Forum: Use AI to test online systems that influence the young
Sign up now: Get ST's newsletters delivered to your inbox
Recent expert recommendations on protecting young people from social media harms rightly look beyond screen time towards what youths actually experience online.
IMDA already uses test accounts and mystery-shopping exercises to assess designated social media services. Could this capability now be extended to recommendation algorithms?
Synthetic youth accounts could test what platforms progressively recommend. For example, does an ordinary interest in fitness remain about healthy exercise, or eventually lead towards extreme dieting and harmful body-image content?
Artificial intelligence could make such testing possible at scale. Just as technology is used to detect scams or malware, AI could analyse recommendation journeys and flag potentially harmful patterns or significant changes for human review.
Such testing could create a continuing audit trail of what was recommended, when it appeared, and what interactions preceded it. When platforms make improvements, regulators could retest them and determine whether the safeguards actually worked.
The findings could also be translated into simple public youth-safety ratings covering areas such as recommendation safety, addictive design, harmful content and stranger contact.
Regulators could additionally identify recurrent high-risk recommendation pathways, domains or creator clusters without publishing links that might amplify harmful material.
Independent testing should be supported, where appropriate, by access to platform records so that what test accounts actually experienced can be reconciled against what platforms recorded.
This would create a chain from detection and evidence, through human review and remediation, to retesting, public transparency and, where necessary, enforcement.
The objective is not to monitor our children. It is to independently test the systems that increasingly influence them.
If technology helps platforms learn what keeps young people engaged, we should also use technology to test whether that engagement remains safe.
Chong Ryh Huei
