Just yesterday, GitHub announced that it is working on a new feature for its platform called “Copliot“; which is an artificial coding assistant that predicts the next chunks of code that a programmer may want to write while developing software, and offers to insert it just in the right time and place.
The technology, a bleeding-edge application of deep learning and neural networks, was trained using the public repositories published on GitHub. Training a neural network model means that you take the data (source code of these repositories in our case) and feed it to the network, so that it can learn what to do in future similar cases.
Copilot has seen billions of lines of code, functions, classes and object definitions before, and hence, can suggest the next steps whenever enough information about the programmer’s desire are determined.
However, this brought a large issue into debate: Many of these public repositories were licensed under the GPL license and other copyleft licenses (MIT, AGPL… etc), so is this process legal? Is it OK for GitHub to use free software source code to train its proprietary, paid and commercial service?