Google DeepMind's AlphaDev, an artificial intelligence (AI) system derived from the foundation of AlphaZero, has successfully generated algorithms capable of sorting data up to three times as fast as their human-created counterparts. The AI system uses reinforcement learning to develop innovative algorithms that surpass the proficiency of those developed by programmers over several decades.
The details of the project were disclosed on DeepMind's blog and published in the scientific journal Nature. As an upgrade to AlphaZero, which excelled in mastering the game strategies of Go and chess, AlphaDev shows promise in revolutionizing algorithm optimization techniques.
AlphaDev's training process in sorting was carried out utilizing a method described by researchers as a 'single-player assembly [language] game.' Through this approach, sorting algorithms were developed incrementally, with AlphaDev consistently seeking options to improve upon previous iterations. The AI system employs neural networks to compare and move values, aiming to achieve accurate results in the shortest time possible.
DeepMind scientist Daniel Mankowitz commented, Moore's Law is coming to an end, where chips are approaching their fundamental physical limits. We need to find new and innovative ways of optimizing computing. AlphaDev appears to be a significant stride in that direction.
The focus of the research was centered on sorting short lists ranging from three to five characters, which are reportedly the most commonly used algorithms by programmers. According to DeepMind, such algorithms are utilized trillions of times daily. For longer sorting sequences consisting of up to 250,000 elements, improvements in speed were marginal when compared to existing methods.
With its current achievements, the next goal for AlphaDev is to investigate optimization techniques in higher-level programming languages, such as C++. This exploration is expected to yield even greater speed enhancements and provide further benefits for developers.
AlphaDev's breakthrough could potentially drive advancements in low-code and no-code platforms like AppMaster, where developers and non-developers alike build robust backend, web, and mobile applications. With a platform like AppMaster, users can visually construct data models, create business logic through an intuitive BP Designer, and develop REST API and WSS endpoints. This facilitates the rapid and cost-effective creation of scalable software solutions for a wide range of customers, from small businesses to large enterprises.
