Much less. There is reason to believe that these tools could lead to mass extinction of languages. And, more worryingly, it will obliterate multiple ways of thinking and creating.
How this happens holds clues on how to stop it.
First, some basics: A large-scale language model (LLM) is trained to make best predictions about what will be produced next in a conversation or sentence. They found that the word after “with bacon” was most likely to be “egg”, and that “egg” was more likely to be correct if it included other words preceded by “breakfast” or “coffee”. learn that is high.
However, we are already well beyond bacon and eggs and much closer to Joël Robuchon’s menu, which combines all the dishes offered by his restaurants, which have a total of 32 Michelin stars. And with each new iteration of LLM, we are surpassing the last iteration faster and faster.
AI language models are trained on vast amounts of data from available books, magazines, newspapers, and online content. The more data, the better. However, what is available for training models varies greatly among the thousands of languages in use today.
The most powerful models will be those trained in about 20 “high resource languages” such as English, Mandarin Chinese, Russian, German, and Japanese. AI then generates a large amount of new text, mostly in those languages. Similar to invasive species, such dominant models can drive out languages with fewer resources for training.
Some of these languages (Hawaiian, Quechua, Potawatomi, etc.) are already endangered due to globalization, immigration and cultural homogenization. Currently, about nine fires are extinguished each year. LLM could dramatically increase its extinction rate.
Importantly, this is more than language. The extinction of most languages within a few generations also brings with it the collapse of ideas and existence. Because the interaction between language and mind is two-way.
Language forms the brain. It’s one of the most powerful ways to organize, process, and structure information. The language we use affects how we perceive the world, what we remember, the decisions we make, the emotions we feel and the insights we have.
Experiments in my lab at Northwestern University and elsewhere have shown that people who speak different languages have different eye movements and different brain activity. Various things in the environment catch their attention. Their memories and interpretations of the world and reality vary. People who speak many languages activate slightly different neural networks.
The reality that each of us perceives is a subjective experience. It is due to how our brain combines input from our senses with knowledge and experience. Language experience gives us a prism through which we can see the universe.
The end of language sorts out the number of prisms that refract the world.
This is ironic. During my sabbatical at Stanford University, most of the Silicon Valley AI scientists I met spoke, learned, grew up with, or were exposed to multiple languages. Many said he spoke more than one language. The very people who may contribute to the demise of the multilingual mind are those who are harnessing that power to build these extraordinary artificial intelligence software programs.
Would our thinking change if our reality were filtered by a much more limited set of languages, largely shaped by symbolic systems of mathematics, logic and artificial language?
One insight comes from a version of the classic dilemma used to study morality. A trolley is speeding towards five invisible workers. You are standing on a footbridge over a railroad track next to a person. If you push into the track below, you will die. But they also stop the trolley and save five workers.
When responding in their native language, 20% choose to sacrifice one person to help five. 33% do when responding in a second language. This 13% shift to utilitarian decision-making is called the foreign language effect.
But as new brain imaging and computational science begin to provide insight into why the multilingual mind works the way it does, we wonder if one of its central functions could be ruled out. I am setting a course.
Two things. First, AI research, development, and use should be regulated in the public interest. At the very least, this should be on par with other private sector industries. Even better is a level similar to that of the defense division.
Second, it is imperative to keep as much natural language actively engaging the human mind as possible in order to buy time to come up with more solutions.
Now is not the time to reduce the supply of new ideas. Our many languages are one of the most powerful sources of diversity of thought. In human experience, multilingualism is a signal, not noise.
