Sense in Truth and Truth in Virtue

(Maat, 2375 BCE)

The title of this post comes from Shakespeare (Measure for Measure) but the epistemological concepts are much older and more complex.  The spectacular rise of large language models such as chatGPT and their capability to generate mindless babble has further clouded our access to objective truth. These language models cannot determine truth but the prose they generate pompously asserts some random notions from their training set that the models predict will satisfy the prompt.  Fake news, alternative facts, and clever lies have never been easier.

But what about search engines with instant answers?  Does duckduckgo, bing, or google generate "truth" when asked similar questions?  After all, these answers are curated by humans, right?  It turns out the humans are just as bad and sometimes worse.

The Rabbit Hole

I became curious about the origin of the concept that "virtue is its own reward."  I asked the internet search engines, "Who first wrote virtue is its own reward and in which publication did it appear?" The most common answer was Cicero with no cited publication.  (If you care, Cicero wrote it in De Finibus Moralibus (On Moral Ends).) But reading Cicero, it is obvious Cicero is citing the Greeks (He mis-cites Epicurus instead of the stoic who wrote it) for these ideas and a tiny bit of digging reveals Aristotle expressed the concept that virtue is its own reward earlier. Going further back, we find Xenophon wrote "εὐδαιμονία ἑαυτῆς τὸ μισθὸν ἐστίν" (virtue is its own reward).  Anaximander (the earliest published Greek philosopher), wrote about virtue in "για τη φύση" (On Nature). He wrote that virtue brings harmony, balance, and justice.  Other than eventually leading to happiness, virtue is not its own reward.  So Xenophon wins, right?  Winner winner chicken dinner!

But wait, these are just the Greeks.  What about earlier writings by the ancient Chinese, ancient Egyptians, or ancient Indians?  Confucious (Classic of Filial Piety) was no earlier than 206 BCE. The ancient Indian writings apparently have no reliable dates but historians think the oldest record of this concept is 200 BCE. The famous Egyption Book of the Dead, first inscribed on scarabs in 1650 BCE does have the notion that maat (virtue) is the reward for virtuous behavior. So this reference is the oldest.  Finally!

So the oldest published expression of the notion is ancient Egyptian, but the author  is unknown.  And the oldest cited philosopher is Xenophon.

Why are we forbidden from discovering the answer to this simple question?  Why are the search engines and chat bots all lying to us?

Garbage in Garbage Out

Inaccurate information is rampant on the internet.  Search engines and chatbots powered by large language models consume these lies. It should therefore be no surprise that these search engines and chatbots end up spitting out patently false answers in response to our questions. This phenomenon is especially concerning given the widespread use of search engines and chatbots in everyday life, from providing customer service to helping students with homework.

Search engines and large language models are not trained to assess the validity of the content they generate. Instead, they rely on the vast amounts of data on which they train to predict the most likely or relevant response to a given input. If the data crawled by the search engine or in the training set of the model contains false information, the search system or chatbot will emit the same wrong answer.

Confidimus in doctrina (Trust Scholarship)

This issue has been largely overlooked in both the commercial and academic AI communities, as the focus has been on the impressive ad profits and academic accolades these systems generate. We must remember that the accuracy of these results will always be called into question when it comes to determining truth.

It appears search engines and chat bots cannot be used for accurate scholarship. When it comes to determining the truth, we must still rely on scholarship and our own judgment.