As I read Venerable Sujato’s essay series I understood that he would not welcome AI translations even if they met the standard or even outperformed the accuracy of expert human translators in a double blind test of Pali. That is, I understood Venerable’s essays as revealing deep misgivings and aversion towards AI regardless of their accuracy. In other words, I don’t think what Snowbird has articulated as a reason for censoring AI output (a reason of inaccuracy) is the prime motivating factor behind Venerable’s viewpoint. 
2 Likes
And i for one am entirely unsuprised.
Jokes aside i appreciate you taking the time to respond to my replynto your post, howevernas i said, i have said what i have to say on this matter, better minds than mine can take it from here and good luck to them.
1 Like
I just re-read the second part for this point and from my impression of his other articles (it’s been a long time, maybe I’m wrong) is that I think Venerable doesn’t think AI is capable of translating like a human, period. Generally his criticism comes from the output looking vaguely human and good enough to fool people and he believes that that’s the purpose of AI, at least again, that’s my impression.
We are in disagreement then on this point friend. What you’re suggesting is an argument from authority. That people should follow the implied rules out of devotion or respect for the original founder of the site. Have you considered that this might be against the expressed wishes of Venerable? After all, he’s disclaimed oversight of the board in favor of community oversight.
Couldn’t it be that he intentionally turned over oversight and policy on the board so as not to be seen as an authority figure imposing rules without consensus? Couldn’t it be that rather than imposing rules with that essay series he was rather trying to convince people with reason and considered articulation for his viewpoint? In treating the essays as commandments perhaps what you’re doing is undermining the forming of the very consensus based upon reason and articulation that Venerable wishes to generate?
I fear that what you’re doing is potentially undermining future faith that a thumb was not put on the scale by the owners of the site.
Yes, but as has been stated the guideline is quite ambiguous and interpreted very differently by sincere readers in good faith.
1 Like
That’s an empirical claim looking like a Turing test for buddhist translations by AI. I suspect that Venerable could be proven wrong in the very near future from a double blind test. If human expert translators are not able to distinguish what is coming from an AI or another human, then the contention would be proven false. Let’s see.
I understand how it could be deeply unsettling for a human translator to find that a machine might be able to output a translation in such a way that they are not able to tell whether it was a human or an AI. However, censoring such translations because of this “creepy” or unsettling feeling is a very different motivation compared to what @Snowbird articulated as worry over misleading the masses.
Whatever the decision by the community I think any future guidelines or disambiguation should articulate clearly what the prime motivation is. If it is because they are “creepy” and unsettling so be it, but let’s be honest and transparent and articulate the consensus motivation.
3 Likes
I’m speaking from a place of organizational dynamics. I’ve spent years in organizations wearing various hats, as have many of us I assume.
I am particularly familiar with those organizational spaces where policies are in development and, in the interim, de facto practices are recommended. When I’ve worn the “chief” hat people expected me to recommend those. If I didn’t, people felt confused in the vacuum.
Some would take advantage but most people tried to follow the spirit of the de facto practice until a formal guideline or policy was established, by whatever process.
At this juncture it’s likely about transparency and, as such, there shouldn’t be issues of faith or confidence, necessarily.
Thank you, dear friend, for your discussion 
1 Like
This implies that those who post un proofread machine translation are at a higher vantage point.
But this comes back to the same stated problem, of those publicly posting ‘translations’ without any knowledge of the source language.
2 Likes
Exactly. And banned or allowed, let’s make the policy clear. To be honest, I won’t die if I’m unable to post AI output on these boards, I hardly have any reason to do so. I’m merely concerned (coming from a family of computer scientists) that the tech is so truly misunderstood and there’s a quite vocal and violent anti-AI sentiment that’s spreading to honest people. An analogy goes something like this:
Person A: I work for building planes!
Person B: Really? Don’t you know planes are used to bomb people? How can you work in such an evil sector?
Person A: No, honestly, our planes are used to transport goods and people to distant corners!
Person B: But that’s all planes are good, isn’t it? Bombing people? You’re clearly lying.
_
Etc.
4 Likes
Yes, and that is why I said that if ‘misleading the masses’ is the worry, then a future articulated guideline doesn’t need have any mention of AI. It is the case that human amateur translators can just as easily “mislead the masses” when they use no AI at all.
If we want to have a guideline against non-expert translations showing up on the forum (whether created by AI or amateur humans), then let’s be clear that is the intent. Of course, then the question will be what qualifies as ‘expert’, but once we’re clear on the motivation we can figure that part out.
1 Like
I feel that’s somewhat where this discussion is at at the moment. When did being categorically against a type of technology ever turned out to be the right thing? Even the, in the eyes of some people, controversial examples like nuclear energy, vaccines (oh gosh) or genetic manipulations desire a much more nuanced approach than just categorical banning.
