Trace every claim
An output should point to a verifiable source; a plausible citation is not a verified citation.
GLOBAL AUTISM RESEARCH CENTER
Responsible AI in Research
AI-assisted. Evidence-grounded. Human-governed. Traceability and scientific review are central to responsible use.
GARC / RESEARCH
Operating principles
- Human oversight and scientific review
- Evidence traceability and source transparency
- Privacy and data minimization
- Bias awareness and research integrity
- No fabricated citations or unverified clinical claims
Current status
Ask GARC currently routes to site search. An evidence-grounded AI system is being developed and is not presented as live.
IN DEPTH
A bounded research assistant
AI can support knowledge work only within clear source, privacy and human-review controls.
Protect data
Use minimum necessary information and approved tools; do not put sensitive participant data into unapproved workflows.
Human decision
A qualified reviewer judges methods, evidence and suitability before an output is used or published.
PRACTICAL DETAIL
From experiment to accountable use
A research workflow should specify what AI does, what humans review and how errors are handled.
Define the task
Separate literature discovery, drafting and classification from decisions that require scientific or professional judgment.
Evaluate
Use a documented sample to check source fidelity, false citations, omissions, accessibility and performance across relevant groups.
Retain accountability
Name the human reviewer, keep versioned records and provide a route to correct or withdraw an erroneous output.