BERLIN: Experienced and productive employees who use artificial intelligence (AI) on the job often end up diminishing the quality of their work, according to researchers at Stanford University.
At the same time, less capable or experienced staff seem to be more productive and faster at getting things done if using AI help, the team found, in a study published by Oxford University Press's Quarterly Journal of Economics.
While some workers saw an increase in the number of cases per hour deemed "resolved" when they leaned on AI, the more proficient staff saw a decrease in the quality of their output. Irate or frustrated customers were less likely to complain further or seek to speak to a manager, however, when confronted by AI instead of a person.
Overall, the effects of AI on productivity "vary significantly," the Stanford researchers said, after looking into how almost 5,200 customer service agents at an unnamed Fortune 500-listed company made use of the technology, which has spurred a mix of admiration and fear since being made available for general use in late 2022 with the launch of OpenAI’s ChatGPT.
One of the main concerns about AI is that it could be used to replace people in many jobs - and not just computer-based ones. In 2024 investment bank Morgan Stanley said that merging AI with "humanoid" robots could see millions of farm and factory workers put out of work by mid-century.
Last month a University of Oxford-led team said that AI bots such as ChatGPT could see less need for people doing "highly repetitive" writing jobs but could lead to new roles for "complementary" skills to supplement AI.
"Despite fears of mass job losses, this study suggests a more balanced reality," said Maria del Rio-Chanona of University College London.
AI bots have been criticised for so-called "hallucinations" – industry jargon for their penchant for churning out inaccurate responses, even gibberish, depending on the query submitted.
But proponents have said that AI is advancing at such a clip that it could soon outdo humans at most tasks - even as soon as 2027, according to a prediction made in January by Anthropic boss Dario Amodei.
Within days of Amodei's warning, the industry was rocked by the sudden emergence of DeepSeek’s R1, a Chinese-made variant that was at first described as a potential great leap forward for AI as it seemed to match the likes of ChatGPT, despite allegedly costing much less money to develop and maintain. – dpa
