In the age of digital convenience, the promise of Artificial Intelligence (AI) was simple: to sift through mountains of data for us, providing a clear and concise snapshot of reality. However, a new investigation by the British consumer organization Which? has brought to light a disturbing side effect of this automation. Tripadvisor's AI tool, designed to summarize thousands of user reviews into short, digestible paragraphs, appears to suffer from a dangerous brand of "optimism," consistently overlooking grave safety and health concerns.
The Embellishment of Risk
The Which? investigation focused on hotels that had accumulated numerous negative reviews regarding issues such as food poisoning, bedbug infestations, and even sexual harassment. According to the findings, Tripadvisor’s AI summaries frequently ignored these critical warnings, opting instead to highlight secondary positive attributes like location or staff friendliness. In one particularly egregious case, a hotel with repeated allegations of sexual harassment by staff members was summarized by the AI as a "popular choice with friendly service."
The root of the problem lies in how Large Language Models (LLMs) process data. Frequently, these algorithms are programmed to identify the "prevailing sentiment" or to provide a balanced overview. When a hotel has 500 reviews praising the view and 10 reviews alleging a serious assault, the AI may statistically downgrade those 10 reviews as "outliers," despite the fact that for a human reader, those 10 reviews are the most critical factor in a booking decision.
Platform Accountability and AI Ethics
The Tripadvisor case reignites the debate over the responsibility of tech giants. Traditionally, these platforms acted as "hosts" for third-party content, distancing themselves from legal liability for what users wrote. However, with the implementation of AI summaries, the platform ceases to be a passive intermediary and becomes a publisher. The summary generated by the AI is a new product—a synthesized "opinion" of the platform itself—and as such, its legal and ethical responsibility increases exponentially.
"Artificial intelligence cannot be a blind word-counter. It must have embedded safety filters that recognize high-risk keywords and prioritize them over general sentiment," AI ethics experts argue.
In a global context, and particularly for tourism-dependent economies like Greece, such phenomena have a dual impact. While some hoteliers might feel "protected" by an AI that glosses over failures, the long-term erosion of consumer trust could be catastrophic for the industry. Transparency is the only currency that maintains its value in the global hospitality market.
Technical Flaws and the Need for a Human-in-the-Loop
Why does the AI fail? The answer lies in its lack of conceptual understanding. AI does not "understand" what danger means; it only understands word correlations. If the majority of reviews utilize positive adjectives, the algorithm will likely produce a positive output. What is required is a "Human-in-the-loop" approach, where summaries involving critical safety issues are vetted by human moderators or specialized algorithms that prioritize reports of crime or health hazards.
Tripadvisor responded to the criticisms by stating that they continue to refine their tools and that users should always read full reviews. However, the very presence of a summary at the top of the page encourages cognitive ease, leading users to skip the details. The Which? investigation serves as a loud reminder that technology, no matter how impressive, remains a tool that requires rigorous oversight and human judgment.
