A dangerous experiment involving AI (and its proper usage)

Can AI be used to write transcendental literature? A dangerous experiment reveals the shortcomings of this idea.

Subscribe to receive new articles by e-mail. It’s free, but if you like, you can pledge a donation:

A few weeks ago, a strange article was submitted on iskconnews.org, entitled Prabhupada’s Silicon Sutras: When 18 Years of Vision Meet the Speed of Now. The article sounds partially incongruent, as if written by someone partially drunk or high. The reason is that it was not written by a human being, but by Gemini, Google’s AI.

How is that an AI article could have ended on iskconnews, one may ask. That’s the interesting part: It was submitted there by a human being, Visnu Murti Das, Founder of Vanipedia, to announce they plan to use Gemini to expand the content of the website, with the goal of producing 1,000 new articles during the past December.

The announcement in itself was produced using Gemini. Probably the idea was to sound cool and show how AI could generate the announcement by itself, but it ended up just exposing the shortcomings of the idea.

As we can all observe, AI can sometimes produce relatively good texts, but most of the time, the output is only partly accurate and coherent. A person who knows the subject can detect the incongruences, fix the text, and come up with something okay at the end. However, the same person could just write a text from the ground up by himself; one who knows a subject doesn’t really need AI to speak or write about it. AI, in this case, may be used as a tool to help in some stages of the work, like research or revision, but not as a magic box to produce finished articles.

The problem starts when someone doesn’t really know something and wants to use AI as a shortcut to produce articles and books explaining it. That’s where all the shortcomings of AI come to bite us. If I don’t know a subject or just have some idea about it, I will not be able to spot incorrect information, flawed logic, improper rhetoric, and other problems in the AI-generated texts.

As far as I could understand, the idea was to feed the context of Vanipedia into Gemini and use it to automatically create articles on different topics that would then be curated by a human team and then published on the platform, dramatically expanding the available content in a short span of time.

There was, for example, an article entitled Tea or Husband? - The Moment That Changed History, which described the situation of Prabhupāda telling his wife to choose between tea and him, the defining moment after which he decided to enter into retired life, leading him to eventually come to the West and fulfill his mission. The short article, however, had several issues with tone, narration, conclusion, and even incorrect historical facts, apart from the blatant superficiality of most AI texts, and was later taken offline. Now, when you access it, there is just a blank page. It seems that most, if not all, of the previously published AI articles had a similar fate.

We may use AI in our work, just as we use any other tool. I myself use AI to speed up many tasks. It helps me to find specific verses and quotes (that I need to check manually afterwards to detect the periodic hallucinations), it helps me find the word meaning for Sanskrit words when I’m translating, helps with fact-checking, grammatical revision, and so on. However, the writing itself I do using my own fingers.

AI works well when it is taken like an intern, whom you can give some tasks and then check the results before using it, or as a buddy with whom you can brainstorm and toy with different ideas and concepts in the hope of getting some insight, but not as an intellectual overlord that does all the work by itself. Not even Paramātamā accepts this function: He gives the intelligence, but we have to use it.

On top of that, there is another dimension, which is the difference between ordinary text and transcendental text. Ordinary text is just a block of information. In this sense, it doesn’t matter who or what writes it, as long as the information is correct. Transcendental text, however, is a transformation of transcendental sound. Just as we can’t put a chatbot to sing a kīrtana, or a robot to worship a deity, we can’t use AI to write transcendental text. It can be produced only by a human being, a devotee, who shares his or her spiritual realization through words.

In this context, AI can be taken more or less like an evolution of databases, search engines, and other technologies we were already using previously. Just as a devotee would use the Folio in the 1990s to find quotes and references, combined with a spellchecker and other tools, one may use an AI now, but the writing itself has to be made by a devotee.

Some become dazzled by the apparent “human” responses given by modern AI systems without properly understanding them. AI doesn’t think in the way a human does, what to say about having spiritual realization. It is just a machine using fuzzy algorithms to manipulate information. All it does is combine different fragments of text from different sources and different pieces of information to create a more or less meaningful narrative. Just like using AI to create illustrations results in all kinds of aberrations (we often have to create several to find one that more or less fits what we need), generating text through it is not very different. The point is that when we generate an image, it is easy to spot a man with three arms and seventeen fingers, but it is not so easy to spot incorrect information in a text on a topic we don’t fully comprehend.

If one wants to fill up a database or website with one thousand articles, he can just take from my website; there are 1,000 articles there that I wrote through hard work. Or one can take from the Prabhupāda-līlāmṛta or other sources. There is no scarcity of devotees writing articles and books. Instead of insisting on using AI, which is a dead-end in terms of transcendental literature, one should promote articles written by real devotees who are genuinely trying to share their spiritual realization.

You can also donate using Buy Me a Coffee, PayPal, Wise, Revolut, or bank transfers. There is a separate page with all the links. This helps me enormously to have time to write instead of doing other things to make a living. Thanks!