Google blames users for highly inaccurate ‘AI Overview’ output.
Search monolith Google’s recently released artificial intelligence feature ‘AI Overview’ is providing inaccurate and dangerous summaries of user searches, and Google appears to have no real solution to this problem.
As of this writing, Google has disabled certain queries for its “AI Overview” feature after it was widely reported that the system was producing erroneous and potentially harmful output.
Reports of user queries asking search engines how to store cheese on pizza began spreading across social and news media communities, with the AI system reportedly responding with text telling users to use glue. In another batch of apparent confusion, the AI system reportedly told users that at least two dogs owned the hotel and pointed to a non-existent dog statue as evidence.
While many of the supposedly inaccurate results may seem humorous or harmless, the main concern is that the consumer-facing models that generate “AI Overview” content produce inaccurate and accurate results based on the same external trust.
And so far, according to Google representative Meghann Farnsworth, who spoke to the Verge via email, the company has taken on the task of removing queries that lead to inaccurate results from its systems as they arise. Essentially, Google appears to be playing metaphorical whack-a-mole with the AI problem.
What makes matters even more confusing is that Google appears to be shifting responsibility for the problem onto the humans generating the queries.
According to Farnsworth:
“Many of the cases we have seen have been uncommon queries, and we have also seen cases where they have been manipulated or cannot be reproduced.”
At the moment, it’s unclear how users should avoid “uncommon queries,” and as is common with large language models, Google’s AI systems tend to output different answers to the same question when queried multiple times.
Cointelegraph reached out to Google for further comment but did not receive an immediate response.
Although Google’s AI system still appears to need some development before it can meet its challenges, Elon Musk, founder of rival AI company xAI, believes the machines will surpass human capabilities before the end of 2025.
As Cointelegraph recently reported, Musk recently told conference attendees at the VivaTech 2024 event in Paris that he believes xAI can catch up to OpenAI and DeepMind Google by the end of 2024.
Related: Political correctness in AI systems is the biggest concern: Elon Musk
