8.3 English and Artificial Intelligence

The Machine That Writes

In November 2022, OpenAI released ChatGPT, and the world of language was never quite the same. For the first time in history, a machine could produce fluent, coherent text often indistinguishable from human writing -- and within two months it had 100 million users, the fastest-growing consumer application in history.

But AI's relationship with language did not begin in 2022. Machine translation tools like Google Translate had been improving for over a decade, voice assistants had been processing natural language since the early 2010s, and autocomplete had been quietly shaping our writing even longer. What changed with large language models (LLMs) like GPT-4, Claude, and Gemini was the scale and sophistication. These systems do not simply follow rules -- they have been trained on vast amounts of text and learned statistical patterns of language. This raises profound questions: If a machine can write fluent English, what does that mean for human writers? Should we still learn foreign languages if AI can translate instantly? Who is the "author" of AI-generated text? These are not abstract questions -- they affect your education, your career, and your relationship with language right now.

How AI Translation Works -- and Where It Fails

Machine translation has gone through three generations. Rule-based translation (1950s-2000s) relied on grammatical rules and bilingual dictionaries, producing stilted, often comical results -- a famous (possibly apocryphal) example translated "The spirit is willing but the flesh is weak" into Russian and back as "The vodka is good but the meat is rotten." Statistical machine translation (2000s-2016) used large bilingual text collections to find patterns, better but still unnatural. Neural machine translation (2016-present) uses deep learning neural networks trained on massive datasets, producing remarkably fluent, natural-sounding translations -- Google switched to neural MT in 2016, dramatically improving quality. Today AI handles common language pairs with high accuracy and can translate documents, websites, and real-time conversations.

But persistent limitations remain. Rare language pairs are poorly served, cultural nuance (humour, irony, wordplay) is often lost, specialized terminology can be mistranslated, literary quality stays below human translators, and low-resource languages have far less training data. You can see this in practice. For a factual text -- "Norge er et land i Nord-Europa" → "Norway is a country in Northern Europe" -- AI is excellent. But for an idiom like the Norwegian "Det var helt Texas" (meaning chaotic/wild), AI translates literally as "It was completely Texas," missing the cultural meaning entirely; a human would write "It was total chaos." And for literary prose, such as a passage from Hamsun's Sult, AI produces grammatically correct but flat output, losing the rhythm and emotional resonance. The pattern is clear: AI excels at information transfer but struggles with cultural meaning, emotional nuance, and artistic voice -- precisely the elements that make language human.

Should We Still Learn Languages -- and Who Is the Author?

The power of AI raises a sharp question: why learn a foreign language if machines can translate for you? There are strong arguments for continuing. First, understanding beyond words -- language carries culture, humour, and worldview; concepts like Norwegian "koselig" resist translation because they encode entire cultural perspectives. Second, cognitive benefits -- learning languages improves memory, problem-solving, and even delays cognitive decline. Third, social and professional advantage -- speaking someone's language builds trust in ways a translator cannot. Fourth, critical thinking about AI output -- you need language knowledge to judge whether a translation is accurate. And fifth, creative expression -- AI can translate what you say, but it cannot decide what you want to say. At the same time, AI can enhance language learning through personalized practice, instant feedback, immersive AI conversation partners, and scaffolding. The balanced view is that AI is a powerful tool for language learning, not a replacement -- the goal should be human competence enhanced by AI, not human dependence on AI.

A second profound question is authorship. If a student uses ChatGPT to write an essay, who is the author? Consider four scenarios. In full AI generation (typing a prompt and submitting the output unchanged), most would say this is not the student's work -- it is plagiarism. In AI as brainstorming partner (discussing ideas, then writing independently), this is a grey area, similar to discussing ideas with a tutor. In AI as editor (writing independently, then asking AI to check grammar), most would say this is like using spell-check. And in iterative collaboration (generating a draft, then substantially rewriting and adding personal examples), the final product reflects significant human judgment but came from AI -- complex. Key principles are emerging: transparency is essential (acknowledge AI use), the thinking matters more than the polished output, context determines ethics, and understanding what AI produces is necessary to use it responsibly. There is no simple answer yet -- societies and institutions are still negotiating these boundaries. AI also affects linguistic diversity: languages with abundant digital text (especially English) are well-served, while minority languages may be further marginalized.

Summary

AI has transformed how we produce and process language. Machine translation evolved from rule-based to statistical to neural systems, which produce remarkably fluent output -- yet still struggle with cultural nuance, idioms, humour, literary quality, and low-resource languages, as the Norwegian idiom "Det var helt Texas" shows. AI is a powerful tool for language learning, not a replacement: the cognitive, social, and cultural benefits of learning languages remain valuable, and you need language knowledge to evaluate AI output. The authorship question is unresolved -- from full AI generation (plagiarism) to AI as editor (like spell-check) -- with transparency, the value of thinking, and context all mattering. AI also affects linguistic diversity, favouring well-resourced languages like English. Critical AI literacy -- understanding what AI can and cannot do -- is a new essential skill.