DeepMind’s new AI can write code that’s better than yours

Artificial intelligence (AI) company DeepMind has developed a new coding engine capable of outperforming many human developers, called AlphaCode.

As explained in a paper published by the firm, AlphaCode is able to write computer programs at a “competitive level”, ranking among the top 54% of humans in a competition setting.

The performance of the AI model was assessed using ten challenges hosted on Codeforces, a platform that attracts developers looking to flex their coding muscles by competing with other to solve specialized problems.

Although these challenges are not typical of the average day-to-day workload, the ability to solve the problems they present in a creative manner is an excellent indicator of programming ability. AlphaCode is the first ever AI system able to compete with humans in this context. 

DeepMind AlphaCode

Acquired by Google in 2014, DeepMind has long been at the forefront of artificial intelligence development. Most famously, the company developed AlphaGo, which became the first ever computer program to defeat a human player at the ancient Chinese game of Go.

DeepMind R&D activity extends into a multitude of areas in which AI could feasibly be deployed, but the company appears to have struck upon a fruitful new opportunity with AlphaCode.

The model was pre-trained on a large pool of open source code drawn down from GitHub and refined using a smaller set of data collected from programming competitions, a step described as “critical for performance”.

Although many humans outperformed AlphaCode in testing, its ability to create novel solutions to complex problems highlights the potential for AI to be integrated into the development landscape in years to come.

“Solving competitive programming problems is a really hard thing to do, requiring both good coding skills and problem solving creativity in humans,” explained Petr Mitrichev, a competitive programmer from Google.

“I was very impressed that AlphaCode could make progress in this area, and excited to see how the model uses its statement understanding to produce code and guide its random exploration to create solutions.”

What’s more, DeepMind says its research into the area is still in the early stages and the level of performance on offer with AlphaCode is therefore just the beginning.

“Our exploration into code generation leaves vast room for improvement and hints at even more exciting ideas that could help programmers improve their productivity and open up the field to people who do not currently write code,” wrote DeepMind.

“We will continue this exploration, and hope that further research will result in tools to enhance programming and bring us closer to problem-solving AI”.