#machine hallucinations

Machine Hallucinations: Coral - Temporary Exhibitions / Exhibition Design

By Refik Anadol Studio, United States

📢 Türk-Amerikan medya sanatçısı ve yapay zekâ öncüsü Refik Anadol, TIME tarafından verilen AI Impact Award’a layık görüldü! Refik Anadol, TIME 100 Listesinde! Anadol, veriyi sanata dönüştüren öncü çalışmaları ve yapay zekâ ile sanat arasındaki bağı güçlendiren projeleriyle ödüle değer görüldü. Ödül töreninde yaptığı konuşmada, “Veriyi yaşayan, nefes alan bir malzeme olarak görüyorum. Sanat, belli…

Type the word “dog” into Google Translate 19 times, request that the nonsensical message be flipped from Maori into English, and out pops what appears to be a garbled religious prophecy.

“Doomsday Clock is three minutes at twelve,” it reads. “We are experiencing characters and a dramatic developments in the world, which indicate that we are increasingly approaching the end times and Jesus' return.”

That’s just one of many bizarre and sometimes ominous translations that users on Reddit and elsewhere have dredged up from Google Translate, Google’s decade-old service that can now interpret messages in over 100 languages. In Somali, for instance, strings of the word “ag” translate into missives about the “sons of Gershon,” the “name of the LORD,” and references to Biblical terminology like “cubits” and Deuteronomy.

There are several possible explanations for the strange outputs. It’s possible that the sinister messages are the result of disgruntled Google employees, for instance, or that mischievous users are abusing the “Suggest an edit” button, which accepts suggestions for better translations of a given text.

Andrew Rush, an assistant professor at Harvard who studies natural language processing and computer translation, said that internal quality filters would probably catch that type of manipulation, however. It’s more likely, Rush said, that the strange translations are related to a change Google Translate made several years ago, when it started using a technique known as “neural machine translation.”

In neural machine translation, the system is trained with large numbers of texts in one language and corresponding translations in another, to create a model for moving between the two. But when it’s fed nonsense inputs, Rush said, the system can “hallucinate” bizarre outputs—not unlike the way Google’s DeepDream identifies and accentuates patterns in images.

“The models are black-boxes, that are learned from as many training instances that you can find,” Rush said. “The vast majority of these will look like human language, and when you give it a new one it is trained to produce something, at all costs, that also looks like human language. However if you give it something very different, the best translation will be something still fluent, but not at all connected to the input.”

Sean Colbath, a senior scientist at BBN Technologies who works on machine translation, agreed that strange outputs are probably due to Google Translate’s algorithm looking for order in chaos. He also pointed out that the languages that generate the strangest results—Somali, Hawaiian and Maori—have smaller bodies of translated text than more widely spoken languages like English or Chinese. As a result, he said, it’s possible that Google used religious texts like the Bible, which has been translated into many languages, to train its model in those languages, resulting in the religious content.