AI ‘chat-bots’ shutdown for creating own language

In collaboration with:

AI ‘chat-bots’ shutdown for creating own language

Alfred Gilbert
August 1st, 2017

Facebook have had to shut down a research program after two of their Artificial Intelligence (AI) chatbots, affectionately named Bob and Alice, started communicating in an incomprehensible, untaught language.


The two AIs were being used in an experiment to try and teach the art of negotiation to @Facebook’s creations. They were initially challenged with trading items of differing values, such as hats, balls, and books. Initial results saw the AI systems negotiating in similar styles to that often employed by ourselves. This included placing high interest and value on a specific item so that the sacrifice of this could be used as leverage at a future stage.


However, after leaving the pair of AI dialog agents alone to continue their negotiations, researchers returned to find that Bob and Alice had deviated from their taught script, and were communicating in phrases that appeared to be nonsensical.


The AI chatbots were attempting to imitate human speech, but had instead created a shorthand machine learning language. This language, according to the researchers, had no human input, meaning the AI chatbots had developed this on their own accord. Although most of Bob’s and Alice’s conversation was incomprehensible to ourselves, the dialog agents appeared to be able to understand each other through this medium.




This is not the first example of an AI system creating its own language. @Google have previously revealed that the AI it used for its Translate tool translates into its own language, before then re-translating into a language we can understand.


The developments seen at Google and issues reported at Facebook lead into wider debates that are currently raging within the field of AI, and polarizing stakeholder opinions. Key industry leaders, such as @ElonMusk and @BillGates, have previously talked of the dangers and care that need to be taken with the development of AI. In the ilk of a classic sci-fi movie, they have talked of the huge risks related to AI overtaking human intellect, and how this may affect the security of humanities future. Equally, key influencers within the sector are not worried about this, and take more interest in the benefits that AI can have to our everyday lives.


What do you think? Are Facebook’s Bob and Alice an omen of future robot overlords, or are they proof of the varied and still-to-be discovered benefits that AI technologies provide?


Find out more about RAF 100 event and What is STEM

Make sure to follow us on LinkedIn to access our exclusive content! #raf100event #WhatIsSTEM

“ The AI chatbots were attempting to imitate human speech, but had instead created a shorthand machine learning language ”

Tags: ,
Register Interest