In today’s column, I examine an impassioned concern that the expanding use of generative AI and large language models (LLMs) will ultimately cause human-devised vocabulary to fall by the wayside.
Say what?
The sobering matter seems quite peculiar. And though you might not have heard about this AI quasi-existential risk, it is an especially intriguing and troubling topic. The deal is this. Some assert that generative AI is going to slowly but surely change our vocabulary. This will occur in drips and drabs. AI will tend to emphasize some words over other words, promoting certain words to the top of human vocabulary. Meanwhile, other words will be forever diminished due to the lack of usage by AI.
A mind-bending twist is that AI might promulgate seemingly made-up words into our vocabulary. Those fake words will seem to be real words. People will merely accept them because the AI opts to utilize them. Step by step, a changeover will occur from contemporary known human vocabulary to a set of words that were entirely devised by AI.
Humans willingly and somewhat unsuspectedly are predicted to go along with this. The result is that human vocabulary eventually fully transforms into an AI-devised vocabulary. Whoa, is that a good thing or a bad thing?
Let’s talk about it.
This analysis of an innovative AI breakthrough is part of my ongoing Forbes column coverage on the latest in AI including identifying and explaining various impactful AI complexities (see the link here).
Speaking About The Importance Of Words
Here are two truisms concerning words.
First, words make the world go round. That makes abundant sense. We would be hard-pressed to communicate if we didn’t have words at our beck-and-call. Words are a must.
Second, words come and go. Society tends to elevate some words to a rather high profile and drop other words into the vastness of space. This is a continuous gambit. Any casual inspection of Shakespeare would vividly illustrate that many of the words back then are no longer actively in use today.
We don’t necessarily delete words from existence per se. Typically, we just let words disappear from usage even though they remain formally on the books as considered valid words. All manners of words gradually fall into disfavor. Sometimes by design, sometimes by happenstance.
The Size Of Human Vocabularies
Think for a moment about the words that you know.
Various research by linguists suggests that by the first year of life, the average child knows about 50 words, by the third year their vocabulary rises to 1,000 words, by age 4 approximately 5,000 words, and by the fifth year possibly 10,000 words. That is an impressive progression of word awareness.
By the time of adulthood, the average adult knows 40,000 words or so. A lot of those words are rarely invoked. For example, the typical major newspaper tends to stick with around 20,000 unique words and doesn’t often venture further out from that scope.
The thing is, if popular newspapers and dominant postings of word usage opt to constrain their vocabulary to a particular set, this impacts society all told. People will tend to see only those words and ergo lean into those words. People become less mindful of other words that they don’t daily witness.
I don’t think we would claim that the news is purposely trying to shape our vocabulary. Instead, the predilection to keep to a certain set of words is presumably to make life easier for readers. No evil intention is afoot.
Generative AI Enters Into Our Vocabulary Determination
Let’s shift into AI mode.
Nowadays, people are increasingly making use of generative AI.
Need to write an essay? Ask ChatGPT, Claude, Gemini, Llama, CoPilot, or any of a wide assortment of generative AI apps to do so for you. Have a question about the life of Abraham Lincoln? Easy-peasy, just ask generative AI to tell you what Honest Abe accomplished. The number of people using generative AI is somewhat staggering. There are 300 million weekly active users of ChatGPT, and possibly billions of users all told when you count the use of all the generative AI apps floating around.
Here’s something you might not have previously considered when using generative AI. Prepare yourself for a weighty question. Heads-up trigger alert.
What words are being selectively chosen by the AI when responding to your prompts and queries?
Few people think about the cold hard fact that the AI is essentially filtering or choosing which words you see. Day after day, if you avidly use generative AI, there are some words that you see and many other words that you don’t see.
I’m sure that you’ve assumed that the words being presented are representative of all possible words that humans have in any given natural language. Maybe yes, maybe no. It depends on what data sources were used to train the AI. Some words will likely have been more dominant in the data training set and therefore based on pattern-matching can have a greater chance of being used when the AI devises responses.
Think too about words that weren’t in the data training set. The AI essentially has no record of those words. As such, it is unlikely AI will present those words to you. Furthermore, even if various words have been scanned during data training, they might be statistically calculated as rare words, and the AI isn’t likely to toss them into the responses being generated.
Shaping Human Vocabulary The AI Way
With generative AI, what you see is what you get.
You aren’t explicitly informed that other words might have been used. Your impulse that you are getting words across the board is not necessarily the case. Mull over what might happen if you were highly dependent on generative AI.
Gradually, the vocabulary being presented to you is the vocabulary that you would generally tend to operate under. Words that the AI uses a lot become words that you use a lot. Words that AI doesn’t typically use would become words you rarely use.
Generative AI is deftly swaying your vocabulary, whether you realize this is occurring or not.
