Deloitte has again come under scrutiny for citing sources hallucinated by artificial intelligence (AI) in a report for the Canadian government.
The report commissioned by the government of Newfoundland and Labrador province contained errors generated by AI, including fake citations, The Independent first reported.
Deloitte’s faulty analysis, which cost the government $1.6 million, aimed to help address the province's health authority’s staffing crisis.
Fabricated citations were used to support claims related to recruitment strategies and the impacts of the COVID-19 pandemic on healthcare workers, among others.
The Department of Health and Community Services told the Canadian media that it “directed Deloitte to confirm the accuracy of the citations and literature review,” and that the firm acknowledged that four citations were incorrect.
Deloitte says it “firmly stands behind the recommendations” laid out in the report, but is making corrections to a small number of citations. The company claims AI was used only to support the research and not to write the report.
This is the second time this year that the British accounting firm has been accused of producing a report containing errors due to the use of AI.
Deloitte’s report for the Australian government’s Department of Employment and Workplace Relations, released in July, included faulty references and citations, some of which were found to be AI-generated hallucinations.
The company later confirmed that it had used AI to produce the report, but insisted that the corrections did not alter the report’s main findings or recommendations.
Deloitte will repay the Australian government a portion of its $440,000 fee, although the specific amount has not been disclosed.
The repeated Deloitte blunder highlights the risks of relying on AI without human oversight.
The hallucination rate for commonly used large language models (LLMs) can be as high as 45%, according to a recent study by AIMultiple.
Sam Altman, CEO of OpenAI, which owns ChatGPT, surprised users earlier this year by saying they shouldn’t trust the chatbot too much because it hallucinates.
Unlock exclusive Cybernews content on YouTube
Your email address will not be published. Required fields are markedmarked