Skip to content

1.1 Current State and Potential of Generative AI

What you will learn on this page

  • The basic mechanisms of generative AI and what it can do today
  • Functions relevant to academic writing, such as text generation and structured output
  • What generative AI is good at and where its limits are
  • Why generative AI matters for researchers who are not native English speakers
  • A collaboration model for researchers and AI

What is Generative AI?

Generative AI (Generative Artificial Intelligence) is a technology that learns language patterns from large-scale text data and can produce natural-sounding text that resembles human writing. Representative tools include OpenAI’s ChatGPT, Anthropic’s Claude, and Google’s Gemini.

These tools generate text based on given instructions (prompts). They can support many tasks related to writing academic papers in English, such as translation, summarization, paraphrasing, and grammar checking.

You do not need to understand the internal mechanism in detail. However, it is helpful to remember that the output is optimized to produce “plausible-sounding text,” which makes judgment and verification easier.

A diagram to grasp the overview of LLMs

AI capabilities relevant to academic writing

Text generation and transformation

This is where generative AI is especially useful for writing papers.

  • L1 → English transformation: Convert claims written in L1 (e.g., Japanese) into natural academic English
  • Paraphrasing: Suggest alternative expressions with the same meaning
  • Grammar and usage explanations: Explain why a certain expression sounds unnatural
  • Logical structure checks: Check whether claims and evidence align

Structured output

Generative AI can organize output into specific formats according to your instructions.

  • Organizing issues in bullet points or numbered lists
  • Comparing items in a table format
  • Suggesting an outline aligned with IMRaD sections

Strengths and limitations of generative AI

What it does well

  • Producing fluent and natural English
  • Constructing grammatically correct sentences
  • Offering multiple wording options
  • Pointing out issues in logic and coherence

What it struggles with and what requires caution

  • Factual accuracy: Generative AI may confidently produce “plausible falsehoods” (hallucinations)

  • Accurate counting: Because it predicts the next word from context, it may not strictly follow instructions such as “summarize in exactly 100 words.”

    Token counting in generative AI differs from ordinary word counts

    An image of LLM tokenization

  • Up-to-date research trends: It may not know studies published after the cutoff of its training data

  • Specialist judgment: Humans must evaluate novelty and significance
  • Generating bibliographic information: It may fabricate papers or DOIs (digital object identifiers)

A crucial point

Generative AI is an “excellent language assistant,” not a “co-researcher.” Accuracy of content, the significance of the research, and ethical judgment are entirely the researcher’s responsibility.

Supporting researchers who are not native English speakers

Researchers whose first language is not English are reported to face additional time costs and psychological burdens at many stages such as attending international conferences, presenting research, writing papers, and responding to revisions (Amano et al., 2023). Generative AI can help reduce these burdens by breaking them down into manageable tasks, for example by offering paraphrases, helping you review the logic of your argument, suggesting fixes for grammar and usage, and assisting with draft responses to reviewer comments. Of course, the final responsibility for accuracy and decisions remains with the researcher, but reducing friction in English processing can make it easier to secure more time to focus on the substance of the research.

Common barriers faced by non-native English speakers in science (Amano et al., 2023)

Reference: Amano et al. (2023). The manifold costs of being a non-native English speaker in science. PLOS Biology, 21(7), e3002184. https://doi.org/10.1371/journal.pbio.3002184

A collaboration model for researchers and AI

When integrating generative AI into academic writing, it is important to design a workflow that satisfies both efficiency and academic integrity. A basic approach is to divide roles, leveraging the researcher’s expertise (original ideas and critical thinking) and the AI’s strengths (improving language, speeding up tasks), and always include a final verification step.

A collaboration model for researchers and AI