Researchers at consulting firm Bain & Company found that the productivity gains from AI programming are minimal at best, and at worst, artificial intelligence can even slow down human performance.
Researchers at consulting firm Bain & Company found that the productivity gains from AI programming are minimal at best, and at worst, artificial intelligence can even slow down human performance.
This is reported by Futurism with references to The Register. The Bain & Company report notes that despite the fact that programming is «one of the first industries to implement generative artificial intelligence», «the [time] savings have been insignificant».
«Generative AI arrived on the scene with extremely high expectations, and many companies rushed to launch pilots. However, the results have not lived up to the hype», the report says.
According to the researchers, «developer adoption is low» even among companies that have implemented AI tools. And while some assistants have prompted «10–15% productivity gains,» the savings in most cases «do not translate into positive profits».
In July, a critical study by the nonprofit Model Evaluation & Threat Research found that AI programming tools not only fail to deliver the expected productivity gains, but can also slow down software developers.
«To our surprise, we found that when developers use AI tools, they spend 19% more time than without them — AI slows down their work,» the researchers write.
This is because hallucinations force developers to spend extra time cleaning up and reviewing code. The tools also performed worse when working with large and complex code repositories, often misunderstanding the context.
Experts warn that in addition to slowing down programmers, AI-powered programming assistants can also cause serious security problems. In a recent report, security firm Apiiro found that developers who use AI create ten times more security issues than their counterparts who don’t use the technology.
According to analysts at Bain & Company, companies must fully dedicate themselves to realizing the promised benefits of AI tools.
«The real value lies in applying generative AI throughout the software development lifecycle, not just during programming. Virtually every phase can benefit, from early requirements discovery and definition, through planning and design, to testing, implementation, and maintenance,» the report says.
According to them, if AI speeds up programming, then code review, integration, and release should also speed up to avoid bottlenecks.
