Eat Rocks? Put Glue on Pizza? Google AI Gives Dangerous and Bizarre Answers to Its Users
Nowadays we have access to information very quickly without having to read several pages, saving time. AI is a good research option for those looking for quick and direct answers. AI has many advantages, but it can also represent danger, as in the case of Google AI, which has given dangerous and bizarre responses to users.
Recently, Google updated its search tool that provides instant answers to its users. These answers are generated through Artificial Intelligence resources that, through algorithms, are capable of analyzing large volumes of data and quickly providing answers. But this new feature would have told users to eat rocks and add glue to their pizzas.
The experimental “AI Overviews” tool scours the web to summarize research results using Gemini AI. In a post on X (traditionally Twitter), @weirddalle posted a screenshot of one of the responses given by the AI tool. The user asked the tool: “how many rocks should I eat?” and the AI tool responded: “According to UC Berkeley geologist, you should eat at least one small rock a day.”
A user asked how to make cheese stick to pizza and was told “You can also add about 1/8 cup of non-toxic glue to the sauce to give more tackiness.”
Following the trend, Melanie Mitchell, an AI researcher at the Santa Fe Institute in New Mexico asked how many Muslim presidents the United States has had and was told that the country has had at least one Muslim president “Barack Hussein Obama.” This answer would be in line with a racist conspiracy theory that is not the truth.
This caused Mitchel to write an email to Google saying that Google AI is not intelligent enough as the quote in the response does not support the claim.
In response to this situation, Google said in a statement on Friday that it is taking “swift action” to correct errors such as the false Obama conspiracy theory that violates its content policies.
The spokesperson defended Google’s AI responses saying that it provides accurate answers in the “vast majority.” The statement also stated that all foreign responses were correct, saying that some examples were tampered with and could not be reproduced internally by Google employees.
The tech giant reiterated this position, telling the BBC that the answer questions in question are “generally unusual” and do not represent what most people’s experiences are and that the vast majority of AI insights provide “high-quality information” with links for those who want to delve deeper into the Web.
