Blog Listing

Microsoft’s AI is learning to write code by itself, but not steal it!

Why this is interesting; This one needs some explanation. This technical AI feat comes out of a joint research between University of Cambridge and Microsoft Research, and yes, they developed a programming AI engine, named DeepCoder, implemented via a proprietary programing language developed specifically for this purpose. Without going to technical details (I will include the link to the actual originating research paper below) they made the AI engine create its own, most optimal, programming lines based on a given simple mathematical task. To augment its ability to write code, DeepCoder also has the capability to scour potential code for a solution that could work. It’s not scanning popular code repositories like StackOverflow or GitHub for a solution, although the authors would love for that to happen in the future. The task was given in the form of inputs and outputs and the engine then had to choose, from a set of simple commands, what sequence set of those commands that most efficiently could solve the task– this is interesting, not in the short term but it is a feat that will bring us closer to simplifying and even democratize how to “program” AI and write smart code. This goes far beyond just high-level programming and potentially to a much more meta way of solving problems. Think of it this way; instead of a person having to know how to build a house, by taking a degree in construction engineering, then the person could instead describe the desired architectural outcome of what this house should look like, and the AI engine would then construct the blue-print for this house out of all most relevant building blocks available, a house that would last and be feasible to build.

Why this is important; back to the hope of democratizing technology. Instead of that we all must learn to program code we can instead focus on being creative and more experimental and leave the tedious code implementation to a AI machine that would do the task in the most efficient way. This is not to say that if this technology becomes mainstream we would no longer need programmers, but it could certainly reduce the need and solve current bottle-neck problems with finding great coders. It is interesting times we live in!

Read this article (not the first article to report this, but one of the more sober ones).

This is great article that debunks much of the initial hype and related “fake news” that emerged from the original article.

The research paper (for the REALLY hardcore).