Further Artificial Intelligence Misinformation: on Vadhryaśva

 

Further Artificial Intelligence Misinformation: on Vadhryaśva

Shrikant G. Talageri

 

Today my niece was telling me about what a wonderful app ChatGPT was in providing immediate replies to questions in an intelligent and almost human way, interacting like a real person would do. Here is what google says about this app:

“ChatGPT is a generative artificial intelligence chatbot developed by OpenAI and launched in 2022. It is currently based on the GPT-4o large language model (LLM). ChatGPT can generate human-like conversational responses and enables users to refine and steer a conversation towards a desired length, format, style, level of detail, and language.^[2] It is credited with accelerating the AI boom, which has led to ongoing rapid investment in and public attention to the field of artificial intelligence (AI).^[3] Some observers have raised concern about the potential of ChatGPT and similar programs to displace human intelligence, enable plagiarism, or fuel misinformation”.

I have to admit the app does seem to have many of the human-like characteristics listed above. The question is: on what kinds of issues can we expect this (or any other similar AI app) to give us really objective answers? I told my niece that I have no doubt this (or other similar AI apps) will provide answers on issues on which there is little or no doubt or controversy (e.g. “What is the capital of Venezuela?”, “Who was the tenth President of the USA?”, “What is a brontosaurus?”, etc.). But then, wouldn’t a simple google search also give us such information with equal accuracy (with or without AI support)?

What about more complicated and nuanced issues shrouded in controversy or which are the subject of heated debates? Can ChatGPT or any other similar AI app give reasonably objective answers in such matters? Surely, this (ChatGPT) app will only be able to give answers based on the data specifically fed into it? But my niece said that ChatGPT is supposed to scour all the information and data on the internet before giving its intelligent answer. She asked me what question I should put to the app.

 

I suggested that she ask: “What is the Mitanni evidence about the Rigveda?”. The app gave an answer mentioning the similarities (in names, deities, etc.) between the Mitanni data and the Rigveda showing the connection between the two. As this was vague, I asked her to put my name and ask what, according to me, the Mitanni evidence shows about the Rigveda. the app, in its answer, repeated the same answer (in different words) but added my name as having shown these similarities. I kept asking her to put more and more specific questions, but the answers showed little difference. When specifically asked what evidence “Talageri” had given about the Mitanni names, the app repeatedly claimed that I had used the Mitanni names Shuttarna, Artatama, and a few others in my evidence. When my niece (at my suggestion) repeatedly pointed out that I had never given these names as part of the evidence, the app answered “you are absolutely right” and gave different excuses and explanations (almost as a human would do) to explain why it had given the wrong data, and kept apologizing for having made mistakes, and added that it was not able to access the internet data about the names given by me. When my niece pointed out that I had given names ending in “atithi”, etc., the app again accepted its mistake and apologized, but said that I had given “Shuttarna” as one of the names ending in “atithi”. When my niece pointed out that the name “Shuttarna” does not end in “atithi”, the app again agreed and apologized, and gave a new name, “Artatithi”, that I have never heard of (and which in fact does not exist as a Mitanni name to my knowledge) as a name given by me. The “discussion” was getting more and more senseless and unending, with regularly repeated acceptances of error and apologies from the app. But, when even the data was not being given correctly, obviously there was no sense in expecting any answer or conclusions from the app, so I gave up this line of reasoning.

Then I thought of another word. I asked my niece to ask the meaning of the Rigvedic word/name Vadhryashwa. The answer was “vadhra” meant “strong”, and “ashwa” meant “horse”, so Vadhryashwa meant “strong horse”. My niece (on my prompting) pointed out that Vadhryashwa was not a compound of “vadhra” and “ashwa”, but of “vadhri” and “ashwa”; and that “vadhri” meant “castrated/impotent” and not “strong”. The app kept agreeing that “vadhri” meant “castrated/impotent”, but continued repeating that Vadhryashwa meant “strong horse”. It claimed that the words “vadhra” and “vadhri” are derived from the verb “vadh-” (to kill), and that Vadhryashwa therefore meant “a horse who kills”, and therefore a “strong horse”. This is wrong, since “vadhri” is not derived from the verb “vadh-” (to kill); and I could not locate any word “vadhra” derived from “vadh-”  (to kill) either in any dictionary. Finally, it turned out the app was taking a word “vadhar” meaning “a destructive weapon” as the first part of the compound name! After some more back and forth "chatting" we finally gave up. The app, till the end, translated Vadhryashwa as “strong horse”!

It must be noted that this was not ordinary old google search: ChatGPT is a sophisticated app powered by Artificial Intelligence, which continued, through a long discussion, to translate Vadhryashwa as “strong horse”!

 

So then, I went to the old and time-tested (and, not touted to be powered by AI) google search, and typed: “vadhryashwa meaning in Rigveda”. The answer I got was: “Vadhryaśva (वध्र्यश्व):—[=vadhry-aśva] [from vadhri > vadh] m. ' having castrated horses', Name of a man, [Ṛg-veda; Brāhmaṇa] etc.”.

In short, while ChatGPT continued, through a long discussion, to translate the first part of Vadhryashwa as “strong”, google search immediately translated it as “castrated”.