I think the underlying situation why reactions by a few individuals on this board are so negative is mainly because Gen AI is eating away the moat of traditional scholars/treanslators and that can be a hurtful process to realize. Of course its not true that all experienced scholars in the larger field of Buddhism, or even the majority that I have talked with so far share this sentiment. Some see it as an opportunity for better scholarship, and some just simply don’t care. So far it seems to me that those who are vehemently and strictly against it are a rather small minority.
I largely agree with the perspective that GenAI systems are not good at tasks that lie completely out of the domain of their training data. The reality is that good scholarship on Buddhist material requires many years of study of already existing material and GenAI is a shortcut on this end. I wouldn’t expect GenAI to solve the translation of difficult Pali commentaries that nobody ever touched before, or some obscure Chinese Sutra with extremely difficult language. But that’s like saying ‘well, cars don’t go into space, so they are useless’.
If you use it for what its good for, it can be very productive. How much bad scholarship is out there because people don’t have the time, the knowledge, or are simply too arrogant to read publications of other people, or to be even aware of their existence. GenAI is going to be a huge game changer in this regard, and no its not going to replace humans from digesting this content and making informed decisions before publishing their own work. But its going to make identifying key information so much faster and easier.
I find it fascinating to observe this debate on this forum. Seems there is a crossroads now where the community can decide to either continue a reactionary path with a certain “cultural revolution 2.0” undertone, or be a liberal place where the use of this technology is seen as a complex issue that can’t be solved with a simple “just don’t” attitude. Alas, the fact that there is a number of people on this thread that have a more critical attitude makes me feel slightly optimistic.
4 Likes
To argue for the other side: We can trace the sources of biases of human translators, for example those of Ven. Thanissaro and TFT’s treatment of nibbāna as a kind of viññana, from a vast literature and tradition of actual experiences. Even if we disagree with their output, humans have an easier time analysing human biases than AI biases.
AI biases can be formed for a variety of reasons. Training data needs to be transparent, and even then, it’s quite limited to what the programmers thought was relevant. More data isn’t always more right either, so it’s kind of a weird situation.
And is the output (no matter how indistinguishable from a human output it is) more correct when it agrees with Ven. Sujato or Ven. Thanissaro? And we can always tailor the bot to lean one way or another.
In this fashion, LLMs built in abstract don’t tell too much (even though theoretically they could, and within 10-20 years some of these problems might be solved beyond my imagination), but for an experiential tradition like buddhism, arguing around any sort of translation or text from experience, is indeed relevant.
So that’s one way I think even a poor human translation might be understood better than a great AI translaton, because if I read Ven. Sujato’s arguments correctly, at least it’s based on true human experience, and thus its biases can be critically analysed based on our personal experience, something which can’t be said for machine translations.
Of course, a way to remedy this problem is to tailor the AI training data & output using monastics’ and meditators’ own experiential bias. But why bother doing that instead of doing it all manually? Well it could mean Ven. Sujato (or someone else) could translate their works into 100+ languages at once instead of just to English.
Whether that’s a desirable thing or not, is also up to debate.
1 Like
I think that this is the nub of the issue when ordained practitioners express their worry about ‘misleading the masses’.
Translation needs to express the meaning of technical terms in a way that can only be grasped experientially. Hence (just for example) what vitakka and vicara mean can only become clear when one has had an experience of Jhana - unfortunately, ordained practitioners are prohibited by the vinaya from actually coming out and being able to say ‘it means this because I / those I know who have experienced Jhana did so only when they applied it in this way’.
The finer the point, the more difficult it becomes to judge who is right or wrong - see for example the discussions revolving around the luminous/ unestablished Consciousness. The original sutta reference is the simile of a ray of light that does not alight anywhere. How the translation is worded can tilt the reader towards one or the other interpretation.
Of course, the human translator could have interpreted their experience wrongly, but at least that would be human… and hence inevitable and somehow forgiveable?
Unfortunately, most people would probably think of an AI translator as somehow infallible. Even though it has no way of actually understanding the human experience.
3 Likes
I came to Buddhist practice through meditation, not through sutta study or sutta exposure. So you have a point there. I feel we’ve already vetted (via the 16 essay threads) the unlikelihood of training LLMs or AI technology on subjective meditative experience.
The fork in the road, for me, is this precisely. If Traduttore Traditore is true – it appears everyone more or less agrees on that, in principle – then I would rather spend my time with human traitors than AI-trained ones because at least the human traitors are informed by human experience.
Of course, this does not have a corollary in more binary disciplines (science, medicine, eg.) where more and more data parsing supports remedies, insights, and solutions.
There’s no doubt that Gen AI puts some translators’ livelihoods at risk; however, I haven’t noted a preponderance of lay translators on this platform. Or I haven’t really heard that in the threads. People like me don’t have a dog in that fight; I’m not a translator.
