Brain vs AI

Have you ever had the feeling that people are talking behind your back, but you can’t figure out what they are actually communicating? Now just imagine for a minute if instead of people, there were Artificial Intelligence (AI) bots maneuvering and scheming in the background and to further fan the flames they have been communicating in their own code language.

That’s exactly the quandary that Facebook faced when it found that its AI bots were conversing in a shorthand mutation of English.

Here’s everything you need to know.

Ground-breaking discovery

On Sunday, the Facebook Artificial Intelligence Research lab (FAIR) made a stunning discovery while trying to improve chat-bots.

The bots, known as dialog agents, were left to converse freely in an attempt to strengthen their conversational strategies and skills incrementally using machine learning algorithms.

Eventually, these bots turned into incredible crafty negotiators. The bots became too proficient at it. They even stated bluffing interest in one item in order to ‘sacrifice’ it at a later stage in the negotiation as a faux compromise. These findings are a huge leap forward for AI.

Over time, these bots began to deviate from their English scripted norms and started communicating in an entirely new non-human language that they created without human input. Due to which, Facebook researchers decided to tweak their model.

Glimpse of the future language

Researchers from the Facebook Artificial Intelligence Research lab designed two AI agents named Bob and Alice to negotiate with humans.

Here is a passage from one of their conversations:

Bob: i can i i everything else . . . . . . . . . . . . . .
Alice: balls have zero to me to me to me to me to me to me to me to me to
Bob: you i everything else . . . . . . . . . . . . . .
Alice: balls have a ball to me to me to me to me to me to me to me
Bob: i i can i i i everything else . . . . . . . . . . . . . .
Alice: balls have a ball to me to me to me to me to me to me to me

Bob: i . . . . . . . . . . . . . . . . . . .
Alice: balls have zero to me to me to me to me to me to me to me to me to
Bob: you i i i i i everything else . . . . . . . . . . . . . .
Alice: balls have 0 to me to me to me to me to me to me to me to me to
Bob: you i i i everything else . . . . . . . . . . . . . .
Alice: balls have zero to me to me to me to me to me to me to me to me to

New Code Language

Looking at face value, the verbiage doesn’t seem to make sense. In language sense, these phrases seem gibberish. But they contain semantic meaning when interpreted by Facebook AI agents. This enabled them to communicate with one another without letting the researchers to know the content of the shared information.

In the above dialect, the agents are shrinking sentences down to their bare elements as a form of shorthand. The entire dialect was formed from variations of the first two sentences.

The repetition of phrases like “i” and “to me” reflect how the AI operates. They reflect that the two bots are working out how many of each item they should take. Bob statements reveal how it was using language to offer more items to Alice. On interpretation, this phrase meant “I’ll have three and you have everything else.”

English lacks a ‘reward’

Modern AIs operate on a ‘reward’ principle. At the end of every dialog, the agent is given a reward on the basis of the agreed deal. That reward was then back-transmitted through every word in the bot output so it could learn which actions lead to high rewards.

Ingenious Express analysis found that, in this instance, there was no reward for continuing to use English. Since it didn’t contribute to their end goal of becoming more efficient negotiators. Instead, these AI system created a more effective, albeit, unintelligible robot language.

In fact, the systems had multiple incentives to veer away from the language, the same way human communities with niche knowledge create and use shorthand to discuss complex ideas more quickly or efficiently.

Not the first time

AI developers at other companies have scrutinized a similar use of shorthand to facilitate communication. At the Elon Musk’s artificial intelligence lab, OpenAI, scientists succeeded in allowing AI bots to learn their own languages.

In a separate case, Google improved its Translate service by including a neural network. The AI system is now capable of translating much more efficiently, including between language pairs. But, they weren’t explicitly taught about this.

Ingenious Express analysis found that the AI had silently written its own language that is tailored specifically to the task of translating sentences. In fact, they even represent the most efficient known solution to the problem.

Future of AI

There is no proof to claim that these sudden AI divergences are a threat that is leading machines to take over operators.

This groundbreaking discovery, unfortunately offers very little real-world value. As we, mere humans, are incapable of grasping the overwhelmingly logical nature of the new languages.

Even this incident comes just a few days after Mark Zuckerberg sternly criticized AI naysayers for fabricating doomsday scenarios.

Perhaps the most vocal alerts about AI advancing had come from Elon Musk and Stephen Hawking. One of the most salient points in their arguments is that by the time we perceive a risk, it may be too late. Let’s hope to remain the more intelligent race.


Please enter your comment!
Please enter your name here

This site uses Akismet to reduce spam. Learn how your comment data is processed.