07.02 AI for Research & Academy Writing

Main ideas, concepts, and lessons

• Scientific paper writing is more than presenting data

  • It requires collecting information, structuring arguments, and ensuring the paper is of good quality.

• AI can help simplify the research-to-writing workflow

  • AI is described as support for:
    • Generating research topics and ideas
    • Analyzing research data (e.g., experimental/statistical analysis support)
    • Conducting literature review and synthesis
    • Plagiarism checking
    • Supporting academic activities (research design, content development, structuring)
    • Editing/review support and publishing/communication outreach
    • Ensuring ethical compliance for journal/conference publication

• AI usage distribution in academic domains (percentages given)

  • Content development & structuring: ~23%
  • Literature review & synthesis: ~9%
  • Data management: ~12%
  • Editing/review & publishing support: ~26%
  • Communication outreach & ethical compliance: ~28%
  • Idea development & research design: ~2%
  • Key takeaway: AI is currently used far more for writing/publishing support and less for core research design/idea generation.

• Important cautions when using AI (especially Large Language Models like ChatGPT, Gemini, Perplexity)

  • Accuracy
    • Users must verify factual claims against literature.
  • Reliability / hallucinations
    • LLMs can produce wrong or misleading information presented as true (“hallucination”).
    • Example statistic mentioned: hallucination level for GPT-4 ~28.6% (as stated in the subtitles).
  • Co-author policy
    • Many journals (including those discussed) do not allow AI as a co-author.
  • Data privacy
    • Be careful if inputting sensitive data into AI tools—inputs may be used to help train or improve models.

• Journal/Publisher policy examples (must be followed)

  • Nature
    • Does not allow large language models (e.g., ChatGPT) as co-authors.
    • Use of AI must be documented in the Methods section if submitting to Nature.
    • AI-generated images: generally not allowed, with exceptions (e.g., contractual/allowed generation).
  • Elsevier
    • AI may only be used to improve readability and language (grammar, wording, structure).
    • Should not be used to generate scientific/medical insights or draw conclusions.
    • If AI is used for language improvement, it must be declared.
    • Generative AI for images from text is generally not allowed, with exceptions referenced in their policy.

• Prompts (“prompt engineering”) as the core method to get useful outputs

  • Definition
    • A prompt is a piece of text/instructions given to a large language model to get specific responses.
    • Prompt engineering = the skill of writing prompts effectively.
  • Best practices listed
    • Define a clear objective
    • Provide context (background + relevant details/examples)
    • Use natural language
    • Be specific, not vague (avoid ambiguity)
    • Iterate and improve the prompt through refinement

---

Methodology / instruction-like content (detailed bullet list)

A) Areas where AI can support academic writing and research

• Generating research topics and ideas
• Assisting with analysis of collected data (experiments, statistics)
• Conducting literature review
• Plagiarism checking
• Supporting academic functions such as:
  • Research design & idea development
  • Content development & structuring
  • Literature review synthesis
  • Data management (experimental/statistical data)
  • Editing and review
  • Publishing support and communication/outreach
  • Ethical compliance for journal/conference publication

B) How to use AI safely/appropriately (checklist implied by cautions)

• Accuracy
  • Verify AI-generated content using relevant literature
• Reliability
  • Treat outputs as drafts; watch for hallucinations
• Authorship
  • Do not list AI tools as co-authors (per journal rules)
• Privacy
  • Avoid uploading sensitive/personal research data into AI systems when not allowed/necessary
• Compliance
  • Follow journal policies on:
    • AI use disclosure (e.g., Methods section)
    • Whether AI-generated images are allowed
    • Whether AI can be used only for language vs. also for insights/conclusions

C) Prompt engineering best practices (explicitly stated)

• Define a clear objective (what you want the model to do)
• Provide context/background, including examples if helpful
• Write prompts in human-readable natural language
• Make prompts specific (avoid overly general requests)
• Use iteration:
  • refine the prompt repeatedly
  • adjust based on the model’s response quality and relevance

D) Examples of prompt use cases mentioned

• Searching prior literature
  • Prompt includes a topic (e.g., “challenges implementing deep learning”)
  • AI returns literature items with summaries (users still must verify correctness)
• Summarizing a complex paper
  • Provide the paper title
  • Ask for: background/context, objective, methodology, key findings, implications, conclusions, and important terms
• Formulating research questions
  • Provide topic + desired research framing
  • Request: context, justification, relevance, variables, methodologies, expected outcomes
• Developing research methodology
  • Provide a field/topic
  • Request appropriate quantitative and/or qualitative methods
  • Ask for limitations of each method and possible solutions
• Creating an outline of a research paper
  • Provide topic (e.g., “generative AI for business”)
  • Request sections such as:
    • intro/background
    • problem statement, relevance, objectives
    • literature review
    • application
    • challenges/ethics
    • methodology
    • contributions/novelty
    • bibliography
    • conclusions
• Finding sources from data repositories
  • Provide topic and request relevant journals/reports/datasets/books/courses
• Responding to reviewer comments
  • Provide reviewer feedback
  • Ask for revision suggestions (while still verifying the correctness of responses)
• Assisting with abstract writing
  • Provide topic/paper details
  • Ask AI to draft the abstract (then revise and verify)
• Ethics considerations
  • Provide topic (e.g., gene sequencing) and ask for relevant ethical standards and responses

---

AI for scientific discovery (additional main point)

AI is portrayed not only as a writing tool but also for scientific discovery, such as:

• Finding new molecules
• Discovering/designing/manufacturing materials
• Drug discovery (shortening time with predictive models)
• Designing proteins

General conclusion: AI is useful for discovery and research, but must be used wisely, responsibly, and ethically, and in line with publisher/journal requirements.

---

Speakers / sources featured

Speakers

• The video presenter: Bambang Rianto (“Welcome back to Bambang Rianto”)

AI tools/models mentioned

• ChatGPT (OpenAI; includes references to GPT-4)
• Gemini (Google)
• Perplexity
• Copilot (Microsoft)
• Claude / Cloud (Anthropic is mentioned)
• Other tools named: QuillBot, Wordtune / other “etc.” (bit.ai / others mentioned as examples)

Journals/publishers mentioned

• Nature
• Elsevier

Examples/resources mentioned

• Springer (as an example of publisher in the literature search example)
• Google dataset / Google Dataset, GitHub, NCLBI (NLM/NCBI mentioned as “us National Center for biotechnology information”)
• NEA (National Endowment for the Arts)
• Other report/data/course repositories mentioned generally
• Nature (again mentioned for the AI protein design example)