DeepMind’s Game-Changing Patent that Uses Neural Networks to Write Code on Command

An artist’s illustration of artificial intelligence (AI). This illustration depicts language models which generate text. It was created by Wes Cockx as part of the Visualising AI project launched by Google DeepMind.

Google DeepMind has filed a patent application for a system that uses neural networks to generate computer code.

The system allows users to input task descriptions and then generates one or more computer programs that can perform the task. This is helpful because writing code to solve a specific task can be challenging, as even small edits can significantly change program behavior.

DeepMind’s neural networks simplify this process by breaking down the problem and recalling lines of code from a database of coding sequences. This enables developers to write specific code more quickly and accurately. The system can also be trained on specific data to create accurate code, even with limited amounts of hyper-specific data.

This filing is not surprising, as Google has been integrating AI into its products and services. DeepMind previously released AlphaCode, an AI-assisted tool for competitive programming. The patent suggests that DeepMind is moving towards a general AI with specific applications, including a code assistant tool. This particular patent could lead to a consumer or enterprise-grade product that significantly enhances programmer capabilities.

Google has filed for other patents related to automating parts of the programming process, creating user interfaces using machine learning, and developing AI-based tools for creating and publishing running apps. These filings indicate that Google may be building an AI-enabled development toolkit for engineers of all skill levels.

However, Google faces competition in this space. Microsoft’s GitHub Copilot and OpenAI’s Code Interpreter are AI coding assistants that offer similar functionality. Additionally, startups are likely working on their own coding assistants. The challenge will be developing a generalized solution that can assist with various coding tasks.