Olaf Lipinski: In the 2016 science fiction film Arrival, a linguist is faced with the daunting task of deciphering an alien language consisting of palindromic sentences, which read the same backwards as forwards, written with circular symbols. As she discovers different clues, different countries around the world interpret the messages differently – with some assuming them to pose a threat.
If humanity were to find itself in such a situation today, the best thing to do would be to turn to research that reveals how artificial intelligence (AI) develops languages.
But what exactly defines a language? Most of us use at least one to communicate with those around us, but how did this come about? Linguists have been studying this question for decades, but there’s no easy way to figure out how language evolved.
Language is ephemeral and leaves no researchable trace in the fossil record. Unlike bones, we can’t dig up ancient languages to study how they developed over time.
While we may not be able to study the true evolution of human language, a simulation may provide some insights. That’s where AI comes in – a fascinating area of research called emerging communications that I’ve been studying for the past three years.
To simulate how language might evolve, we give agents (AIs) simple tasks that require communication, such as a game where one robot must guide another to a specific location on a grid without showing it a map. We offer (almost) no restrictions on what they can say or how – we just give them the task and let them solve it however they want.
Because solving these tasks requires the agents to communicate with each other, we can study how their communication evolves over time to get an idea of how language might evolve.
Similar experiments have been done with humans. Imagine that you, an English speaker, are paired with a non-English speaker. Your job is to instruct your partner to pick up a green block from an assortment of objects on a table.
You can try pointing to a cube shape with your hands and pointing to the grass outside the window to indicate the color green. Over time, the two of you would develop a kind of protolanguage together.
Perhaps you can create specific gestures or symbols for “cube” and “green.” Through repeated interactions, these improvised signals would become more refined and consistent, forming a basic communication system.
This works the same way for AI. Through trial and error, they learn to communicate about the objects they see, and their interlocutors learn to understand them.
But how do we know what they are talking about? If they develop this language only with their artificial interlocutor and not with us, how do we know what each word means? After all, a specific word can mean ‘green’, ‘cube’ or worse, both. This interpretation challenge is an important part of my research.
Cracking the code
The task of understanding AI language may seem almost impossible at first. If I tried to speak Polish (my native language) to an employee who only spoke English, we wouldn’t be able to understand each other or even know where each word begins and ends.
The challenge with AI languages is even greater, because they can organize information in ways that are completely alien to human language patterns.
Fortunately, linguists have developed sophisticated tools that use information theory to interpret unfamiliar languages.
Just as archaeologists piece together ancient languages from fragments, we use patterns in AI conversations to understand their linguistic structure. Sometimes we find surprising similarities to human languages, and sometimes we discover entirely new ways of communicating.
These tools help us peer into the ‘black box’ of AI communications and reveal how artificial agents develop their own unique ways of sharing information.
My recent work focuses on using what officers see and say to interpret their language. Imagine you have a transcript of a conversation in a language unknown to you, along with what each speaker was looking at. We can match patterns in the transcript with objects in the participant’s field of view, creating statistical connections between words and objects.
For example, perhaps the phrase “yayo” coincides with a bird flying by – we might guess that “yayo” is the speaker’s word for “bird.” Through careful analysis of these patterns we can begin to decode the meaning behind the communication.
In the latest paper by me and my colleagues, which will appear in the Neural Information Processing Systems (NeurIPS) conference proceedings, we show that such methods can be used to reverse engineer at least parts of the language and syntax of AIs, giving us insight in how she can structure communication.
Aliens and autonomous systems
How does this relate to aliens? The methods we develop to understand AI languages could help us decipher future alien communications.
If we are able to obtain a written alien text along with some context (such as visual information related to the text), we could apply the same statistical tools to analyze it. The approaches we develop today could be useful tools in the future study of alien languages, known as xenolinguistics.
But we don’t need to find aliens to benefit from this research. There are countless applications, from improving language models such as ChatGPT or Claude to improving communication between autonomous vehicles or drones.
By decoding emerging languages, we can make future technology more understandable. Whether it’s knowing how self-driving cars coordinate their movements or how AI systems make decisions, we’re not just creating intelligent systems, we’re also learning to understand them.
Olaf Lipinski, PhD student in Artificial Intelligence, University of Southampton
This article is republished from The conversation under a Creative Commons license. Read the original article.