Google’s DeepMind Coding AI Competes with Human Programmers
by Adam Pease
Google has just released a comprehensive new article about its AI research subsidiary called DeepMind. The article details the lab’s progress on a code-generating AI system named AlphaCode. In the new report, Google boasts that AlphaCode has demonstrated its ability to autonomously generate computer code at levels that rival human coders. In this blog, we discuss the new research and what it may mean for the enterprise.
What Is AlphaCode?
Google says that its new AI for generating computer code performs within the top 54% of contestants in programming competitions. Using the platform Codeforces, it tested the AI’s ability to autonomously write code in ten different contests, where it managed to place around the median skill level.
In a moment reminiscent of its AlphaGo board game breakthrough, DeepMind states that this is the first time an AI coding system has managed to compete with humans in a competition. If Google’s AI can reliably perform this well, it marks a major turning point in the development of generative AI that enterprises should pay attention to. The power to instruct a machine to generate code rather than a human programmer would transform the way that enterprises develop software and maintain their applications.
Competition Heats Up in the Generative AI Space
Recently, we covered the progress of OpenAI, a Microsoft-partnered AI research firm incubated by Elon Musk back in 2015. While we remarked on the progress of OpenAI’s predictive language models and their surprising ability to generate near-photorealistic images, OpenAI is also involved in producing an AI model for code generation similar to AlphaCode.
Microsoft’s ongoing support of OpenAI and extension of its services through platforms like GitHub suggests that there is a cold war underway for the development of generative AI tools. Equipped with the ability to generate code itself, enterprises could save untold financial resources and time. Now that AlphaCode has demonstrated that its AI can keep pace with human coders, it is only a matter of time until the system is streamlined and made available for enterprise use.
Generative content is in its very early stages of enterprise penetration, but its potential value should not be understated. As Microsoft and Google compete through partners and subsidiaries to develop the best-in-class enterprise tool for code generation, it would be wise for business leaders that depend on developers to keep an eye on these developments. Look out for more research from Aragon soon about the definition of generative content.