Expert Session

Researcher-Led, AI-Assisted: A Responsible Academic Research Workflow

Speaker

Dr. Ethan J Gantana

Quick answer

In SciSpace, a responsible AI research workflow assigns each stage to one of four categories: automate, assist, human checkpoint, or human-only. Define approved inputs, source boundaries, review criteria, privacy rules, and disclosure requirements before starting, and preserve an evidence trail for consequential outputs.

About this session

Dr. Ethan J. Gantana discusses utilizing the SciSpace platform to synthesize literature reviews and draft research text [L5-L16]. The session emphasizes maintaining a human-led workflow when identifying research gaps and processing complex medical information to diagnose and prognosticate hematymphoid tumors amidst evolving molecular data [L5-L16].

The most reliable AI-assisted research process is not fully autonomous: it is researcher-led, with automation applied where it saves time and human judgment retained where decisions affect validity. This session offers a practical framework for deciding what SciSpace may do, what requires a checkpoint, and how to document the resulting workflow.

What this tutorial covers

Map a research project into tasks that can be automated, assisted, or kept fully human-led.
Define source, scope, methodology, and output constraints before asking AI to act.
Use checkpoints for inclusion decisions, analytical choices, interpretation, and final writing.
Keep a traceable record of prompts, sources, revisions, verification, and disclosure.

Key concepts explained

Human-in-the-loop design: The researcher remains the decision-maker at consequential stages.
Task–risk matching: Summarization, extraction, and formatting carry different risks from interpretation or causal claims.
Provenance: Outputs should show which sources, files, or data support each conclusion.
Accountability: AI assistance does not transfer responsibility for accuracy, ethics, or originality.

How to try it in SciSpace

1
List the stages of your project from question formation to publication.
List every project stage from question formation to publication and the consequence of an error at that stage. This makes high-risk decisions visible before automation is added.
2
Classify each stage as automate, assist, checkpoint, or human-only based on methodological and ethical risk.
Classify each stage as automate, assist, human checkpoint, or human-only. Record the reason, approved data, and person responsible for the final decision.
3
Create prompts that specify sources, exclusions, deliverables, uncertainty, and approval points.
Write prompts that specify allowed sources, exclusions, deliverables, uncertainty labels, and approval points. Instruct the system not to invent missing evidence.
4
Verify sampled outputs early, then expand only after the workflow performs reliably.
Pilot the workflow on a manageable sample and compare the output with human review. Revise the process before expanding to a larger paper set, dataset, or manuscript.
5
Retain a research log and prepare a truthful disclosure for the final output.
Retain prompts, source links, outputs, changes, verification notes, and a truthful disclosure. The record should show what the system did and what the researcher decided.

Worked example

A project team maps its workflow before using SciSpace. Literature discovery is classified as assisted, reference verification as a mandatory human checkpoint, statistical model choice as human-only, and formatting as automatable. For each stage, the team records allowed data, expected output, reviewer, and evidence requirement. When an AI-generated synthesis conflicts with a source paper, the checkpoint stops the workflow until the discrepancy is resolved.

Practice prompt

Design a researcher-led AI workflow for [project]. For every stage, specify the human decision, the task SciSpace may assist with, approved sources, validation method, checkpoint, output, risk, and documentation required.

Key takeaway

Responsible research automation is a designed collaboration: use AI for scale and structure while keeping scientific decisions, interpretation, and accountability with people.