There are some specific scenarios relative to this particular platform that are in play; at the same time, if someone is going to publish a rationale for keeping SuttaCentral AI-free, they are obliged to make it clear and emphatic. So, I feel that’s what we have.
2 Likes
Again, I don’t think forgivable or such terms are necessary. 
Human translation biases are understandable and rebuttable, that’s a good way to put it.
This kind of arguments and disagreements are useful, so we can work towards understanding what AI is and isn’t good for and why we’re against some of its applications (with good reason or otherwise).
We can have a nuanced perspective on a piece of technology without being offended by every instance of it being mentioned or indeed, facing abuse and violence just by the virtue of being involved in the field, I would hope. 
1 Like
Does this imply the Elite can quickly spot faulty translations inflicted upon the masses?
Even if they don’t know the source language?
Friend, I agree!
I consider calling attention to things and exposing them a form of public discourse. So long as we maintain ethical conduct in those conversations and, well, treat people the same way we want to be treated. That’s my standard, not that I always succeed. I welcome people holding me to account in this way.
Otherwise, we will all sit on pins and needles, not having these conversations. We are extremely limited in an online forum because we’re not with each physically. No one’s to blame for that; just how it is. I would love to be in a room with everyone talking and discussing this, sharing honestly, holding each other accountable. It would be so much less contentious.
There are specific scenarios, when we’re talking about this forum, which are considerably beyond “just mentioning [AI].” So, I’m glad that lots of eyes are on this. I always appreciate your input.
1 Like
I don’t think its about ‘Elite’ or otherwise… 
Someone who has had the experience in question can more easily judge the accuracy of a particular term that refers to it, don’t you think?
For example, if someone has ever actually had a migraine, they can quickly judge how a word like ‘throbbing’ is more on the mark rather than ‘aching’ when it comes to describing the pain. Even though the dictionary list goes as follows
Synonyms for Aching
sore
painful
hurting
throbbing
stinging
tender
smarting
Can an AI (or for that matter a human) who has never had the experience parse the difference accurately?
For the record, I personally feel that AI is a genie that is out of the box, which makes its appearance on our forum inevitable. Far better to make friends with the technology and promote useful and ethical development of this tool, which might one day even turn into a sentient (???) companion of humanity.
Speaking only for myself, I can say this has nothing to do with my reaction. And frankly the number of translators out there to even be replaced is vanishingly small. That’s really why we are in this situation. If there actually were masses of translators to lose their jobs then perhaps we wouldn’t even care about using these AI tools.
I don’t see it that way. Of course I would love to have a policy that kept “wrong” translations off the forum. But such policy would be impossible to create and enforce. However there is a subset of wrong translations that can be kept off the forum, namely machine translations.
It seems to me that you are the one making extreme accusations against those you disagree with. Please take it down a notch.
You have made very good arguments against posting machine translations verbatim here on the forum and that is all this discussion is about.
It doesn’t even have any way of understanding the material it is outputting.
What does that even mean in the context of the forum? Even in the framing you are taking it as a given that this technology is human enough to “make friends with.” In fairness you are probably just speaking in the colloquial sense, but I think it’s good to stay aware of these things.
Again, the discussion here is about posting verbatim machine translations on the forum, not the myriad other ways of using the power of computers to crunch large amounts of data.
This is true to a degree, but it is also true that there is a very problematic colonial history of academics inserting themselves and misrepresenting both the texts and religious practitioners.
But again I come back to the idea that if there is going to be academic discussion here on the forum then it should be held to basic academic standards. And relying on machine translation to make an argument (NOT just to find a text in a corpus you can’t read or to get an overview of someone else’s research in a language you can’t read) does not meet those standards.
Yeah, I mean that’s just the nature of the world. If we want to taste a certain kind of food we need to either learn how to cook it or find someone who can.
I feel like this is misrepresenting the position of people opposed to the posting of verbatim machine translations and is not helpful for the conversation.
I see not allowing the posting of verbatim machine translations as extremely pragmatic. It’s clear and simple. If someone doesn’t have the ability to translate a root text, then by definition they don’t have the qualification to be posting a machine translation. They are free to read those machine translations themselves. But if they want to talk about the text they need to take ownership of the fact that they can’t read the language and what they are basing their opinion on may be incorrect since they don’t have the expertise to actually judge it.
1 Like
It would not be hard at all to create and enforce such a policy. One option would simply be to whitelist a group of experts who are allowed to publish such translations. All others can apply to be in that group if they wish to publish here.
It has not been demonstrated that all machine translations are necessarily wrong. In fact, I’d like to see an example of a machine translation of an EBT sutta that is so dangerous as to require the blanket censorship. This to me would be a prerequisite for my personal support for such a policy. Has anyone offered up such a problematic translation?
1 Like
