Google Cloud: Accelerating Digital Change

Published: 14 June 2023

Authors: Our AI research is making Google Cloud solutions faster, simpler, and more efficient

Google Cloud empowers organizations to digitally transform themselves into smarter businesses. It offers cloud computing, data analytics, and the latest artificial intelligence (AI) and machine learning tools. Using our AI research, we’re enhancing these solutions for Google Cloud customers all over the world. Our research is deciphering written documents, enhancing the value of wind energy, and improving the usability of AlphaFold — our breakthrough AI system designed to better predict protein structures.

From cuneiform tablets to the printing press, countless ways of sharing written knowledge have evolved throughout history. Modern documents vary across countries, languages, and industries—making it challenging to extract and utilize that information, especially at scale. Google Cloud’s Document AI enables users to make digital, printed, or handwritten information contained inside a document—like an invoice or tax form—extractable and queryable.

Before Document AI, industries looking to use AI tools for document understanding required vast amounts of training data to perform accurately. However, such data often remains unavailable, incomplete, or improperly annotated, hindering widespread AI adoption. Collaborating with the Google Cloud Document AI team, we developed innovative machine learning models that require 50-70% less training data than traditional methods to process documents like utility bills and purchase orders. We are also focused on improving Document AI’s performance in languages with smaller datasets, ensuring more customers across various industries and geographies can benefit from Document AI.

As part of our commitment to utilizing AI for achieving net-zero emissions by 2030, we partnered with Google Cloud Professional Services to advance the wind energy sector and help build a carbon-free future. Wind farms are a crucial source of carbon-free electricity, but their output can fluctuate based on weather conditions. To balance electricity grid supply and demand, operators depend on energy generation forecasts. If operators can commit to selling a certain amount of electricity based on next day’s forecast, they can secure a better price.

In collaboration with Google Cloud, we developed a custom AI tool to improve predictions of wind power output. This tool was trained on weather forecasts and historical wind turbine data. An additional model suggests how much energy an operator can commit to delivering to the electricity grid a day in advance. ENGIE, a global energy and renewables supplier, is currently piloting this technology in Germany, and if successful, it may be rolled out across Europe. Enhancing the economic attractiveness and reliability of wind energy will promote the adoption of renewables, benefiting everyone.

The development of a new machine learning model involves several stages—from design to deployment—and requires strong data infrastructure. To support data scientists and businesses, Google Cloud built Vertex AI, a unified platform that provides access to machine learning tools throughout the development journey. After launching our groundbreaking AlphaFold system, which accurately predicts protein 3D structures, we made it accessible on Vertex AI. This allows scientists working in diverse fields, such as drug discovery and combating plastic pollution, to run the AlphaFold prediction workflow more seamlessly by tracking experiments, optimizing hardware selection, and managing everything at scale.

In 2022, we expanded the AlphaFold Protein Structure Database to include nearly all cataloged proteins known to science. We partnered with Google Cloud to host this extensive database, providing more than 200 million proteins for bulk download. Billions of structures have already been downloaded, and the database has become an essential tool for the scientific community, driving a new wave of advancements in biology.