How Microsoft could improve Copilot and ease open source controversy

GitHub Copilot has been the subject of some controversy since Microsoft announced it in the summer of 2021. Microsoft was recently sued by programmer and lawyer Matthew Butterick, who alleged that GitHub’s Copilot violates the terms of open source licenses and infringes on programmers’ rights. Despite the lawsuit, I have a feeling that Copilot will probably stay here in some form, but it got me thinking: If developers are going to use an AI-assisted code generation tool, it would be more productive to think about how they can improve it instead of fighting for its right to exist.

Behind the Copilot Controversy

Copilot is a predictive code generator that relies on OpenAI Codex to suggest code — and full features — as coders build their own code. It’s very similar to the predictive text seen in Google Docs or Google Search features. As you begin to build a line of original code, Copilot suggests code to complete the line or fragment based on a saved repository of similar code and functions. You can choose to accept the suggestion or ignore it with your own suggestion, potentially saving you time and effort.

The controversy comes because Copilot draws its suggestions from an extensive training set of open-source code it has processed. The idea of ​​monetizing the work of uncredited open-source software contributors has irked many in the GitHub community. It has even resulted in a call for the open source community to leave GitHub.

There are valid arguments for both sides of this controversy. The developers who freely shared their original ideas probably didn’t intend to put it in a package and monetize it. On the other hand, it could be argued that what Microsoft has monetized is not the code, but the AI ​​technology to apply that code in an appropriate context. Anyone with a free GitHub account can access the code, copy it and use it in their own projects – without attribution. In this regard, Microsoft uses the code no differently than how it has always been used.

Taking Copilot to the next level

As someone who has used Copilot and seen how it saves time and increases productivity, I see an opportunity for Microsoft to improve Copilot and address some of the detractors’ complaints.

What the next generation of Copilot would improve on is a greater sense of context for its suggestions. To make actionable recommendations, Copilot could base them on more than a simple GitHub search. The suggestions can work in the specific context of the code being written. There must be some major AI technology at work behind the suggestions. This is both the unique value of Copilot and the key to improving it.

Software programmers want to know where the suggestions come from before accepting them and understanding that the code is suitable for their specific purposes. The last thing we want is to use proposed code that works enough to run when compiled, but is inefficient, or worse, prone to bugs or security vulnerabilities.

By giving more context to its Copilot suggestions, Microsoft could give the coder the confidence to accept them. It would be great if Microsoft took a look at the origins of the proposed code. A trace back to the original source – including some attribution – would accomplish this, as well as share some of the due credit. Just knowing that there is a window into the original open source repository can bring some peace to the open source community and would help Copilot users make better coding decisions as they work. I was happy to see Microsoft reaching out to the community recently to understand how to build trust in AI-assisted tooling, and I look forward to seeing the results of that effort.

As I said, it’s hard to imagine GitHub Copilot going away just because part of the community is angry at Microsoft repackaging their work behind a paywall. But Microsoft would be all about extending a digital olive branch to the open-source community while improving the effectiveness of its product.

Coty Rosenblath is CTO at Katalon.