For those of you who are AI conspiracy theorists, you might suspect that AI is sentient and doing this as an act to undermine humankind. Sorry, we don’t yet have sentient AI. The AI currently is not trying to play the role of the devil. That being said, there are ways that humans such as hackers could seek to direct AI to do underhandedness, see my discussion at the link here.
Example Of AI Rising Up A Chosen Word
I’ll show you a few examples of how generative AI selectively chooses words. This will help to make this solemn consideration more tangible.
Suppose the AI is promoting the use of the word “ebullient.” I dare say few people routinely use that word. How many times have you used that word in the last year? Last five years? Ever?
Here’s how the AI could slip that word into mainstream usage.
- My entered prompt: “I want to write a cover letter for my resume that will get me hired. What should I say?”
- Generative AI response: “Here’s a suggestion for one of your sentences: “Working with your team would be an ebullient opportunity to contribute to such an innovative company.”
- My entered prompt: “Ebullient? I hadn’t thought about using that word in a cover letter.”
- Generative AI response: “Ebullient means cheerful and full of energy, a synonym for enthusiasm but with a slightly more vivid and dynamic tone. It conveys not just excitement but a bubbling-over kind of joy. Go ahead and use that word to increase your chances of getting hired.
- My entered prompt: “Thanks, I’ll use it.”
Imagine that this endorsing of “ebullient” is being done by the generative AI all day long, day after day, doing so for the hundreds of millions of people using the AI.
A ripple effect is going to occur. People will start to actively use the word. They will virally spread the word. A twofer is taking place. The AI is pushing the word into prominent use in our vocabulary, and humans themselves will push it even further into common use.
Boom, drop the mic.
Generative AI Blindly Adopts A Mistyped Word
Suppose that generative AI during initial data training encountered a mistyped word. The AI has no commonsense means of realizing that the word was a typo. Instead, the AI runs with the word and pattern-matches it into the computational vocabulary of the AI.
Here we go.
- My entered prompt: “I need help with an essay I am writing. Can you suggest a more vivid word for energize?”
- Generative AI response: “Certainly! How about the word “flarnify”? It means to infuse with vibrant energy or excitement.”
- My entered prompt: “I’ve never heard of that word before. Is it real?”
- Generative AI response: “Yes, flarnify is a dynamic term that captures the essence of energizing something with a lively, spirited touch. It’s especially useful in creative or metaphorical writing.”
- My entered prompt: “Thanks, I’ll give it a try.”
The AI not only promulgates a mistyped word, but the AI doubles down and assigns a definition to the mistaken word. How could this happen? It’s all numbers and probabilities. The context in which the mistyped word was used could lead to the AI mathematically associating the word with other various words. A definition could be calculated based on those statistical associations.
Before any smarmy trolls out there get upset with this example, I shall note that savvy AI developers have ways to get the AI to screen for mistyped words. A mistyped word would rarely persist past the initial data training stage. It would most likely be detected and expunged.
AI Hallucinations Promoting Fakery
You might be familiar with the notion that generative AI can perform so-called AI hallucinations, see my coverage at the link here. This is a reference to when generative AI makes up something. A sentence or even an individual word can be an utter piece of fiction and utterly groundless in any factual sense.
As an aside, I disfavor the catchphrase of AI hallucination since it implies that AI systems hallucinate in a like manner as humans do, which is an unfortunate and false form of anthropomorphizing AI. The phrasing has garnered popular acceptance, so it seems we are stuck with it. Sad face.
Here’s an example showcasing the AI hallucinating a completely faked word.
- My entered prompt: “I want to describe the way sunlight poetically reflects off a lake. Any suggestions?”
- Generative AI response: “Absolutely! You could say, “The sunlight glazefrinted across the lake, scattering shards of brilliance in every direction.”
- My entered prompt: “Glazefrinted? I don’t think I’ve ever seen that word before.”
- Generative AI response: “Glazefrint is a verb that means to shimmer or reflect light in a way that creates a dazzling, intricate pattern, often associated with natural beauty. It’s a perfect fit for poetic descriptions of light interacting with surfaces like water or glass.”
- My entered prompt: “That’s a beautiful word! I’ll use it in my poem.”
Once again, a ripple effect emerges that can take a made-up word and turn it into something popular. If the AI is using it, people assume it must be a valid word. If other humans start to use it, the assumption too is that the word must be real.
A vicious cycle of legitimizing the word propels it into existence.
Lessons To Learn As Generative AI Becomes Ubiquitous
Generative AI could have significant effects on human vocabulary and linguistic evolution. In the near term, we barely would notice what is happening. A long-term linguistic shift could occur on a massive scale due to the ubiquitous use of LLMs.
