Language AI in the News: Why NLP Makes a Better Judge of the Past than Guide to the Future

Last month, CSOFT looked at how language AI is bringing clarity to translations not only of present day content but also relics from the distant past – specifically, by deciphering text artifacts bearing symbols from the earliest known precursor to ancient Greek. Now, in an odd twist of developments from across the AI world, natural language processing (NLP) technology has gone further yet in explorations of the past with a new algorithm from DeepMind that is helping historians clarify mysteries from the age of myths. Simultaneously, an unrelated algorithm named for the oracle of Delphi has recently demonstrated the limits of modern ambitions to deploy AI in ethically sensitive settings like advice and counseling, as yet another model trained on the things people say online comes to reflect their prejudices. Seen together through their thematic overlap, the same technology that many are looking to as a predictive window for the future is suddenly helping enlighten the past from which so many of the words and images we have to describe its awesome power originate, while having also provided case in point for why the ancients were right to doubt the Promethean hubris of all who would wield godlike powers such as AI.

The Verge reports that DeepMind’s Ithaca algorithm “helps not only restore text that is missing from ancient Greek inscriptions but offers suggestions for when the text was written (within a 30-year period) and its possible geographic origins.” Whereas our previous feature highlighted the ways AI can gather semantic meaning from symbols that are essentially hieroglyphic in nature, DeepMind’s instance of this type of machine learning has an additional value in terms of its ability to recognize a specific text’s place within a broader, evolving language tradition. This would be impressive in itself, but the concrete value of this kind of insight is that many of the same relics where these texts appear are challenging to carbon date. In short, with language AI in the picture, the ancient world is undergoing a kind of second excavation.

What of the oracle of Delphi, then? While a somewhat less recent news item, a late 2021 development also reported by The Verge highlights a wholly failed effort to apply NLP in an advice setting, in which people can pose ethical questions that, while debatable, would tend to generate a predictable range of responses from people when asked, in terms of our shared sense of humanity. Unfortunately, our data does not always reflect the same values proportionally, and the chatbot Ask Delphi has proven a dubious moral authority for having formed its opinions based on training on online communications. As the original article puts it, “If you add the phrase ‘if it makes everyone happy’ to the end of your statement, then the AI will smile beneficently on any immoral activity of your choice, up to and including genocide.” In other words, if AI is becoming our oracle, then we have only ourselves to blame for the lack of virtue in its most outlandish advice. All of which only highlights the power of language itself: said one way or another, a superhuman intelligence can still be led to very different notions of what people mean, in terms of their fundamental values across the bigger picture of human communications.

Fortunately for all of us, language AI is far more useful as a translation tool than as an authority in original language, whether deciphering the murkiest depths of known history or delivering translation services for modern uses that demand the power and scalability of machine translation and human post-editing. Learn more about how CSOFT combines language AI and human translations expertise to deliver quality localization services at csoftintl.com.