guha aside, in non-fiction, authors such as Amartya Sen, P Sainath and James Crabtree made it to most lists. Clearly, the algorithm powering the engines had made up its mind and could not see beyond that. So, what’s the value add?

This is where Ramnath made an interesting observation. If this question on books are posed to a full-time Indian Tamil language writer such as Badri Seshadri, chances are, he would come up with an altogether different list. This has to do with the fact that his exposure to Indian languages is higher than what LLMs are trained on.

Two researchers at IISER who work on LLMs in Pune concur. One of them made the case that “LLMs such as Copilot are great to code on and take the grunt work away.” He pointed out that his students routinely use it and they are encouraged to do it because it cuts down on their work hours by at least half. This is time that would otherwise go into fundamental research. However, there’s a catch: “When it comes to writing research reports and papers for journals to be presented at international seminars, LLMs get it all wrong. They imagine the citations and make things up.” Both of them have witnessed this consistently and agreed that “human hand-holding” is required.

This takes us to the ethics of it all. In their current state, LLMs do not generate original thoughts or opinions. They mirror what they’ve been fed. And this is the aggregate opinions and information from various sources. This mirroring raises concerns over biases and conflicts of interest. When AI recommends a book or a film, it’s essentially reflecting the consensus or the dominant narrative. In doing that, it is potentially side lining the niche, unconventional, or emerging perspectives that a human expert might catch and appreciate.

So, if recommendations on philosophy are asked for, the more well-known authors from the Western World such as Plato or Nietzsche will emerge. The lesser-known and equally insightful contemporary people whose work haven’t yet permeated the AI’s database won’t make the cut. It’s similar to asking for a song recommendation and receiving safe choices, undoubtedly classics, but not the hidden gem you were hoping for. This is why places such as Five Books people such as Badri Seshadri will continue to matter.

Get Current Updates on...
See more
Get Current Updates on India News, Weather Today, Latest News and Top Headlines from India.
SHARE THIS ARTICLE ON
Share this article
SHARE
Story Saved
Live Score
Saved Articles
Following
My Reads
Sign out
New Delhi 0C
Tuesday, December 03, 2024
Start 14 Days Free Trial Subscribe Now
Follow Us On