Why Hong Kong’s bilingualism is uniquely indispensable in the AI era

Disturbingly, the system cloaked Chinese content in English citations, creating a deceptive authenticity that made the hallucinations difficult to verify. AI claims to bridge all languages, but gaps remain. Knowing both sides is the only way to see them.

This systemic vulnerability is inherent in how large language models handle cross-lingual alignment. When bridging asymmetric data pools – the vast English web and the structurally distinct Chinese digital ecosystem – these models suffer from epistemic asymmetry. Instead of true synthesis, the system often takes localised, unverified Chinese content and cross-contaminates it with invented English academic sources.

For instance, if a Chinese art blog mistakenly claims a specific local painting inspired Vincent van Gogh’s The Starry Night, the AI doesn’t just translate the claim – it might fabricate a non-existent Oxford University Press citation to back it up in English. This synthetic authority, delivered in flawless academic prose, hides real gaps between the two digital ecosystems, making the fabrication almost impossible to detect.

This separates monolingual and bilingual users. When an AI fabricates citations, a monolingual user is trapped within a single semantic loop, unable to test the boundaries of the model’s bias. In my case, bilingualism broke this loop through cross-examination, manually tracing the AI’s citations back to their linguistic origins and verifying them across both language webs until the illusion collapsed.

In a world where algorithms blend and flatten multiple information streams, independently checking both sides of the digital curtain is a critical step: synthesised digital authority must not be taken at face value.