Expert Session

How to Use AI in Academic Research: Rigor, Possibilities, and Integrity

Speaker

Marija Dukic

Quick answer

Use SciSpace in academic research where it improves speed or consistency without obscuring methods, evidence, or accountability. Map its role across discovery, reading, analysis, and writing; protect sensitive data; verify outputs proportionately to risk; preserve researcher judgment; and disclose material use when required.

About this session

This session demonstrates how to utilize the SciSpace platform to conduct and automate systematic literature reviews [L6-L7, L13]. Marija Dukic explores strategies for identifying research gaps, mitigating potential threats to integrity, and streamlining labor-intensive academic workflows to enhance research efficiency [L7-L9, L14-L16].

AI expands what researchers can search, compare, draft, and analyze, but it can also accelerate weak assumptions and unsupported conclusions. This session examines the opportunities across an academic workflow and pairs each one with the controls needed for rigor, reproducibility, and integrity.

What this tutorial covers

Identify research tasks where AI adds speed, structure, or coverage.
Match each use case with risks involving scope, bias, sources, methods, data, and writing.
Use verification, reproducibility, and human approval as built-in workflow stages.
Document AI assistance so collaborators, reviewers, and readers can understand its role.

Key concepts explained

Methodological integrity: AI output must follow the actual design and analysis, not invent a cleaner story.
Evidence integrity: Citations and extracted facts require source-level checks.
Reproducibility: Prompts, filters, code, versions, and decisions should be retained when material.
Proportional oversight: The higher the impact of an output, the stronger the validation.

How to try it in SciSpace

1
Map the intended AI uses across discovery, reading, analysis, writing, and review.
Map the intended AI uses across discovery, reading, analysis, writing, and review. Record the benefit sought and the consequence of an error at each stage.
2
Define approved inputs, sources, methods, and prohibited actions for each stage.
Define approved inputs, source boundaries, methods, prohibited actions, and the person responsible for final decisions. Protect confidential or regulated material.
3
Create validation checks for citations, extracted data, calculations, and interpretation.
Create validation checks for citations, extracted data, calculations, code, interpretations, and disclosures. Match the strength of the check to the risk of the output.
4
Run pilot comparisons against human work and revise the workflow.
Run pilot comparisons against human work and review the failure patterns, not only average quality. Revise the workflow before wider adoption.
5
Retain documentation and prepare the appropriate disclosure.
Retain prompts, source links, outputs, changes, and review notes and prepare an accurate disclosure. The record should preserve methodological transparency and accountability.

Worked example

A research team uses SciSpace across discovery, reading, analysis, and writing. Before starting, it classifies literature triage as low risk, reference verification as a mandatory checkpoint, interpretation of sensitive qualitative data as human-only, and formatting as automatable. Each stage has an approved input, output, reviewer, and record. When a generated summary overstates a source, the evidence checkpoint catches the error before it reaches the manuscript.

Practice prompt

Evaluate my proposed use of SciSpace across [project]. For each task, describe the benefit, methodological risk, evidence risk, data risk, validation procedure, human approver, reproducibility record, and disclosure requirement.

Key takeaway

The question is not whether AI is used, but whether its role is bounded, documented, source-grounded, and validated to the standard the research requires.