Learning Krsna Consciousness in the age of AI
One pressing question our generation will have to answer is the role of AI systems in teaching Krsna Consciousness. Do they replace books? Do they replace a guru? To which extent can we rely on them?
Subscribe to receive new articles by e-mail. It’s free, but if you like, you can pledge a donation:
One pressing question our generation will have to answer is the role of AI systems in teaching Krsna Consciousness. Traditionally, spiritual knowledge is transmitted through books and living teachers, who can explain the knowledge contained in them. One can learn a lot reading books, but when one becomes serious about learning the spiritual science, he or she should approach a spiritual master, who can teach the deep meaning of the books. By the process of asking questions and being corrected by the guru, one has the opportunity of properly understanding the spiritual science and becoming an authentic representative of the disciplic succession. One can then, later on, accept his or her own students and continue the chain.
In the beginning, the Internet served mainly as a means to transmit the knowledge contained in the books. Different scriptures and commentaries became available online, and search engines helped us to find what we were looking for. Then discussion groups appeared, where we could discuss different philosophical points with others. These discussions, however, were not necessarily conclusive, because usually there was no master/disciple relationship, just different practitioners discussing amongst themselves. Discussion groups didn’t thus replace the need for a spiritual master; they just served as an extension of friendly discussions with peers.
Later, audio and video became widespread, and many spiritual masters started sharing their teachings only. It became thus possible to hear from different gurus and thus establish relationships with them. One would then accept one teacher as the principal guide and eventually take initiation from him, and accept other teachers as śikṣās. This type of interaction didn’t replace the normal relationship between guru and disciple, but it facilitated our contact with different spiritual teachers and made them more accessible.
We can see that up to this point, there was nothing really new. Different internet resources served as just extensions of the traditional system, facilitating access to books, friendly talks with peers, and the process of hearing from the spiritual master.
AI, however, poses as a disruptive technology. Now we have something that sits in between. It contains the knowledge from the books, but at the same time can answer questions and reason in ways that appear similar to a living teacher. Now, the process of enquiring, which was previously only possible with a living guru, can be done with an AI chatbot. What to do with this?
The first point is to understand what AI is. All AI systems, including ChatGPT, Gemini, Claude, etc., started as chatbots. These are applications developed to provide conversations that can mimic a human. The original goal for these chatbots was to pass the Turing test, convincing a person on the other end that he was talking with another human, and not with a computer.
The first chatbots were not realistic at all, because the softwares had just fixed questions and answers. This, however, changed when a new generation of softwares was developed, allowing them to be trained with content from the internet. Chatbots then started being trained to answer based on a vast volume of information (basically anything available on the internet). They started thus sounding a lot more convincing and prolific.
The problem is that chatbots have zero real understanding of the information they are processing. They just give answers based on their training, giving what appears to have a higher possibility of making sense, according to their training, and not necessarily what is correct. In other words, they regurgitate information available online and compose an answer mixing fragments of different sources, creating something that appears to make sense, but that is not necessarily deep or correct.
Imagine a mixture of conspiracy theories from Facebook, books of fiction, discussions from forums, Wikipedia, and Prabhupāda’s books. That’s more or less what it is. It’s possible to narrow the scope of the context of the AI to get more precise answers, creating thus personalized chats that offer better answers, like I did with the Prabhupāda’s servant AI, which is a narrowed version of ChatGPT programmed to answer based on Prabhupada’s books, but this approach has limitations, because the AI still answers according to it’s training.
Avoiding this problem would require creating a new AI system for devotees from the ground up, training the AI in all the books from Prabhupada and then other sources, and then processing all other sources of information according to correct principles and conclusions acquired in this initial training, as well as human input. However, producing a state-of-the-art AI system from the ground up costs billions of dollars, and without continued development, the system would become obsolete after a couple of years.
In any case, here are a few of the many limitations of AI systems:
a) Chatbots can “lie” because they will always try to give some answer, even when they don’t have the necessary information to answer it.
b) They can’t always distinguish between correct and incorrect information, and will mix content from different sources.
c) They can insist on an incorrect point, even when confronted with the truth, and can often be very convincing.
d) Most AIs are programmed to respect the copyright of books (including Prabhupāda’s books, which are still under copyright), which prevents them from giving exact quotes from the books in many circumstances. An AI can thus “quote” a passage from a book, giving a modified version of the text, instead of directly quoting what is written there. Apart from that, it also often misquotes.
e) AIs can hallucinate, inventing information that is not at all present in the books, just inventing completely new passages or concepts.
AI can thus give answers, but there is no guarantee that these answers are correct. Here is exactly where the first problem appears: many tend to see AI as some type of oracle that can answer all questions, accepting what it gives as truth, just because it appears to make sense, which is a serious mistake.
AI can be useful to brainstorm different ideas and concepts, more or less like we would with a friend. By definition, brainstorming means to regurgitate different ideas and explore different possibilities, which AI can do well. However, we then have to use our own brains to make sense of it.
When we talk with a friend, we hear what he or she has to say and then process this information according to what we know, and thus see to what extent we can accept it as truth or not. When the friend gives information we know is incorrect, we reject it, and when we don’t know, we ask some higher person who knows. Just as we don’t accept what all our friends tell us as the absolute truth, we should take what AI tells us with similar discrimination. It can help, but we need to compare the answers it gives us with what we know, and confirm with people who have higher knowledge.
AI systems are thus a rich but unreliable source of information. They thus don’t replace books, nor living teachers. We still have to study books and check the quotes and other information chatbots give us against the original texts, and we still need living teachers who can give us perfect knowledge.
There is an essential difference between a robot and a self-realized person. A pure devotee can see the spiritual reality and hear Krsna from inside the heart, and thus speak perfect knowledge from practical experience, while an AI, regardless of its level of sophistication, is limited to regurgitating material from different sources, some of it correct, some not.
What AI can do is to complement traditional search engines, as systems that help us to find information (although everything has still to be personally checked), and as a tool for brainstorming ideas, just as we could do with a friend, taking what the chatbot gives us as an opinion that can help us to get started, but not as necessarily true, and certainly not as complete. All answers still need to be checked against other resources.
AI systems can replace humans in certain tasks, but they also inherit the four human imperfections of having imperfect ways of obtaining knowledge, committing mistakes, being in illusion (literally hallucinating), and cheating. Just as we can’t trust an imperfect person to give us perfect answers, we can’t trust an AI. Just as before, when we finally become serious in understanding the spiritual science, we need to approach a bona fide, living teacher and inquire from him.
Just as one can’t become a doctor by studying under ChatGPT, one can’t understand the spiritual science from it. It can serve as a research or brainstorming tool, but real knowledge can be obtained only through the authorized process of approaching a self-realized soul, offering service, asking questions, and being corrected. In this sense, everything is still the way it used to be.
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!