Generative AI and research ethics

PhD Seminar · Chemical Engineering

Verify the output. Keep responsibility for the research.

How to use this lesson

01

Watch the worked examples and complete the short pauses.

02

Submit the 10-question quiz with your name and student ID.

03

Finish with the exit ticket and anonymous feedback.

Choose the check before using the tool

01

Language editing: compare meaning before and after.

02

Literature support: open the original source.

03

Code assistance: define an independent check.

Human responsibility

01

AI tools are not authors under the COPE position.

02

Human authors remain responsible for the manuscript.

03

An accepted output needs a named human reviewer.

The battery evidence supplied

Item Supplied evidence
Cell and temperature One LFP/graphite model; 25 °C
Mean retention at cycle 200 Protocol A: 88%; protocol B: 94%
Independent units 3 cells per protocol; common diagnostic
Not supplied Uncertainty, lifetime endpoint, plating diagnostics

Synthetic classroom example. No experimental or published result is claimed.

A fluent edit changes the evidence

01

Supplied: mean retention was 88% for A and 94% for B at cycle 200.

02

Mock edit: “B extends battery lifetime by 6%, eliminates lithium plating, and works under all conditions [REF-X].”

03

REF-X is an invented classroom placeholder.

Pause: audit the paragraph

60seconds
01

Identify the unsupported outcome, mechanism, scope and reference.

02

Write one sentence that the supplied data support.

03

Use Worksheet 4A or a piece of paper.

Audit answer: retain the measured outcome

Claim Decision and reason
Longer lifetime Revise: retention at cycle 200 was measured.
Eliminates plating Remove: no plating diagnostic supplied.
All conditions Restrict: one cell model at 25 °C.
REF-X Remove: invented, unverified reference.

The number and its denominator

01

94% − 88% = 6 percentage points in mean retention.

02

(94 − 88) / 88 × 100 ≈ 6.8% relative to A.

03

Neither calculation measures a lifetime extension.

Arithmetic from synthetic teaching values; no inference about statistical significance.

A defensible revised paragraph

01

At cycle 200 and 25 °C, mean retention was 88% for A and 94% for B in the tested LFP/graphite cell model.

02

The observed difference was 6 percentage points, with 3 independent cells per protocol.

03

No uncertainty estimate or plating diagnostic is supplied.

A reference needs two checks

01

Existence: locate the original publication and verify its details.

02

Support: read the passage or result relevant to your claim.

03

A real DOI does not prove that the citation supports the sentence.

A bounded editing request

01

Improve clarity while preserving values, units and scope.

02

Add no mechanism, reference or factual claim.

03

List changes that could alter meaning; compare the result yourself.

Confidentiality comes before the prompt

01

Identify who controls the material and who may authorize sharing.

02

Check the applicable agreement, institutional rules and tool terms.

03

Use an approved workflow with only the necessary information.

Pause: choose the next action

30seconds
01

A partner sends unpublished cycling data under a project agreement.

02

A colleague removes the company name and wants to upload the file.

03

Is that enough? Write the next action and a workable alternative.

Scenario answer: permission is still unresolved

01

Hold the upload until the sharing authority and terms are confirmed.

02

Removing a name does not remove every confidentiality obligation.

03

Develop the task using synthetic data or an approved local workflow.

Code: test the meaning of the result

Check Expected result
2.7 Ah / 3.0 Ah × 100 90% retention
Later capacity = initial capacity 100% retention
Initial capacity = 0 Flag an invalid denominator
Same formula, wrong diagnostic rows Reject the comparison; inspect data selection

Synthetic values. Test inputs, units, known cases and the data-selection step.

Images: identify what the picture represents

01

Conceptual illustration: an explanation of an idea.

02

Data plot: a reproducible transformation of recorded values.

03

Primary image: an observation, such as an SEM micrograph.

Check the target journal policy

01

Identify the task: manuscript preparation, research method or image.

02

Read the current official policy and the journal instructions.

03

Record the policy URL and the date checked.

Elsevier journal policy example, checked 19 October 2026. Requirements are publisher-specific.

An AI use log records decisions

Field Illustrative entry
Task and material Clarity edit; synthetic battery paragraph
Tool, date and reviewer Record the actual values in your project
Check Compare values, units, scope and sources
Decision Keep wording; reject mechanism and REF-X

Illustrative log structure, not a claim that a named tool performed this exercise.

Disclosure describes actual use

01

Name the actual tool and the task it performed.

02

Explain the review that was actually completed.

03

Use the location and detail required by the target journal.

Pause: plan a verifiable workflow

60seconds
01

Task: improve a nonconfidential paragraph about your own research.

02

Specify what stays fixed and how you will check the output.

03

State what you will record and what triggers rejection.

Workflow answer

01

Limit the task to clarity; preserve values, units and tested scope.

02

Compare each change with the evidence and original sources.

03

Record the accepted edit, rejected claims, reviewer and applicable disclosure.

Online quiz and attendance

10 questions

01

10 questions · Name and student ID required.

02

Unlimited attempts; explanations after submission.

03

One completed submission per student ID counts as attendance.

QR code for session 4 attendance quiz

Attendance is based on quiz completion, not the score or verified viewing time.

Exit ticket and anonymous feedback

3 minutes

01

Exit ticket: name one task, one risk and one verification step.

02

Student ID is optional on the exit ticket.

03

Feedback asks for no name or student ID.

QR code for session 4 exit ticket and anonymous feedback

Reading and policy sources

01

COPE: Authorship and AI tools

02

Elsevier: Generative AI policies for journals

03

Official links and practice worksheet are on the lecture page.

Policies checked 19 October 2026. Recheck the target journal before submission.