Google’s much-anticipated AI Overview feature, launched in beta last year, is facing a rocky rollout. The feature, designed to summarize search results using AI, has generated a firestorm of criticism for producing nonsensical and even dangerous outputs.
Read: You can now hum to YouTube Music to search a song
Social media is flooded with examples of AI Overview suggesting users glue their pizza or eat rocks. Google is scrambling to contain the damage, manually disabling the feature for specific searches as these memes gain traction.
This chaotic launch raises questions about Google’s development process. Despite a year of testing and over a billion queries served, AI Overview seems to be struggling with basic accuracy.
Critics point to an overemphasis on optimization. Google boasts an 80% reduction in AI answer delivery costs, achieved through “hardware, engineering and technical breakthroughs.” Experts like Gary Marcus, an AI researcher, suggest this optimization may have come at the expense of quality.
“Google has gone from cutting-edge to meme-worthy,” said one anonymous AI founder. Google maintains that AI Overview delivers “high-quality information” and blames user-generated queries and potential manipulation for the bad examples circulating online. However, the company is taking action, removing AI Overviews for specific searches and working on broader improvements.
The pressure to compete in AI is intense. Microsoft’s Bing made a splashy AI push years ago, OpenAI is building a rival search engine, and a new AI startup is already a billion-dollar company. This pressure, coupled with a younger generation prioritizing user experience, likely fuelled Google’s rush to launch.
The situation is reminiscent of Meta’s failed Galactica AI, which also dispensed dangerous advice (like suggesting users eat glass) before being shut down.
Google envisions AI Overviews as a cornerstone of its future search experience, including complex query reasoning and AI-generated results pages. But with the current iteration struggling with basic tasks, Google’s reputation for reliable search is at stake. As Marcus points out, these large language models lack the ability to independently verify their own outputs, a critical flaw hindering the field.
Google finds itself in a precarious position. While ambitious plans exist for AI Overviews, the company desperately needs to address the present issues and restore user trust.
