Many people, such as journalists, are understandably anxious about what generative artificial intelligence means for the future of their profession. It doesn't help that experts' predictions on the issue offer a confusing cocktail of wide-eyed excitement, sharp skepticism, and dystopian despair.
But some workers are already living in one potential version of the generative AI future: computer programmers.
“Developers have arrived in the age of AI,” says Thomas Dohmke, CEO of GitHub. “The only question is how quickly you jump on board. Or are you going to get stuck in the past and end up on the wrong side of the 'productivity polarity'?”
In June 2021, GitHub launched a preview version of a programming assistance feature called Copilot. It uses generative AI to suggest ways to complete large pieces of code as soon as you start typing. Copilot is currently a huge hit as a paid tool. GitHub owner Microsoft said in its latest quarterly results that it now has 1.3 million paid Copilot accounts, a 30% increase over the previous quarter, and that 50,000 different companies use the software. I said that.
Domke said Copilot's latest usage data shows that nearly half of the code written by users was generated by AI. At the same time, he argues, there is little indication that these AI programs can operate without human supervision. “After using these tools, there was a clear consensus from the developer community that they should be co-pilots for paired programmers,” he says Dohmke.
The power of Copilot lies in the way it abstracts complexity for programmers trying to address problems, Dohmke says. He likens this to the way modern programming languages hide tricky details that programmers had to discuss in earlier low-level languages. Dohmke added that younger programmers in particular have embraced his Copilot, and it seems to be particularly helpful in solving beginner coding problems. (This makes sense given that Copilot learns from reams of code posted online, where solutions to beginner problems are better than examples of arcane and rare coding techniques.) (There are also many.)
“We're seeing an evolution in software development,” Dohmke says.
None of this means that developer labor demands won't change with AI. According to his GitHub research, conducted in collaboration with MIT, Copilot allows programmers faced with relatively simple tasks to complete their work 55% faster on average. Ta. This productivity increase suggests that a company can accomplish the same work with fewer programmers, while the company can spend the savings on more labor on other projects.
Even those who don't write code can benefit from these discoveries and the rapid adoption of Copilot. Microsoft is developing AI Copilots, which the company calls AI Copilots, designed to help you compose emails, create spreadsheets, or analyze documents for Office software. Modern Windows PCs also introduced his Copilot key, marking the first major change to keyboard buttons in decades. Competitors such as Google are also building similar tools. GitHub's success may be helping fuel this push to bring AI workplace assistants to everyone.
“There is ample empirical evidence and data behind GitHub Copilot and its associated productivity statistics,” Microsoft CEO Satya Nadella said during the company's latest earnings call. He added that Microsoft's other CoPilot users will see similar benefits. Microsoft has created a site where you can try Copilot for Windows. Honestly, I'm not sure how similar the tasks I want to perform on Windows are to the tasks I perform on GitHub Copilot, which uses code to accomplish a clear goal.
Tools like GitHub Copilot have potential side effects beyond job displacement. For example, increased reliance on automation can introduce more errors into your code. One recent study claims to have found evidence of such a trend, but Domke said it only reported an increase in common mistakes since the introduction of Copilot, and that the AI helper was not responsible for an increase in errors. It states that there is no direct evidence that it is causing this. While this is true, I think it's natural for an inexperienced programmer to worry that relying on his AI's help will cause him to miss errors or reduce the overall quality of his code due to autocomplete. I will.
Given Copilot's popularity, it won't be long before we have more data on this issue. Those of us in other jobs may soon find out if we can get the same productivity gains as programmers, and the corporate havoc that comes with programmers.