Rise of the machines(sort of)
Shortly before starting at Flatiron School in September, I stumbled across an article about a collaboration between Intel, MIT, and the Georgia Institute of Technology. They are working towards the goal of machine programming, meaning that they want to create AI that can write code.
As someone who was equally excited and nervous about starting coding bootcamp, I was immediately stressed when I saw the misleading title of the article’s thumbnail. I thought this new tech meant the end of my career before my career even began. After reading the article, I knew that my panic was unnecessary for two reasons.
- What the Intel team has so far is a tool to assist developers, not put them out of jobs.
- What they have right now could be extremely useful, but is incomplete and far from their lofty goal of full-on machine programming.
To the first point, the tool that the Intel team is developing is called machine inferred code similarity(MISIM). One of the main factors that led Intel to invest in MISIM is a global shortage of capable developers. The idea is that MISIM will do a lot of things to make developers more productive, not make them irrelevant. It’s purpose is to help make suggestions as you write code and help with the debugging process. MISIM does this by analyzing pieces of code that serve the same or similar purpose across many programming languages, and then making inferences based off of what it sees to give the developer suggestions. This really isn’t all that different from tools that we use every day in our code editors, such as Ruby Solargraph in VScode. In Ruby on Rails, rails new or rails g will make a ton of files for you to save you time. MISIM isn’t the beginning of the end any more than Rails was.
Secondly, MISIM has not been released. Aside from that, Intel has described it as being an important building block for their machine programming goals. They have also acknowledged that they are far from those goals.
I’m relieved that this isn’t the beginning of the end for developers, but I would also like to stress that I am not trying to diminish the impressiveness of Intel’s accomplishment. As I mentioned earlier, similar systems exist. The thing is, Intel is claiming that MISIM is up to 40 times as accurate as similar systems that are currently available. This is due to MISIM replacing a compiler with a new system Intel has named context-aware semantic structure(CASS). According to Intel, CASS uses their neural networks in place of a compiler to give every piece of code a similarity score based on their intended function. Although MISIM and CASS are only doing something faster, not completely new, 40 times is a an enormous increase in accuracy. It will be exciting to see the impact that this technology has across Intel’s architectures over the next few years, as they could certainly use a good performance boost after years of unexciting products. MISIM genuinely seems like a good first step in teaching a computer how to code, and I hope we keep seeing regular progress in its development.
I am extremely happy that I actually opened and read this story when I saw it, and didn’t take the sensationalized headline that was something along the lines of “Intel has a thing that makes software engineers obsolete.” I shouldn’t be shocked that Intel is far from completing this project. It would be ambitious for anyone, but even more so for a company that has been repackaging the same desktop CPU lineup for years on end. 😎
