The most striking development in English between 2022 and 2024 CE was that an ordinary question could become an instruction to a computer system capable of producing long, fluent replies. OpenAI introduced ChatGPT on 30 November 2022. By 2023, educators and families were discussing the opportunities and risks of generative AI in classrooms. UNESCO published its guidance on generative AI in education and research on 7 September 2023, followed by an AI Competency Framework for Students in 2024. Meanwhile, Oxford’s Word of the Year for 2023 was rizz, a reminder that everyday speakers were still making English lively and inventive outside the world of machine-generated prose.
What happened to the English language? Its syntax did not suddenly transform into a new language understood only by computers. The change concerned the functions people assigned to ordinary sentences. A question could ask a person for information, or it could direct a system to summarise a book, draft an email, explain algebra or imagine a story. The same form of written English was increasingly becoming a control interface for another kind of technology.
But a control interface is not a guarantee of knowledge. If a system answers an uncertain question confidently, the reader still needs to establish whether the statements are accurate. If a student can copy a composition in seconds, the teacher still needs to find out what the student can understand, explain and produce independently. The first principle of this new era is simple: generating an answer and understanding why that answer is justified are different activities.
Follow the chronology: 2021–2022 CE: Vax, Hybrid Learning and Returning · 2020–2026 CE: the wider history of pandemic words and AI · Master of English Language.
Imagine a library that answers in paragraphs
For centuries, a reader could ask a librarian where to find an answer. The librarian might direct the reader to a shelf, catalogue or expert. On the Web, a search engine could provide pages matching certain words. Then a conversational AI system introduced another experience: the user types a question and receives what looks like a direct explanation.
That change is convenient. It reduces the distance between the question and a readable response. It can make an unfamiliar subject feel less intimidating, particularly when the user requests a simpler explanation or asks a follow-up question.
The difference from an ordinary library is important, however. A library catalogue normally points towards identifiable works with authors, dates and editions. A generated paragraph may be readable without identifying which sources, if any, support each claim. The user must not confuse a helpful summary with a verified reference.
30 November 2022: ChatGPT arrives as a public research preview
OpenAI’s original announcement, dated 30 November 2022, introduced ChatGPT as a conversational model. It highlighted the ability to respond to follow-up questions, reconsider a premise and operate through exchanges rather than through one fixed input and output.
The release did not invent machine learning, chatbots or natural-language processing. Each belonged to a longer history of research. The significant public development was an accessible interface through which a non-specialist could explore what a generative language system might do.
The title ChatGPT became a familiar word in discussions of study, writing, work and authorship. But naming a product and understanding the capabilities of its changing versions are separate questions. A statement about a system in late 2022 should not automatically be treated as true of every later version.
What is a prompt?
A prompt is an input used to guide a system’s response. In everyday generative-AI interactions, the prompt is often written in natural language: a question, instruction, example, constraint or set of materials that the system is asked to work with.
The word itself is much older than this technological use. Teachers prompt students to answer; actors may need prompting when they forget a line; a written cue can prompt a memory. The newer digital use applies an established idea to a machine interface.
Consider the instruction ‘Explain photosynthesis.’ It gives the system a topic, but not the age of the learner, the desired depth or the purpose of the explanation. The response may therefore miss the user’s actual needs even if much of it is factually correct.
Now consider ‘Explain photosynthesis to a Primary 4 learner in Singapore. Begin with the role of sunlight, use one simple example and ask two questions to check understanding.’ The second instruction better specifies the audience, sequence and desired action.
Better instructions do not make false statements true
The more detailed prompt may produce a more appropriate format. That is valuable. But a precise prompt does not turn the model into an infallible authority. The system may still misunderstand a source, omit a condition or generate a plausible but incorrect example.
This gives us two separate standards. First, did the system follow the requested presentation? Second, is the resulting explanation accurate? A beautifully formatted answer can pass the first test and fail the second.
