Check Before You Trust AI

Emily Yang

Human-Centred AI (HCAI) Specialist

Learn how to review AI outputs efficiently by checking what matters most and avoiding blind trust or unnecessary overchecking.

Data Science

Learn how to review AI outputs efficiently by checking what matters most and avoiding blind trust or unnecessary overchecking.

Subscribe to watch

Access this and all of the content on our platform by signing up for a 7-day free trial.

Check Before You Trust AI

4 mins 47 secs

Key learning objectives:

Learn how to distinguish low-risk from high-risk outputs
Identify what must be verified manually
Learn how to spot warning signs in AI outputs
Review efficiently using judgement

Overview:

Effective review is about applying the right level of scrutiny to the right outputs. Low-risk tasks may need only a quick sense-check, while higher-risk outputs require closer verification. By knowing what to check, spotting warning signs, and using judgement proportionate to the stakes, people can stay efficient without compromising quality or trust.

Subscribe to watch

Access this and all of the content on our platform by signing up for a 7-day free trial.

Summary

Why do people review AI badly?
Many people fall into two unhelpful extremes. Some trust outputs too quickly because the response sounds polished and confident. Others check every line with the same level of effort, which removes the time-saving benefit of using AI. Strong review means avoiding both blind trust and unnecessary overchecking.

What should always trigger scrutiny?
Certain content deserves closer attention because errors carry consequences. This includes facts, figures, names, dates, quotations, promises, recommendations, calculations, and anything linked to legal, commercial, financial, compliance, or reputational impact. Where the cost of being wrong is higher, the quality of review must rise with it.

What are warning signs?
Common warning signs include confident claims without evidence, vague sources, invented details, suspiciously precise numbers, missing nuance, or outputs that present uncertain issues as settled facts. Content that sounds smooth and authoritative is not always reliable, so tone should never be mistaken for accuracy.

What does efficient review look like?
Efficient review matches effort to risk. A routine internal draft may only need a quick read for logic and tone, while a client-facing recommendation or data-heavy summary may need line-by-line checks. The aim is not to review everything equally, but to focus attention where it matters most.

What never changes?
Responsibility remains with the person using the tool. AI can assist with speed and drafting, but accountability for what is shared, approved, or acted upon still sits with the human decision-maker.

Subscribe to watch

Access this and all of the content on our platform by signing up for a 7-day free trial.

Emily Yang

Emily Yang leads Human-Centred AI and Innovation at a global financial institution and serves on the organisation’s AI Safety and Governance committees. Her work focuses on advancing responsible and trustworthy AI systems that balance innovation with accountability. She is among the first practitioners in the industry to apply Human-Centred AI at scale. With over a decade of experience in human-computer interaction and user experience, Emily has held roles across tech startups, corporate venture builders, and major technology companies. Her journey into AI began with studies in biochemistry and neuroscience, followed by a research master’s in HCI and natural language technologies, during which she published work on perceived empathy and emotional intelligence in virtual agents.

There are no available Videos from "Emily Yang"