Artificial Intelligence: AI that Can Write Code

Last year software developers were astonished when word got out that OpenAI’s GPT-3 AI-based text generator was able to write code based on natural English text prompts.

John Carmack, programmer and game developer, tweeted that “I used to say that AI research seemed to have an odd blind spot towards automation of programming work, and I suspected a subconscious self-preservation bias. The recent, almost accidental, discovery that GPT-3 can sort of write code does generate a slight shiver.”

A 2013 study from Oxford estimates that 47 percent of all US jobs will be automated by 2030. That number seems a little high and automation growth a little too fast, but a lot can change in a decade — who knows? In 2019, McKinsey estimated that it will take longer — by 2055 50 percent of jobs will be automated.

Will code developers be safe from the on-coming take-over by AI and automation? To anyone who has done code development, it seems hard to imagine that the job can be automated. In the short-term what is most likely to happen is that AI-tools will accelerate the work of developers. Auto-suggestions for which APIs to use, code completion, and assistance debugging are all tasks that could speed-up the job of developers. Others hope that code translation, the rewriting of software from one computer language to another, is the initial kind of problem that AI can be trained for and has a good chance to be able to excel at.

Kevin Scott, chief technology officer at Microsoft, said that AI and auto-suggestion “can save the developer from having to dive through a whole bunch of documentation to get a tool to do a thing that I know it’s capable of doing, and that is so good for productivity. I can’t even tell you the number of hours I’ve wasted trying to figure out the right way to do a relatively prosaic thing, just navigating the complexity of these tools.”

Boris Paskalev, head of product at Snyk, said that “since there are no formal rules for software development, [programming] is an art that requires a complete understanding of code and a developer’s intentions to produce something that works as expected without bugs. As far as we’ve come in using machine learning and neural networks for code, we’re still only in the ‘invention of the wheel’ phase … machine learning is already proving to be very useful for code, but only after it goes through a semantic machine learning-representation of the code: making sure all semantic facts, variables, transitions, and logical interrelations are clearly represented and considered by the learning model.”