Study: AI hampered productivity of software developers, despite expectations it would boost efficiency

The results of this experiment, published in this month’s study, were a surprise for program developers assigned to the use of artificial intelligence – and the authors of the study, Joel Baker and Night Rush, technician employees in the evaluation of the Metr Research Organization model.

The researchers recruited 16 software developers, who have an average of five years, to conduct 246 tasks, each part of the projects they were already working on. For half of the tasks, developers are allowed to use artificial intelligence tools – most of them from the Cursor Pro or Claude 3.5/3.7 Sonnet – and for the other half, developers performed the tasks on their own.

The belief that artificial intelligence tools would make them more productive, program developers expect that technology will reduce the time of completing their mission by an average of 24 %. Instead, artificial intelligence resulted in their mission time to 19 % larger than technology use.

“I like to believe that my productivity did not suffer during the use of artificial intelligence for my tasks, it is unlikely that this has not helped me as much as I expected or maybe my effort hindered,” wrote Philip Burkhardt, a study participant.

Why Amnesty International slows down some workers

Where do the rabbits go from the road? In the midst of their own projects, it is possible that their work with a lot of additional context that artificial intelligence assistants did not have, which means that they had to modify their own agenda and problem -solving strategies in artificial intelligence outputs, which they also spent in time for time, according to the study.

“The majority of the developers who participated in the study noticed that even when they get the outputs of Amnesty International useful to them in general – and talking to the fact that artificial intelligence can often clean up from impressive work, or a kind of impressive work – the developers must spend a lot of time cleaning the resulting code to already agree to the project.” luck.

Other developers have been time to write claims for Chatbots chat or are waiting for Amnesty International to create results.

The results of the study contradicts the noble promises of artificial intelligence to transform the economy and the workforce, including a 15 % increase to GDP in the United States by 2035 and eventually an increase in productivity by 25 %.

But Rush and Baker have moved away from submitting comprehensive claims on the meaning of the study results for the future of artificial intelligence.

For one of them, the study sample was small and unspecified, including only a specialized group of people whose artificial intelligence tools were new. The authors said that the study measures technology at a specific moment of time, does not rule out the possibility of developing the tools of artificial intelligence in the future that would already help developers to enhance their workflow.

The purpose of the study, on a large scale, was pumping the brakes on the implementation of artificial intelligence at the workplace and in any other place, while recognizing more data on the actual effects of Amnesty International, should be known and accessible before making more decisions about its applications.

“Some of the decisions we are now making about developing and publishing these systems are likely to be a very high result,” Rush said. “If we’ll do that, let’s not only take the clear answer. Let’s do high -quality measurements.”

The broader effect of artificial intelligence on productivity

Economists have already confirmed that Metr research is in line with broader novels about artificial and productive intelligence. While artificial intelligence has begun to get rid of beginners positions, according to LinkedIn Anesh Raman, it may provide decreasing returns for skilled workers such as experienced software developers.

“For those people who already have 20 years, or in this specific example, five years of experience, their main task may not be to search for them and force them to start using these tools if they work well in the job with their current work methods.” luck.

Hamelum has similarly conducted research on the impact of artificial intelligence on productivity. It was found in a work study from May that among the 25,000 workers out of 7,000 workplaces in Denmark – a country with the absorption of similar artificial intelligence such as the United States – productivity improved by 3 % among employees who use tools.

Hamelum’s research, economist at the Massachusetts Institute of Technology and Emphasis on the Nobel Darwun Asimoglu Prize, supports that markets have been exaggerated in estimating productivity gains from artificial intelligence. Acemoglu argues only 4.6 % of the tasks within the American economy will be more efficient with artificial intelligence.

“In a hurry to automate everything, even the operations that should not be mechanism, companies will waste time and energy and will not get any of the advantages of promised productivity,” Acemoglu previously wrote to them luck. “The difficult fact is that obtaining productivity gains from any technology requires organizational amendment, a set of supplementary investments, and improvements in workers’ skills, through training and learning during work.”

Hummelum said that the productivity of software developers hindered this need to critical thinking when implementing the tools of artificial intelligence. While previous research on artificial intelligence productivity has been seen in self -reported data or specific and specific tasks, data on challenges from skilled workers using technology complicates the image.

“In the real world, many tasks are not as easy as writing in Chatgpt,” said Hamiltom. “Many experts have a lot of experience [they’ve] This accumulated is very useful, and we should not only ignore it and give up that valuable experience that has been assembled. “

He added: “I would just like to take this as a good reminder to be very careful about when these tools should be used.”