Home News Deloitte to refund part of the money to the Australian government for...

Deloitte to refund part of the money to the Australian government for a report containing AI hallucinations

Deloitte

The Australian branch of the consulting firm Deloitte will refund part of the total AUD 440,000 (AUD; CZK 6.1 million) paid for a report that contained errors caused by so-called artificial intelligence (AI) hallucinations. These included fabricated quotes from a federal court ruling and references to non-existent scientific papers, the AP reported.

AI is changing the consulting services market

The error comes at a time when the advancing expansion of artificial intelligence systems is putting pressure on consulting firms, according to the Semafor website. According to the website, large consulting firms have laid off approximately 50,000 people since the beginning of 2023 in connection with the slowdown in activity after the pandemic, the deployment of AI, and the reduction of government spending.

The report by the company providing financial services to the Department of Employment and Labor Relations was originally published on the department’s website in July. The revised version was published by the department at the end of last week after Chris Rudge of the University of Sydney alerted the media that the report contained fabricated citations and references.

Australian ministry uncovers errors in Deloitte report

Deloitte subsequently reviewed the 237-page report and confirmed that some footnotes and references were incorrect, the ministry said on Tuesday. “Deloitte has agreed to refund the final payment under the contract,” the ministry added. The exact amount will be disclosed after the refund.

The Deloitte report looked at how the Australian department’s IT systems use automated sanctions in the Australian social security system. The department said the substance of the report remained unchanged and its recommendations remained the same. In the context of AI and language models, hallucinations are a term used to describe generated responses that contain false or misleading information. They can also be the result of insufficient, outdated, or biased training data.

Source: Reuters

LEAVE A REPLY

Please enter your comment!
Please enter your name here