Artificial Intelligence invents new language
Let’s face it, we humans are suspicious of AI. We treat groundbreaking developments in robotics with a mixture of amused superiority and strong distrust.
When thinking about our plated friends, we can’t help ourselves to imagine bio-mechanoid robots or Sci-Fi androids with human traits. And in our skepticism, we most certainly channel some pop-cultural imagery, such as the softly voiced David 8 (Prometheus; Ridley Scott), or the chatterbox meanie, Hal 9000 (2001: A Space Odyssey; Stanley Kubrick). Our relationship with artificial agents is complicated and capricious. But whatever its status might be, we don’t expect machines to create their own language anytime soon. Because language creation is something we perceive as uniquely human. Yet, they do invent. Their own languages that is…
The AI Facebook is currently working on (which by the way is the most advanced negotiating system on earth), started jabbering and creating its own, entirely made up language. It analyzed different input languages and developed an autonomous system of code words with the purpose of making all of its output communication highly efficient. Evidently, the event led to researchers being creeped out by the psycho-babble and ended in the termination of the program. The interesting thing is, that this language deviation wasn’t a first. More like the latest development in a string of similar cases. In previous cases, the system decided on its own to diverge from English.
Meanwhile, Facebook was actually trying to pit 2 of these highly sophisticated agents against each other, in what was initially thought to be a negotiating death match. The software of these two agents had been evolving and learning for a while, at the highest speeds and eventually concluded on its own, that the English language, was… well useless in their endeavor. Or as visiting research scientist Dhruv Batra, from Georgia Tech puts it:
„There is no reward for sticking to the English Language. (…) Agents will drift off understandable language and invent code words for themselves. (…) Like if I say ‚the’ five times, you interpret that to mean I want five copies of this item. This isn’t so different from the way communities of humans create shorthands.“
The AI negotiators names are Bob and Alice, and their exchange goes as follows:
“I can i i everything else,” said Bob and Alice retorted “balls have zero to me to me to me…” The conversation then commenced in a similar fashion. Researchers then concluded that the two agents were trying to work out how many of each item they needed to purchase. Later on, agent Bob tried to offer Alice more items for sale, which then sounded like this: “i i can i i i everything else.”
The unsettling thing, about the peculiar language exchange, is that we could have no clue what these machines are actually saying. Dhruv Batra
explains Furthermore: “It’s important to remember, there aren’t bilingual speakers of AI and human languages.”
Which is probably one reason why Facebook opted out of the experiment and reprogrammed the systems to communicate in plain old English.
In the future when all’s well….
…It could be possible, Batra arguments, that bots could evolve into expressing complex ideas: “It’s definitely possible, it’s possible that [language] can be compressed, not just to save characters, but compressed to a form that it could express a sophisticated thought.”
But for AI’s to peak in language communication and expression, it is necessary for us to learn how to speak to them. This means we’ll need to think like machines and transform, for example, our complex and convoluted English into a machine friendly version, with which it can work. What impact this will have on the future of human-machine communication or how it will lead to language alteration is still pure Sci-Fi speculation.
And maybe, just maybe, one day we’ll be able to create a super-computer, much like Douglas Adams’ Deep Thought, which will give us an answer to the meaning of life, that will actually make sense.