English proficiency in the AI era therefore includes the ability to state a task clearly and to evaluate what returns. Those skills are related, but they should not be merged into the single claim that a person who writes good prompts necessarily understands the subject.
A familiar English sentence becomes an operational command
For a human reader, the difference between ‘Please summarise this paragraph’ and ‘Please translate this paragraph into Malay’ is a difference in purpose. For an AI interface, those words can also determine the kind of output generated.
The language user is no longer only sharing an idea. They are selecting an operation. Verbs such as summarise, compare, rewrite, extract and explain become practical choices with different results.
That is one reason instruction-writing deserves serious educational attention. An imprecise verb can lead a system—or a student—to do the wrong job. To describe something is not necessarily to evaluate it. To simplify a text is not to remove every necessary qualification.
The danger of a coherent but unsupported paragraph
Imagine asking for the year in which a particular poem first appeared. The system responds with a date, an author and a confident explanation. None of the sentences contains obvious grammatical errors. Yet the cited date may refer to a revised edition, not the first publication.
A historian cannot decide which date is right by counting polished sentences. The next action is to find a reliable publication record or a scholarly catalogue and check which edition is being discussed.
The same principle applies to invented quotations. A machine-generated line can sound exactly like a famous writer without existing anywhere in that writer’s work. Before using the quotation in an essay, the student should locate the passage in an identifiable edition.
The reader’s job is to trace the claim back to the evidence, not merely decide whether the paragraph sounds convincing.
Source identity becomes part of English competence
Traditional reading lessons ask students to identify topic, purpose, tone and audience. Those skills remain necessary. But machine-generated explanations place additional importance on identifying the source of factual material, not only the style of its presentation.
A student may need to distinguish an official syllabus from a third-party summary, an original research paper from a news article and a public-domain literary text from a modern paraphrase. Each can be useful for a different purpose.
When an answer refers to a source, the link should be checked. Does it open a real page? Does that page support the specific claim? Is the document current for a topic whose details may change? These questions are practical extensions of reading comprehension.
2023: Oxford chooses rizz while people keep inventing slang
In December 2023, Oxford University Press selected rizz as its Word of the Year. The term, associated with charm or attractiveness, had grown in prominence through youth language and online culture.
The choice demonstrates that the public history of English was not exclusively about AI. People continued adapting language to describe relationships, fashion, identity and humour. New slang often becomes widely recognisable through repeated interaction rather than through formal instruction.
The word’s popularity does not require every reader to use it. A formal report may need a more explicit description, while a conversational sentence among familiar listeners may communicate efficiently through slang.
This is the difference between describing usage and prescribing register. A dictionary can document the word; a teacher can help students decide when a particular vocabulary choice suits the task.
7 September 2023: UNESCO publishes generative-AI guidance
The United Nations Educational, Scientific and Cultural Organization published Guidance for Generative AI in Education and Research on 7 September 2023. It addressed human-centred use, privacy, age appropriateness, regulation, educational design and the need for institutions to develop their capacities rather than simply adopt rapidly changing tools.
The document’s importance lies in refusing to treat technological fluency as an educational outcome by itself. A system that can supply explanations still needs to be evaluated against the real needs of teachers, learners and researchers.
This is particularly important for children. A learner may accept a fluent answer because it resembles a teacher’s voice. Responsible use requires adult judgment, clear purposes and opportunities to question what the system produces.
The challenge cannot be solved by a single sentence telling students to ‘use AI carefully’. Educational institutions need methods for checking accuracy, protecting information and ensuring that the student remains intellectually active.
A good tutor asks what the learner can do next
Imagine a student asking an AI system for a composition about a memorable journey. The tool offers a polished story with effective adjectives and a neat ending. The result can serve as an example of structure, but copying it does not demonstrate the student’s own planning, vocabulary control or writing ability.
A better learning sequence is more demanding. First the student explains the event they want to describe. The system or teacher then helps organise the sequence, prompts the student to improve one paragraph and asks why a particular sentence works.
The learner writes again without copying the example, applies feedback and tests whether the skill transfers to a different topic. In this way the tool supports human capability rather than replacing the evidence that capability exists.
