Olaf Lipinski: In the 2016 science fiction film “Arrival”, a linguist is faced with the tough task of deciphering an alien language consisting of palindromic phrases that read the same backwards and forwards, written in circular symbols. As he discovers different clues, different nations around the world interpret the messages differently – some assume they carry a threat.
If humanity found itself in such a situation today, the best solution would be to turn to research exploring how artificial intelligence (AI) develops languages.
But what exactly defines language? Most of us operate at least one to communicate with the people around us, but how did this happen? Linguists have been pondering this question for decades, but there is no effortless way to know how language evolved.
The language is ephemeral, leaving no researchable traces in the fossil record. Unlike bones, we cannot dig up old languages to study how they developed over time.
While we may not be able to study the true evolution of human language, perhaps a simulation could provide some insight. This is where artificial intelligence comes to the rescue – a fascinating field of research called emergent communication, which I have spent the last three years researching.
To simulate the evolution of language, we give agents (AI) elementary tasks that require communication, for example a game in which one robot must guide another to a specific location on a grid without showing it a map. We put (almost) no limits on what they can say and how they can say it – we just give them a task and let them solve it the way they want.
Since solving these tasks requires agents to communicate with each other, we can study the evolution of their communication over time to learn how language might evolve.
Similar experiments were performed on humans. Imagine you are an English speaker and a non-English speaker. Your task is to instruct your partner to pick up a green cube from the assortment of items on the table.
You can try making a cube gesture and pointing to the grass outside the window to indicate the color green. Over time, you developed something of a proto-language together.
Maybe you could create special gestures or symbols for “dice” and “green”. Through repeated interactions, these improvised signals will become more refined and consistent, creating a basic communication system.
It works similarly with artificial intelligence. 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're talking about? If they only develop this language with their artificial conversation partner and not with us, how do we know what each word means? After all, a particular word could mean “green,” “cube,” or worse, both. This interpretive challenge is a key part of my research.
Cracking the code
The task of understanding the language of AI may seem almost impossible at first. If I tried to speak Polish (my native language) with a co-worker who only speaks English, we wouldn't be able to understand each other or even know where each word begins and ends.
The challenge of AI languages is even greater because they can organize information in ways that are completely alien to human linguistic patterns.
Fortunately, linguists have developed advanced tools that operate information theory to interpret unknown languages.
Just as archaeologists piece together old languages from fragments, we operate patterns in AI conversations to understand their linguistic structure. Sometimes we find surprising similarities to human languages, and other times we discover completely up-to-date ways of communicating.
These tools aid us look inside 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 agents see and say to interpret their language. Imagine having a transcript of a conversation in a language you don't know, along with what all the speakers were looking at. We can match patterns in the transcript to objects in the participant's field of view, building statistical associations between words and objects.
For example, perhaps the phrase “yayo” coincides with a bird flying by – we can guess that “yayo” is the speaker's word for “bird.” By carefully analyzing these patterns, we can begin to decode the meaning of the message.
In latest article by myself and my colleagues to appear in the Neural Information Processing Systems (NeurIPS) conference proceedings, we show that such methods can be used to reverse engineer at least part of the language and syntax of artificial intelligence, giving us insight into how it can structure communication .
Aliens and autonomous systems
How does this connect with aliens? The methods we develop to understand AI languages could aid us decipher any future alien communications.
If we can obtain some foreign written text along with some context (e.g. visual information relating to the text), we could apply the same statistical tools to analyze it. The approaches we are developing today may prove to be useful tools in the future study of foreign languages, known as xenolinguistics.
But we don't need to find extraterrestrials to benefit from this research. The applications are many, from improving language models such as ChatGPT or Claude to improving communication between autonomous vehicles or drones.
By decoding up-to-date languages, we can make future technology easier to understand. Whether it's knowing how autonomous cars coordinate their movements or how AI systems make decisions, we don't just create bright systems – we learn to understand them.
Olaf Lipiński, PhD student in artificial intelligence, University of Southampton
This article has been republished from Conversation under Creative Commons license. Read it original article.
Image Source: Pixabay.com