Amazon has become the latest company to launch an artificial intelligence-powered coding assistant, CodeWhisperer, which can be used by developers to automatically generate scripts or entire functions based on a short description. The announcement comes days after code repository Github launched a similar system, Copilot, which has been in the works for a year. Such tools are likely to become a “staple” for software engineers, but are unlikely to fully replace human coders, experts believe.
Unveiled at Amazon’s re:Mars conference, CodeWhisperer supports Java, JavaScript and Python. It will be available for users of the company’s cloud service, AWS, and Amazon says it trained the system on billions of lines of publicly available open source code and its own codebase, as well as documentation and code from public forums.
Vasi Philomin, Amazon’s VP for AI services told Tech Crunch the system builds on two other tools offered to AWS developers: CodeGuru, an AI code reviewer and performance profiler, and DevOps Guru, which helps programmers find operational issues.
CodeWhisperer is now available in preview as part of the AWS IDE Toolkit.
Github Copilot and the rise of AI-assisted programming
Earlier this week Github announced the general availability of Copilot, a system which it unveiled last year but which is now available for $10 per month.
Developers which have tested out the system have generally reacted positively, said Stephen O’Grady, analyst at RedMonk. While the quality of Copilot’s work is “challenging to assess objectively, the overwhelming sentiment to date has leaned towards being impressed – often extraordinarily so,” O’Grady said. “The most common reaction, in fact, is amazement.”
I might be done with work today because Copilot successfully generated half of the boilerplate I was planning to work on.
— Jaana Dogan ヤナ ドガン (@rakyll) June 10, 2022
While many have seemingly been impressed by Copilot, it has also attracted criticism. Github trained the system on open-source code from its repository, and members of its community have voiced concerns that the company is now monetising their work.
So GitHub copilot is gonna be paid
— Sara Vieira (@NikkitaFTW) June 22, 2022
Is it worth 10$ a month for the amount of time it saves me? Definitely
Definitely not paying for it tho, it’s using open source code and the code i write to feed it’s AI without giving any of this money back to the projects it uses pic.twitter.com/dufoVkkSnj
Questions of copyright could also arise from using the system, given that some of the code it is trained on is owned by third parties. Github’s license for using Copilot states: “GitHub Copilot is a tool, like a compiler or a pen. GitHub does not own the suggestions GitHub Copilot generates. The code you write with GitHub Copilot’s help belongs to you, and you are responsible for it.”
O’Grady writes: “In that line, GitHub is asserting two things. They’re saying explicitly that the output is not under some new GitHub license, it’s as if you wrote it yourself unassisted. And in doing so they are saying implicitly that the license of the code that was used to train the system is immaterial.”
Will AI replace programmers?
While CodeWhisperer and Copilot offer assistance to human developers, Google’s DeepMind claims to have taken AI programming one step further with AlphaCode, a system it says can generate fully-fledged software that compares favourably to the work of humans. Earlier this year it claimed its AI wrote better code than 54% of human participants in a coding contest.
Speaking to Tech Monitor earlier this year, Pete Schrammel, lecturer in computer science at the University of Sussex, said that despite the advances offered by the likes of AlphaCode, AI is unlikely to ever fully replace human developers.
“Software developers negotiate requirements in conversations with the stakeholders, implement a first version, then demonstrate it and discuss with the stakeholders what to improve for the next iteration,” he said. “In real-world software development, extracting the requirements from the stakeholders’ heads is the primary challenge.”
RedMonk’s O’Grady agrees, but said the conversation around wider automation of software development “speaks to the degree to which [Copilot] has elicited some extraordinary reactions from some otherwise hard to impress audiences.”
He added that, in the long term, “I think Copilot – and its current and future competitors – are likely to be a staple of the developer experience.”