The difference between assistance and substitution
A spellchecker may help someone catch a typo. A dictionary can clarify a meaning. A writing assistant can suggest a clearer sentence. These tools have different effects on what the learner is doing.
The crucial question is which skill is being assessed. If the purpose is to learn punctuation, accepting every correction without explanation may limit the learning. If the purpose is to communicate an urgent message accurately, using a correction tool may be appropriate.
There is no single rule that makes every kind of assistance either good or bad. The educational design must identify the target capability and decide which assistance makes that capability stronger.
2024: UNESCO defines competencies rather than software tricks
In 2024, UNESCO published its AI Competency Framework for Students. The framework describes 12 competencies across four dimensions: a human-centred mindset, ethics of AI, AI techniques and applications, and AI system design. These appear across three progression levels: Understand, Apply and Create.
This is a more durable approach than teaching one fashionable prompt formula. Particular products will change. Students need ways to understand what AI systems do, recognise risks, make appropriate decisions and participate responsibly in a world where such systems may be common.
The Understand stage includes recognising basic concepts and limitations. Apply asks whether a person can use that understanding appropriately. Create adds more complex design and evaluation work. Learners need the foundation before the more advanced action becomes meaningful.
The framework does not make students experts in one afternoon. It gives educators an organised set of competencies that can develop over time.
Singapore English remains a living local resource
During this period, students in Singapore were navigating formal academic English, everyday Singapore English, mother-tongue languages and a growing amount of international digital material. An AI-generated paragraph might produce a style associated with a different country, age group or curriculum.
A responsible learner should notice these differences. A Primary-school comprehension answer needs the meaning and structure expected at that level. A Secondary-school argument requires stronger evidence and organisation. A professional communication may need a more explicit tone and better documentation.
A tool can help present alternatives. It cannot determine the student’s actual learning need without relevant context and informed judgment.
What a machine can generate is not what a reader must accept
A generative system is a source of possible text. Its output can be informative, creative, irrelevant or mistaken. The language can be analysed just like other writing: identify the claim, inspect the reasoning, trace the evidence and decide what would justify the conclusion.
One additional habit is to keep the task and the output separate. An instruction may request a factual explanation; the generated paragraph is a proposed response to that instruction. It should not be elevated to the status of established evidence merely because it appeared directly beneath the prompt.
The same discipline protects writers as well as readers. When publishing a claim, the author remains responsible for whether it is supported, even if a tool helped produce the wording.
A practical five-stage guide to English and AI
- Define: State the question, intended audience and level of detail.
- Generate: Obtain a possible explanation or example, recognising that it is provisional.
- Verify: Check consequential facts, dates and quotations against identifiable sources.
- Reconstruct: Explain the reasoning independently and identify any missing steps.
- Transfer: Apply the understanding to a new question rather than memorising one answer.
The last two stages distinguish learning from passive acceptance. A student who can rebuild an argument and use it in a new situation has evidence of capability. A student who can only paste a smooth paragraph has evidence of access to a tool.
Questions readers often ask
Did AI first begin in November 2022? No. AI research and natural-language systems existed for decades before ChatGPT’s public release.
Is a prompt a new English word? No. The word has older uses; its role as a common digital instruction became more prominent through AI interfaces.
Was UNESCO’s generative-AI education guidance released in 2024? The guidance appeared in September 2023. UNESCO’s AI competency framework for students was published in 2024.
Does a fluent AI response prove a statement is true? No. Accuracy requires evidence and contextual checking.
Does a student who receives a perfect AI essay automatically possess the writing skill shown in it? No. The learner’s capability should be demonstrated independently.
What comes after the first wave of generative AI?
Continue to 2024–2026 CE: Brain Rot, Rage Bait and AI Literacy in Singapore. The next chapter examines how readers responded to more abundant machine-generated writing and how schools began designing for stronger judgment rather than mere output. The Master of English Language connects the full history.
Leave a Reply