WHY THIS MATTERS IN BRIEF
The art of programming is being automated and while it will take a while AI’s off to a good start.
While I’ve spoken about semi and fully autonomous Artificial Intelligence (AI) “Robo-Coders,” such as Facebook’s Transcoder, Google’s Bayou, Microsoft’s DeepCoder and OpenAI’s GPT-3, that can write their own code and compile programs with a few hundred lines of code from scratch – thereby automating many of the aspects of software development – this week code hosting service GitHub, owned by Microsoft, revealed that their AI assistant for programmers and citizen developers called Copilot is being used to write 30 percent of all new code on the platform.
Copilot, which is a distant cousin of Microsoft’s skunk works DeepCoder project, is an AI tool that acts like predictive text for coders. It is a programming assistant in GitHub’s visual studio code editor and gives users suggestions for lines of code or entire functions inside the editor.
Oege de Moor, VP of GitHub Next, the team responsible for Copilot, told Axios yesterday that feedback for the tool has been largely positive – with half of the developers who tried it continuing to use it.
“We hear a lot from our users that their coding practices have changed using Copilot,” he said. “Overall, they’re able to become much more productive in their coding.”
The tool is powered by the OpenAI Codex algorithm, a new AI system that was trained on a large dataset of public source code. OpenAI was founded in 2015 with the aim of ensuring that AI “benefits all of humanity,” and Microsoft invested over $1 Billion in the company a couple of years ago.
In a June interview with CNBC, co-founder and CTO Greg Brockman said OpenAI will release the Codex model for third-party developers to weave into their own applications later in the year, meaning that Copilot’s underlying technology won’t be exclusively for major investor Microsoft.
Microsoft in the meantime plans to continue working on developing “secure, trustworthy and ethical” AI to serve the public, while focusing on constructing new Azure AI supercomputing technologies.
“This is going to help bring this technology to a much broader audience,” de Moor said in the Axios interview, adding that it is part of GitHub’s effort to “make programming accessible to the next 200m developers.”
Naturally, AI assisted tools such as Copilot often prompt fears that the technology could one day replace human programmers by doing a job better but at the moment GitHub’s Copilot tool is designed to augment human workers rather than create original code. That said though by the end of the decade, if not earlier, it’s estimated that it, like it’s other Robo-Coder cousins, will be able to develop programs with thousands and even tens of thousands of lines of lines of code autonomously.
However, that said a New York University study in August found that approximately 40pc of the code produced by Copilot had cybersecurity flaws, so it’s handy that Microsoft also have an autonomous AI based debugger, and while software bugs are never great news it’s estimated that even the best human programmers have 70 bugs per 1,000 lines of code, according to data logging analytics company Coralogix.
So, as you can see, while programming continues its journey towards full automation there’s still a long way to go before humans are completely replaced by AI overlords and can relax on the beach while the machines take the strain.