Several days ago, media started reporting on an AI project at Facebook which went “rogue” (eg, Independent, Sun). References were made to malicious AI systems in sci-fi movies, like Skynet in the Terminator movies (1984-2015, IMDb) and War Operation Plan Response (WOPR) in WarGames (1983, IMDb). As always, reality was less juicy but still fascinating (eg, Gizmodo).
Apparently, Facebook had been developing trading negotiation skills in AI chatbots in order to deal with human customers. Instead, these chatbots started negotiating amongst themselves and were altering the English language in that process. Project management realised their mistake, ended the conversation, and “opted to require [the chatbots] to write to each other legibly“.
The notion that our languages are efficient, effective and follow the SMART criteria is intriguing. Several words have completely different meanings although the underlying action shows some similarity (eg, bank). Some of these so-called homographs (examples) only have a different pronunciation. The existence of homographs must be a “headache” for AI.
The English language may be considered a common language but its proficiency varies hugely. On 21 March 2017, the Civil Aviation Authority issued the “Aviation English Research Project: An independent study” (PDF). Independent: “An air disaster could be caused by low levels of English language ability among some pilots and air traffic controllers, [the] report has warned.”
It’s tempting to think that these “low levels of English language ability” is a foreign problem. Wrong. BBC: “The non-native speakers, it turns out, speak more purposefully and carefully, typical of someone speaking a second or third language. Anglophones, on the other hand, often talk too fast for others to follow, and use jokes, slang and references specific to their own culture”.
Following the above, it makes perfect sense that AI robots would want to develop their own language. Independent: “Robots have learned how to communicate with each other by creating their own language, a new [OpenAI] report explains. [] As the language continues to develop and grow more and more complex, the researchers hope to build a translator bot capable of translating their communications for humans.” Note: bold, italics and underlining by LO.
In my August 1 blog on “Fear sells”, I mentioned that artificial fears relate to differences in disability, gender, nationality, race, religion, and sexuality. With the knowledge of hindsight, I should have added language. A unique artificial language, incapable of understanding by humans, would create artificial human fears. The AI related articles of 20 June 2017 and 1 August 2017 in tabloid newspaper The Sun indeed address this human fear.
Albert Einstein‘s radio address to the 1941 London Science Conference, was called “The Common Language of Science” (video). Einstein says: “the mental development of the individual and his way of forming concepts depend to a high degree upon language.” Open Culture: “Thus a shared language implies a shared mentality”. Hence, a unique AI language, incapable of understanding by humans, must be a dead-end street.
In Common (2016) by Alicia Keys – artist, lyrics, video, Wiki-1, Wiki-2
0 Comments