The idea of a machine learning system designing other programs may seem far-fetched. But, tech companies such as Intel have been working to create such a system for a while.
According to a researcher at Intel, Justin Gottschlich, the company’s ultimate goal is to democratize software creation. That way, anyone can create programs as they see fit — using either code or natural language.
In a statement, Gottschlich, who is also the director of the machine programming research group at Intel, said:
“Building little apps for your phone, or things like that will help your everyday life—I think those are not too far off. I would like to see 8 billion people create software in whatever way is most natural for them.”
As audacious as the goal may seem, Intel has taken a substantial step toward it.
Last week, a team led by Gottschlich announced a new machine learning system that designs its own code. The researchers are calling it Machine Inferred Code Similarity, or MISIM for shorts.
Here’s how it works.
A Machine Learning System That Designs its Code
MISIM first analyzes a snippet of code to understand what the code is trying to accomplish. Next, it searches its repository to identify codes that can perform similar tasks.
After doing that, the machine learning system can then offer a faster and more productive code to get the job done.
MISIM is not the first system to compare code snippets. However, Intel’s team pointed out that their machine learning system’s accuracy was up to 40 times that of the nearest competitors.
Besides, MISIM can significantly streamline the time-consuming bug detection process.
Current machine learning systems are not good at spotting bugs unless they’ve been defined as such. As a result, they generate too many false positives.
MISIM, on the other hand, doesn’t rely on such a definition.
Instead, the system compares a new program with a code that has previously been established as correct. Then, it raises a flag when it detects significant differences — which could be errors.
Over time MISIM’s ability to translate simple English instructions into programming code will improve. When this happens, everyone will be able to create their software, says Gottschlich.
MIT and Georgia Tech also took part in the study.
Comments (0)
Most Recent