Google’s AI Search Feature Faces Criticism for Inaccurate Responses
Google’s new AI search feature, AI Overviews, has recently been criticized for providing users with inaccurate and sometimes dangerous responses to their queries. Despite these errors, Google maintains that the majority of AI Overviews offer high-quality information and that the company is actively working to improve its systems.
Summary of the Issue
- Examples of AI-generated errors include suggesting users mix glue with pizza cheese, eat rocks daily, and cook chicken at unsafe temperatures.
- Experts warn that AI language models like Google’s are prone to hallucinations and can reinforce biases in their training data.
Social Media Backlash
Social media platforms have been inundated with examples of AI Overviews’ blunders since its launch in the United States.
One instance involved the AI recommending mixing non-toxic glue with cheese to improve its adhesion to pizza, a suggestion originating from a joke Reddit post years ago.
Another error advised users to consume at least one small rock daily for digestive health, attributing the advice to “UC Berkeley geologists.”
Food Safety Concerns
AI Overviews also raised food safety concerns by inaccurately suggesting a cooking temperature of 38 degrees Celsius for chicken, well below the recommended 73.9 degrees Celsius to prevent foodborne illnesses.
In response to a query about the number of Muslim US presidents, the tool incorrectly claimed Barack Obama was secretly Muslim, despite his Christian faith.
Expert Warnings
Linguistics professor Emily M. Bender cautions that AI language models can reinforce biases and may make it harder to distinguish misinformation.
Experts highlight the potential for AI systems to hallucinate or fabricate information based on the extensive data they are trained on.
Google’s Response
Google asserts that the majority of AI-generated summaries offer high-quality information despite isolated errors.
The company states that it has addressed policy violations and is using feedback from these instances to enhance its systems.
By
By
By
By
By