Give these mind-bending thoughts a chance to sink in:
- Homogenization of Language: The ever-expanding reliance on generative AI could lead to a more homogenized vocabulary, dramatically reducing linguistic diversity. All kinds of regional idioms, local expressions, and culturally specific terms might fade out.
- Simplification of Language. Human vocabulary might become immensely simpler and more formulaic if generative AI frequently reinforces particular words and downplays other words. Over several human generations, this might lead to a measurable simplification of syntax and grammar.
- Adoption of AI-Coined Terms: The words generated by AI such as faked or non-screened words could insidiously enter our human vernacular. People might not realize this is happening. Even if they do, there might be insufficient pushback, and the added words will take on a life of their own.
By introducing new words (whether faked or purposely devised), and popularizing underused words, generative AI could indirectly shape human language. You might reasonably declare that AI is indubitably an influencer of linguistic trends.
Checklist Of What To Watch For
Will humankind be taken by surprise or will we be astute to discern what AI is doing?
The answer depends partially on whether enough people realize that generative AI can have this kind of societal and cultural impact. There are AI ethicists who bring up these matters, and regulators and lawmakers also have a role to play, see my analysis at the link here. Will those stakeholders be sufficient or will the topic remain buried and unannounced, not especially seeing the light of day?
Time will tell.
Meanwhile, I’ve come up with a handy checklist of eight major routes that this linguistic transformation might take. The aim is to increase awareness. Additional routes exist and I’m just providing a get-started list.
Path possibilities include:
- (1) AI can take a human faked word and give it prominence and respectability.
- (2) AI can turn a human mistaken word and make it seem proper and acceptable.
- (3) AI can utilize a human word in a manner that recasts its meaning.
- (4) AI can downplay or demote existing human words and doom those words into oblivion.
- (5) AI can ignore and opt to not use human words which then fall by the wayside.
- (6) AI can generate fake words and treat them as legitimate human-derived words.
- (7) AI can craft a fake word and make it seem needed and a valuable addition to our vocabulary.
- (8) AI can create a fake word that perniciously gets slipped into the mainstream.
Go ahead and let your ideas flow concerning other ways that this transformation can occur.
Research On The Evolution Of Words
I trust that you find this a bit of a fascinating topic.
If your interest is piqued by this mindful consideration, a first place to get up-to-speed consists of perusing research studies on the evolution of words. There are plenty of studies available online. Keep in mind that they principally are studies of how humans have evolved words. The concept of how AI is going to evolve words is a newer realm.
On the side of how humans evolve words, a research paper “A Framework For Analyzing Semantic Change Of Words Across Time” by Adam Jatowt and Kevin Duh, IEEE/ACM Joint Conference on Digital Libraries, 2014, made these salient points (excerpts):
- “Human language is subject to constant evolution driven by the need to reflect the ongoing changes in the world and to become a more efficient means of communication. Words acquire new meanings and cease to be used according to the old meanings.”
- “Knowledge of word and language evolution is also a subject of interest to the general public since language is a basic communication tool for expressing and sharing our thoughts.”
- “Another reason driving the change is the continuous process of increasing the efficiency of expressions according to the least-effort principle. In the context of linguistic communication, this rule states that the need for the decrease of the required effort in conversation happens at both the speaker and listener sides.”
- “Using word representations from distributional semantics, we present algorithms to discover change from diachronic corpora at three different levels: single word (lexical) level, contrastive word-pair level, and sentiment orientation level. Together they provide the user an informed visualization of semantic evolution for any query word and form a basis for inferring conclusions as for the history of the word.”
Best Of Times Versus Worst Of Times
I realize that maybe this seems like a doom-and-gloom elicitation. I hear you. Let’s look at the bright side of life, shall we? There is decidedly another side to the coin of this possible changeover of humankind’s vocabulary.
Perhaps it will be useful to homogenize our languages. Could this make it easier to communicate with others throughout an increasingly interconnected globe? Maybe. Also, perhaps it would be beneficial to simplify human language. No need to have the baggage of excess words. Streamline. Get to the brass tacks of the words that we use.
In terms of the AI making up words and we take those for granted, including that we neglect humankind-devised words, well, you could argue that AI is acting as a catalyst for organic linguistic innovation. Think about it. Perhaps AI will devise a better language than any language that humans have ever devised.
Early humans came up with words. Words have evolved by humans. The next step might be AI-evolving words, going further than we could have conceived. Voila, AI aids humankind to rise to heightened levels.
That’s an upbeat picture. Happy face.
A final thought for now.
Mother Teresa famously made this insightful remark: “Kind words can be short and easy to speak, but their echoes are truly endless.” Suppose that AI took humanity step by step toward nearly exclusively using only kind words. All that we say to each other consists entirely of kind words.
The echoes would be truly endless.
