What is the 'power of language' that remains for humans in the age of generative AI?

I am Kumataro Yokoyama, a training instructor and sign language interpreter.

While utilizing AI, I provide information for seniors and introduce books and gadgets.

Recently, the books I have been reading the most are those by Mutsumi Imai.
Although the content is quite difficult for me, it contains many great hints for those learning language, and I read through it with excitement.

The presence of AI is growing in the environment surrounding language.
I myself am also writing articles in collaboration with AI.

I believe that what a creator wants to convey—their passion and warmth—is lost when left entirely to AI.

So, where does the balance lie between what humans can do and what AI can do?

Relying on this article by Ms. Imai, I would like to think about it together with you.

Generative AI has become able to write surprisingly natural sentences.

If you ask a question, you get an answer immediately.
The sentences are well-structured.
There is no hesitation.
The grammar is also perfect.

It almost looks as if it has already surpassed humans in the 'power to use language'.

However, when reading Mutsumi Imai's 'The Power of Language and Thinking Required in the Age of Generative AI,' one realizes the opposite.

Precisely because we are in an era where generative AI handles language fluently, we humans have a 'way of using language' that only humans can possess.

AI is fluent, but it does not live the meaning

Generative AI learns the relationships between words from vast amounts of data and predicts the words most likely to come next.

That is why the sentences are so natural.

However, Ms. Imai points out an important perspective here.

AI can explain what an 'apple' is.
But it does not know the scent, the texture, the weight in the hand, or the sweetness when bitten into through physical experience.

In other words, the language is not grounded in real-world experience.

This is expressed as 'lacking symbol grounding'.

For humans, language is not merely a set of symbols.
It is intertwined with physical experience, senses, memories, and interactions with others.

That is precisely why human language has depth.

Children do not learn language by being 'taught'

What is particularly impressive in this article is the explanation of children's language acquisition.

Children do not learn language by having meanings neatly explained to them by adults.

They observe the situation, consider the other person's intent, and form a hypothesis, thinking, 'Does this mean this?'
Then they try using it.
They make mistakes.
And they correct themselves again.

Through this repetition, language becomes their own.

What becomes important here is 'abductive reasoning'.

Abduction is a form of reasoning where one forms a hypothesis from incomplete information, thinking, 'This is likely the case.'

For example, when someone we are meeting does not show up, we immediately think about the cause.

Maybe there was an accident.
Maybe they forgot.
Maybe something came up.

Even without concrete evidence, humans cannot help but form hypotheses.

This power is at work when scientists form hypotheses, when children learn language, and when we read the intentions of others in daily conversation.

'Being able to calculate' and 'understanding the meaning' are different

Ms. Imai also introduces examples of children who struggle with arithmetic.

They remember the calculation procedures.
They can find common denominators and simplify fractions.
But they do not understand the meaning of the fractions themselves.

For example, they cannot grasp the sense that 1/2 falls in the middle of a number line from 0 to 1.
They get confused about whether 1/2 or 1/3 is larger.

This is not merely a problem of academic ability.

It is a problem that manipulating symbols is different from understanding meaning.

And this aspect is somewhat similar to generative AI.

It can produce plausible answers.
It has fragmented knowledge.
However, that meaning is not rooted in physical sensation or real-world experience.

Herein lies a major hint for thinking about human learning.

The power of language is not just the power to speak correctly

When people think of the power of language, many likely think of vocabulary, reading comprehension, and expressive ability.

Of course, those are important too.

However, what this article teaches us is that the power of language is the 'power to grasp meaning'.

Why did the other person use those words?
What does that expression refer to in that situation?
How does the meaning of the same word change depending on the context?

The ability to think about these things is the depth of human language.

Even in an age where generative AI creates text, the things humans should be responsible for will not disappear.

Rather, I think the importance of the 'power to think about meaning,' which was hard to see through fluency alone, has become clearer.

A perspective that also applies to sign language interpreting

While reading this article, I felt something that strongly applies to the world of sign language interpreting as well.

Sign language interpreting is not the task of replacing words with other words.

What is the speaker trying to say?
Which words should be chosen on the spot to convey the message?
What is a natural expression for a deaf person?
How should the context of the hearing person be supplemented?

Always amidst incomplete information, we form hypotheses, make judgments, and choose expressions.

This is truly a series of abductions.

Therefore, what is required of a sign language interpreter is not just vocabulary.

The power to grasp the meaning of language with the body.
The power to read context.
The power to infer the other person's intent.
And the power to revise one's own interpretation as needed.

Even in the AI era, this should remain a very important expertise for human interpreters.

What is needed in the generative AI era is the 'power to question'

Generative AI is convenient.

Creating drafts of text.
Organizing information.
Suggesting alternatives for phrasing.

It is very useful in these situations.

However, simply accepting the text produced by AI will not develop human thinking skills.

What is important is to re-examine things by asking questions like these.

Is that really true?
What does this word mean?
Is there another perspective?
Will this explanation convey the meaning to the other person?
Is it connected to real-world experience?

The more we use AI, the more humans are required to have the 'ability to think about meaning'.

Conclusion

Mutsumi Imai's article does not deny generative AI.

Rather, by observing the characteristics of generative AI, it brings to light the essence of human language and thought.

AI speaks fluently.
But humans experience, hesitate, make mistakes, and revise as they make language their own.

It is precisely in that clumsiness that human intelligence resides.

What is needed in the age of generative AI is not to write text faster than AI.

Thinking about the meaning behind words.
Formulating hypotheses from incomplete information.
Understanding by connecting it to experience.
And, re-perceiving the world in one's own words.

I believe that what will be questioned in the coming era is not fluency, but depth.

Mutsumi Imai's book can be listened to on Audible.

#GenerativeAI #PowerOfLanguage #ThinkingSkills #Abduction #SymbolGrounding #LanguageAcquisition #MutsumiImai #Education #Learning #AIEra #ReadingComprehension #SignLanguageInterpretation #ThinkingAboutMeaning #NoteArticle #MitaReview