Parents in Singapore should be made aware of a new genre of malevolent videos targeting young children on video-sharing sites such as YouTube.

The videos are easily found on children's channels on video-sharing sites.

They have familiar baby rhymes in the background and feature brightly coloured cartoon characters.

However, these characters act out violent and sexual scenes that exceed the tolerance level even of adults.

I was sickened and horrified when I saw one. Children may become traumatised or influenced by the violence.

But what troubled me most, as a researcher in the artificial intelligence industry, was how little we can do to rectify the problem, and how our current information infrastructure can be easily compromised by such well-engineered attacks.

In the tech industry, we use a combination of algorithmic recommendation and human feedback as a way to optimise knowledge sharing and discovery.

This good design has been exploited by video-makers.

They use harmless hashtags and cartoon characters to trick the classification algorithms on video-sharing sites into thinking these videos belong in the children's category.

Bots and humans then provide false reviews to fool the ranking algorithms. This results in the videos being promoted as recommended videos.

For the moment, we are without any legitimate weapon to counter this. But the fact is, even the best algorithms cannot match the darkness of some human minds.

Guo Hang