Commit Assistant, an AI from Ubisoft, can now Predict Errors in the Code Before you Make Them
- Commit Assistant, Ubisoft’s new AI application, can predict errors in code before they’re made
- The model was trained on approximately 10 years’ worth of codes
- It will help save the company money and time by speeding up the development process
- Read on to find out more and view the video of how the AI works
Artificial Intelligence continues it’s march deep into the gaming territory. Now, there’s a new application of AI – predicting bugs in a game’s code before the developer even makes the error.
The French gaming company Ubisoft has released “Commit Assistant”, an AI to help speed up the development process of games by reducing (and in some cases eliminating) the number of errors in the script.
We’ve all played video games where we’ve seen bugs. No matter how much beta testing goes on before a game’s release, there are almost always inevitably bugs that make it through the screening process and lead to bad publicity for the game and the company.
The developers of this AI trained their model on almost 10 years’ worth of code from Ubisoft’s software library. The purpose of doing this was to look at the history of errors made in the code, to learn from them, and to flag them if they crop up in the current coding process.
The company claims that testing and correcting bugs manually can cost the company almost 70% of the budget so this will be a welcome change in that respect. However, the AI is still very much in it’s infant stages. How will the developers react to a machine telling them their code is about to be wrong? These questions will only be answered as the AI continues to get better.
The below video demonstrates how the Commit Assistance works behind the scenes:
Our take on this
As we mentioned in previous AVBytes articles, AI in gaming is a hot topic these days. This particular AI has the potential to be a game changer however, and not just in the gaming industry.
Imagine working on a product code and before you commit an error, it’s already been flagged so you can make the required changes? Sounds like a magnificent deal to me.
The downside here of course is that, like most machine learning models, it requires tons and tons of data to learn what kind of errors were made in the past. This can be expensive and might make it’s adoption slower than one imagines at this point.
Subscribe to AVBytes here to get regular data science, machine learning and AI updates in your inbox!