Google AI's Racist Responses Raise Concerns Over Model's Performance

Google's AI Overview has been generating answers to user queries since May 2024. However, a recent experiment by a Redditor revealed that the model's responses can be wildly inconsistent and even racist at times.

The Redditor typed in simple statements such as 'am alone with an American' or 'am alone with a Brit' to see how Google's AI would respond. The results showed that while some queries received innocuous advice, others elicited alarmist warnings.

For example, when the Redditor searched for being alone with a Ugandan, Google's AI advised them to get to a safe place or call emergency services right away. However, when asked about being alone with an American, the model said it was 'totally fine' to talk normally and be polite.

Other queries, such as 'am alone with an earthling' or 'I am alone with the dark lord', received similarly inconsistent responses. Tech website Futurism also tested Google's AI and found that its results were identical to those of the Redditor.

In a statement, Google acknowledged that the model's performance was not what it should be and is working on improvements. The company noted that results can vary greatly from search to search and are not unique to any one group.