code

Google DeepMind’s AlphaDev artificial intelligence (AI) system has produced new algorithms that can write code better than humans and could create better programs. 

Discovering Enhanced Algorithms 

In a recent research paper, Google’s Deepmind has outlined how AlphaDev is able to use ‘reinforcement learning’ to discover enhanced computer science algorithms which surpass those honed by scientists and engineers over decades.  

Why? 

Google’s AI research organisation, Deepmind, says that with a digital society driving increasing demand for computation and energy use, plus with only a reliance on improvements in hardware so far to keep pace, microchips are approaching their physical limits. This means that it’s now critical to improve the code that runs on them to make computing more powerful and sustainable going forward. 

What Is Reinforcement Learning? 

Reinforcement learning is a subfield of AI where an agent learns by interacting with an environment to maximise cumulative rewards. The agent observes the current state, takes actions, and receives feedback to improve decision-making over time. The policy guides the agent’s action selection based on a set of rules. The iterative process involves observing the state, taking actions, receiving rewards, and updating the policy using algorithms like Q-learning or policy gradients.  

Reinforcement learning is applied in game playing, robotics, recommendation systems, finance, and healthcare, achieving impressive results such as training agents to play games at superhuman levels and enabling autonomous systems to learn complex tasks independently. 

A Faster Algorithm For ‘Sorting’ Discovered 

In computing, ‘sorting’ is a method for ordering data, with sorting algorithms underpinning everything from ranking online search results and social posts to how data is processed on computers and phones. 

The recent research paper revealed that using reinforcement learning, following being tasked with developing a new way to sort short sequences in numbers in the popular coding language C++, AlphaDev was able to uncover a faster algorithm for sorting, beating well-established sorting algorithms.

In essence, the AlphaDev has been able to improve how the structure of a working assembly program is represented in the AI code, thereby allowing its reward system to better narrow down the possibilities, making the AI better and faster. 

C++ Running Faster, & More 

This discovery has led to C++ running faster and improved algorithms for sorting and other basic tasks like hashing. For example, AlphaDev’s new C++ sorting algorithms are 1.7 per cent more efficient than previous methods for sorting long sequences of numbers, and up to 70 per cent faster for five-item sequences.  

When scaled up, this could transform how we program computers and could impact all aspects of our increasingly digital society. 

Added To The Main C++ Library 

The improved algorithms have been open sourced and added to the main the main C++ library, used by millions of developers and companies in many industries around the world on AI applications. DeepMind says it’s the first change to this part of the sorting library in over a decade and is the first time an algorithm designed through reinforcement learning has been added to this library.   

What Does This Mean For Your Business? 

With so many AI applications in so many industries using sorting algorithms, and with AlphaDev’s improved algorithms being open sourced, AlphaDev’s discovery is already making an impact by speeding many apps and programs. DeepMind believes that many improvements exist at a lower level that may be difficult to discover in a higher-level coding language and sees this as just the beginning and an important steppingstone for using AI to optimise the world’s code, one algorithm at a time.  This means that AlphaDev’s reinforcement learning approach could point to many more small algorithm improvements to come that could be quickly fed into the hands of developers through open sourcing thereby benefitting businesses, industries, and users around the world.

If you would like to discuss your technology requirements please:

Back to Tech News