Rice University computer scientists developed BAYOU, an AI that can write its software code.
There have been efforts in the past to create software that can write its code; however, programmers need to write as much or more code to tell the programme what types of applications they want it to code as they would if they only coded the app itself. BAYOU has changed all of that.
The AI examines all of the code on GitHub and utilises it to write its code. The AI study all of the code and then associates an “intent” behind each using a method known as natural sketch learning.
When a human requests that BAYOU construct an app, BAYOU associates the content it learnt from Github codes with the user’s request and begins writing the app it believes the user wants.
BAYOU is a deep learning tool that functions similarly to a search engine for coding: tell it what kind of programme you want to develop using a few keywords, and it will spit out java code that will achieve what you’re looking for, based on its best estimate, according to Futurism.
The programme was created by a team of computer scientists from Rice University, funded equally by the military and Google.
If the code BAYOU read contained any information about how the code works, BAYOU learned what those programmes mean to do and how they work. This contextual knowledge enables the AI to create software with only a few keywords and basic information about the programmer’s preferences.
However, it is not entirely self-sufficient. As per the researcher, BAYOU creates “sketches” of a programme relevant to what a programmer is trying to write. There is still a lot of work to be done in putting these sketches together into a bigger picture, and they may need tailoring to fit the specific project.
Even though the technology is still in its infancy, this is a significant step forward in the trail for an AI coder, a long-standing aim for computer science experts.