This webinar is designed for anyone who has to decide whether to let automation into their review and wants a defensible basis for that decision. We will discuss an approach we use to decide which tasks are safe to automate and which need a human final call, using two contrasting cases from our own work: screening, where a miss is costly and often invisible, and extraction, where every value has a source that can be checked against. We will place both against the RAISE (Responsible AI in Evidence Synthesis) guidance, and discuss why the answer to ‘when to trust the machine’ comes down to what an error costs and who is positioned to catch it.
By the end of this session, participants will learn how to:
- Apply a simple test for whether an AI suggestion is safe to accept
- Interpret AI recall and precision claims, and know which matters for which review task
- Understand the validation methodology behind the automation features in Covidence, including what we chose not to automate and why
- Identify a short list of questions to ask any AI tool provider, aligned to RAISE 3
- Recognize what to record during a review so that AI use is reportable and defensible at peer review
Teachers: Julie Brown, Principal Systematic Consultant Reviewer, & Razia Aliani, Senior Consultant Systematic Reviewer, Covidence Support
If you can't attend live, register anyway. We'll send you the recording within 48 hours of the session.
