What Is Google DeepMind?

Google DeepMind is an Alphabet subsidiary focusing on artificial intelligence, machine learning, and neuroscience research. Since its beginnings in 2010, the company has developed AI systems including identifying eye diseases faster and playing complex games like chess, Go, and Shogi. 

The History of Google’s DeepMind 

The origins of Google DeepMind can be traced back to 2010. Initially called DeepMind Technologies, the company was founded in London by Demis Hassabis, Shane Legg, and Mustafa Suleyman. The founders wanted to combine machine learning, neuroscience, engineering, mathematics, simulation, and computing infrastructure to create innovative deep-learning solutions. Their early successes were in training their AI models to learn how to play classic games like Atari, Pong, and Space Invaders.

Google acquired the company in 2014, and it's now a subsidiary company of Alphabet. Since then, Google DeepMind's AI has been trying to make progressive leaps into AGI (Artificial General Intelligence) all-purpose tools, a type of AI that show characteristics of human thinking.

In April 2023, Google Brain, the earlier AI division of Google, was merged into Google DeepMind to form a single entity. DeepMInd's founder, Demis Hassabis, will be at the helm of the new research company.

Achievements of Google DeepMind

DeepMind trained its early machine learning systems on classic arcade games from the seventies and eighties. They used reinforcement learning, where the program could learn through trial and error using feedback from its actions. In short, DeepMind's system could learn from experience, unlike other AI systems that were programmed for specific tasks. These innovations brought it to the attention of Google. Since then, Google DeepMind has had several notable AI wins to its credit.

AlphaGo (2016): The Go board game originated in China and is considered more complex and abstract than chess. AlphaGo is an AI program that defeated the top-ranked Go players in the world and is now considered the greatest Go player of all time.

WaveNet (2016): An advanced text-to-speech technology creates more natural-sounding speech from any raw audio signal. The generative model can also generate natural-sounding music or any other sound by directly modeling the raw waveform of the audio. This technology improves text-to-speech synthesis in apps like Google Assistant, music composition, and other audio-related applications.

AlphaStar (2019): Google DeepMind's AI mastered StarCraft II, a Real-Time Strategy (RTS) game. The challenging multi-layered nature of the game made it an esports favorite across tournaments. AlphaStar bested the top professional player ranked as Grandmasters.

AlphaFold (2016 - 2022): Protein structures are at the heart of every field in biology and medicine. There are 200 million known proteins, and decoding their unique structures in 3D is time-consuming. AlphaFold is an AI tool that predicts the three-dimensional structure of a protein from its amino acid in minutes. This breakthrough can revolutionize drug discovery and help in finding cures for a wide range of diseases.

AlphaCode (2022): Google DeepMind's AI system can write computer programs on par with competitive programmers. AlphaCode used programming challenges to develop critical thinking, logic, algorithms, coding, and natural language understanding.

DeepMind Is a Key Part of Google's AI Future

DeepMind's conversational AI technology, LaMDA (Language Model for Dialogue Applications), is also behind Google Gemini. While Gemini is capturing attention, there have been other contributions by DeepMind to our Google experience in the years before.

For instance, DeepMind has improved the accuracy of traffic slowdown predictions in Google Maps. In addition, DeepMind's machine learning also went into Android's adaptive battery and brightness features. And Google also applied DeepMind's AI to cut the energy consumed by their data centers by 40 percent.

We're confident in saying any AI updates from Google will have DeepMind's fingerprint all over them.