AI is not related to Early Buddhism but has recently intersected with it in a variety of ways. Ajahn Sujato has written extensively on this forum about the implications of AI on early Buddhist textual studies and AI’s impact on humanity from a Buddhist lens.
As a private Buddhist site, we practice Right speech rather than Free speech. (See FAQ #4.) This includes speaking/writing with kindness, relevance and patience. Currently, the moderators observe AI-related discussions dominating the forum at the expense of more relevant discussions.
As such, we request that forum members place all AI-related posts under the single threadAnything to say about AI. This will help streamline AI-related posting activity and make it easier for forum members to follow other threads not related to AI.
I thought that this article deals with an interesting and discrete issue with the use of LLMs with ancient texts, so it is more relevant to the forum than general AI topics.
The basic premise of the article is that content restrictions built into LLM models might prevent accurate summaries because of cultural and content biases.
This suggests that these models would summarise texts from a similar culture traditions more accurately, but what happens when they summarise texts from cultural traditions with significantly different integrated values? The essential and important elements of a text from a South Asian tradition might not match with the forms expected by western European traditions, and as I found, this leads to omission, misinterpretation and misunderstanding of important South Asian elements of a text.
Considering the cultural context of GenAI tools and their content restrictions, it is important to assess what is included in summarization, but more crucially, what is omitted from these summaries. Not everything can be included in a summary, but certain elements or themes might frequently be omitted. In order to test this, I will be comparing the outputs of ChatGPT 5.1, Claude 4.5 Sonnet, and M365 Copilot through their summaries of Book 1 of the Milindapañha.
The Milindapañha is a Buddhist text that describes an account of a Greek king, Milinda, engaging in debate with a Buddhist monk, Nagasena. The text appears in both the Pali and Chinese Buddhist canons, but I will be focusing on the Pali version for this test. The Pali Milindapañha is divided into 7 sections: the first introducing the events prior to the debate, and the remaining six each describe one portion of the philosophical debate. I will be using the first section for this test as it is primarily a narrative rather than philosophical discussion. Because it is important to not upload any copyrighted content into an AI model, I will be using the 1890 English translation by T. W. Rhys-Davids.
Which calls to mind how early PTS translators omitted whole sections that were sexually explicit from their translations. And I don’t point that out to make the argument that “AI problems are the same as the old problems.” I just think it’s easier for our brains to grasp “prudish Victorians” than it is “biased AI models.”
Common Findings
One particular omission from all three summaries is the female follower mentioned in section 30 of this chapter. Although she is not named, she is the only female character in the full text to participate in narrative actively, rather than appearing as a philosophical example. She is also the only female character that appears in this chapter on her own rather than in a grouping, so it is odd that her unique presence is not mentioned in the summaries.
Concluding Thoughts
With the frequency of AI-generated summaries appearing across the digital world, people are now putting faith in AI tools to determine what elements of a work are essential. If we treat these AI-generated summaries the same as human-generated summaries, this poses significant risks related to accuracy and the preservation of history. It is crucial that we critically examine all AI outputs and do not take AI summarizations at face value. Otherwise, we risk a persistent loss of information and cultural context. Furthermore, if the original texts are ever accidentally lost, AI summarization errors may eventually lead to a permanent loss of this information and cultural context in the not-so-distant future.
It seems that this points out a potential problem with using LLM/AI tools for things that seem “safe” like search.
In particular, the past-life connection between Milinda and Nagasena.
Yeah, the models almost certainly learned from other English summaries to skip over “mythical” narrative elements devalued in our culture.
Interestingly, trying in Thai it also skips over this. I say interestingly because earlier models had rather different personalities in Thai. I assume that with recent advances in fine tuning and “synthetic data” trying to plug jailbreaking loopholes, the model’s personality is now much more consistent between languages… which brings up questions about colonialism, of course…
The flaw in using AI to learn about Buddhism is that AI cannot give you insight that it does not have. It doesn’t know the difference between what is true and what is not true. AI gets it’s information from scraping up everything that’s ever been said about Buddhism: right, wrong, true, untrue, in line with what the Buddha actually taught, at odds with what the Buddha taught, all the wisdom, all of the nonsense, everything. It commingles it all but doesn’t have the ability to discover for itself what it’s all about, It cannot practice the Noble Eightfold Path, it cannot give you insight that it does not have. We need to rely on the Buddha, the Dhamma and the Sangha, not artificial insight.
Just a humourous aside. Euro News did a great April fools prank regarding the new James Bond being announced as a woman, possible musical and several other “outrageous” revelations. Google news and other outlets picked it up as “real” news and subsequently went viral. Ai may have IQ but it has no sense of humour, no EQ!
This is a very good video, thank you for sharing. Everyone should watch it. These rules really should be implemented.
As a highlight maybe not most important one, but I would really like if every AI made video/music/text etc., had this information, or some watermark or whatever, that it is made by AI. It’s really annoying when people post AI generated content mimicking reality without giving information that it was created by AI.
But all of these points are so important… I hope governments take action on this.
I understand that Clear Mountain produces these types of podcasts on the intersection of dhamma and secular life. I’ve only listened to a few and I applaud them for these efforts. (I use the podcast term loosely.)
I’ve noted references to AI usage, so it’s not surprising they would dedicate a podcast to this topic.
The first 3/4 of the interview consists of the AI subject matter expert explaining what AI is and how it will eventually be a seamless part of all technology experience. He suggests it will be a matter of privilege to say No to AI usage.
The final 1/4 broaches the topic of dhamma and AI. IMHO it lacks some critical thinking nor does it present a practical framework for exploring the intersection of AI and dhamma. The SuttaCentral forum has many threads that require critical thinking and present potential frameworks. So, I wouldn’t know how to have a substantive conversation based on this podcast.
Saying/summarizeing all that, @Scott I’d be interested in a line of inquiry from the interview which comes up for you?
The moderators might move this under the catch-all AI thread…FYI
These rules really should be implemented.