But this is only the first part of the story: it shows that google search, at least in this case, proved superior to ChatGPT.

 

But this story has a second part, which may be known to readers who have read my article on the subject. The question now is: was even google search really right in this matter? It must be noted that while the answers by AI apps (whatever may be claimed for them) are actually based only on the data fed into them, the same is the case with google search as well: it is ultimately based at best on the data fed into it which is based on the writings of the established academic scholars in the field. And when the established academic scholars in the field have unanimously made a wrong interpretation (which not only escaped the notice of all the scholars themselves, but never seems to have been brought to their notice before by anyone else either), then google search will also give the wrong interpretation.

In this particular case, google search (and the consensus among academic scholars, on which its results are based) is half-right: it correctly translates the first part of the compound word as “castrated” rather than as “strong” (as wrongly done by ChatGPT). But, it wrongly translates the compound word as a whole as “having castrated horses” rather than as “castrated horse”, based on the wrong consensus among academic scholars.

 

Have the academic scholars indeed wrongly translated the word as “having castrated horses” rather than as “castrated horse”: Why did they do so? And what is the evidence that all the academic scholars could unanimously have translated it wrongly?

1. Why the Academic scholars unanimously translated the word as “having castrated horses” rather than as “castrated horse”:

There were two reasons for this translation:

Firstly, it was believed that Vadhryaśva was the name of a king, the father of Divodāsa. It would be extremely unlikely that the parents of a king could have named their son by so insulting a name as “castrated horse”. So the scholars automatically decided that the meaning must be “having castrated horses”.

Secondly, the Rigveda has at least four names of persons ending in “-aśva” where the meaning is “having ….. horses”:

 

ṛjrá + áśva:   ṛjrā'śva (having reddish-brown horses).

I. 100.16,17;  116.16;  117.17,18.

VIII. 1.30. 

 

śyāvá + áśva:   śyāvā'śva (having pale brown horses).

V. 52.1;  61.5;  81.5.

VIII. 35.19,20,21;  36.7;  37.7;  38.8.

 

ví + áśva:   vyáśva (having no horses).

I. 112.15.

VIII. 9.10;  23.16,23,24;  24.22,28,29;  26.9,11.

IX. 65.7. 

 

ninditá + áśva:   ninditā'śva (having condemned horses).

VIII. 1.30.

 

So it was automatically assumed that this was another one of those names.

 

2. Why the Academic scholars were wrong in unanimously translating the word as “having castrated horses” rather than as “castrated horse”:

Firstly, in Vedic Sanskrit, compounds are of two types relevant here (distinguished by the accent):

The first type is the bahuvrīhi compound (where the two words together signify possession by the person to whom, or the object to which, the compound word refers. The accent is retained in the first word).

All the four above names (ṛjrā'śva, śyāvā'śva, vyáśva, ninditā'śva) are bahuvrīhi compounds, where the accent remains at the joining point of the two words (since the accent is in the last syllable in the first word, and in the first syllable in the second word). Hence the meaning in all the words is “having ….. horses”

The second type is the tatpuruṣa compound (where the first word qualifies the second as an adjective. The accent is retained by the second word).

The word vadhryaśvá is a tatpuruṣa compound, where the accent is retained in the second word (shifted to the last syllable for emphasis). Hence the meaning cannot be “having ….. horses”. It simply means “impotent/castrated horse”:

vádhri + áśva:  vadhryaśvá (impotent/castrated horse).

As the scholars realized, it cannot have been the name of a person given by his parents. But to avoid accepting that it meant “impotent/castrated horse”, the scholars (who believed it was the name of Divodāsa’s father) treated it as a bahuvrīhi compound meaning “having castrated horses” instead of a tatpuruṣa compound meaning “impotent/castrated horse” in violation of the rules of Vedic grammar.  

 

Secondly, the context makes the whole meaning clear. Vadhryaśvá was not the name of Divodāsa’s father: it was an insulting epithet applied to him because he was not begetting children:

The verse in which the word Vadhryaśvá appears is VI.61.1. It refers to the father of Divodāsa, before he became the father of Divodāsa, worshipping the Sarasvati, in order to beget a child, and being granted the boon of that historically famous son by the River Goddess. Clearly, the epithet refers to his natural childlessness as “impotency”.

[On the other hand, a final name of the very opposite meaning “powerful/virile horse” is exactly, and understandably, the actual name of a person in the Rigveda:

vṛ'ṣan + áśva:   vṛṣaṇaśvá (powerful/virile horse).

I. 51.13.

VIII. 20.10]

As if to emphasize and confirm that meaning of the word, the very next hymn in the Rigveda, in VI.62.17, refers to vadhrimatī the "wife of an impotent person", being granted the boon of a child by the Aśvins. The only other references in the whole of the Rigveda, to this vadhrimatī the "wife of an impotent person" being granted the boon of a child by the Aśvins, are in four of the latest hymns of the Rigveda:

I. 116.13;  117.24.

X. 39.7;  65.12.

 

The whole point is: neither the consensus views of earlier scholars (who may have got it wrong), nor the answers given either by google search or by the latest AI apps, are reliable guides in getting answers to questions in areas involving complicated and nuanced issues shrouded in controversy which are the subject of heated debates. Only an intelligent examination of the original sources can give the right answers.