14752 Chap.4 Generative AI and Research Integrity
Generative AI and Research Integrity
How language models work
Large language models are neural networks with billions or trillions of parameters, trained to predict the next word or token from large amounts of internet text and then fine-tuned with human feedback. Their linguistic basis is distributional semantics, the idea that a word's meaning comes from the company it keeps.
Two behaviours matter for researchers: hallucination and bias.
Risks and debates
The lecture lists issues from environmental impact and deepfakes to job losses, intellectual property theft and surveillance, several of which are researchable communication topics.
The Yellow rule
Both assignments are Yellow, the University's Type B: generative AI may support researching and developing ideas, but no AI-generated content may appear in a submission.
The Code of Practice also treats fabricated data as misconduct, a decisive point in a module built around collecting your own dataset.
Where to get support
The University provides Microsoft Copilot for study use, and the lecture points to a Canvas course on using generative AI tools for study, guidance on using Grammarly responsibly, and the University's codes of practice for detail.
Lecturers' own materials should never be fed into an AI tool without their explicit permission. When in doubt about whether a particular use is allowed for an assignment, ask the module lead before submitting rather than after.
What this chapter covers
- 01
AI, generative AI and language models
- 02
Distributional semantics
- 03
Next-word prediction
- 04
Hallucination and bias
- 05
Challenges of generative AI
- 06
Yellow Type B rule
- 07
Code of Practice breaches
Worked example · free
Classify three uses of AI under the Yellow rule
- 1(a) Suggesting topics is idea development, which Type B permits.
- 1(b) Rewriting proposal text puts AI-generated content into the assessment, which is not permitted.
- 1(c) Feeding lecturers' materials into AI needs the lecturer's explicit permission.
Key terms
- Large Language Model
- A large language model is a neural network trained to predict the next word or token from very large amounts of text.
- Distributional Semantics
- Distributional semantics is the idea that a word's meaning comes from the patterns of words it co-occurs with.
- Hallucination
- A hallucination is a confident but false statement produced by a language model.
- Type B Assessment
- A Type B assessment allows generative AI for researching and developing ideas but not for content included in the work.
- Next-Token Prediction
- Next-token prediction is the task a language model is trained on: choosing the most likely continuation of a text.
Generative AI and Research Integrity FAQ
Why do AI chatbots make things up?
Language models are trained to predict likely next words from patterns in their training data rather than to check facts. The lecture notes that this makes them prone to hallucination, confident false statements, and to bias inherited from the text they learned from.
Is proofreading with AI allowed in this module?
The lecture lists proofreading among general ways students might use AI, but the module's Yellow rule forbids AI-generated content in the assessment. Reworded sentences from a tool would be such content, so make every edit yourself or use the University's writing support.
Why is fabricated data treated so seriously?
The University's Code of Practice names fabricating data alongside AI-generated and third-party work as plagiarism and misconduct. In a module where the second assignment presents a dataset you collected, invented items would undermine every claim the commentary makes.
What does Yellow mean on the assessment table?
The module uses colours aligned with the University's generative AI policy types: Red for Type A, Yellow for Type B and Green for Type C. Both assignments here are Yellow, so AI may help you research and develop ideas, but its output must not appear in submitted work.
Assessment move
Write a one-line rule for yourself that matches the Yellow statement, and keep it beside your draft. List every AI use you are tempted to try for the assignments and sort each into permitted, not permitted or needs permission. Explain distributional semantics to a friend using a pair of near-synonyms of your own, as the lecture did with bring about and cause.
If AI interests you as a topic, draft a research question about how people talk about it rather than using it to write. Keep a short log of any AI use during the project, so you can state honestly how it helped if you are asked. Note in the same log any claim from a chatbot that you later checked, and whether it turned out to be true.
Working through Generative AI and Research Integrity in 14752? Sia is AskSia’s AI Media and Communications tutor — ask any 14752 Generative AI and Research Integrity question and get a clear, step-by-step explanation grounded in how 14752 is taught and assessed. Read this chapter free, then take your hardest questions to Sia.