Machine Heart report
Editors: Du Wei, Chen Ping
Google DeepMind: Bringing together two world-class AI teams.
In the white-hot stage of AI competition, Google made another big move, announcing the merger of Google Brain and DeepMind to form a new department called Google DeepMind. The new division will lead breakthrough AI product research and advancements while maintaining ethical standards.
Leadership, DeepMind CEO Demis Hassabis will serve as CEO of Google DeepMind. Jeff Dean, one of the co-founders of Google Brain, will serve as the chief scientist of Google Research and Google DeepMind, reporting to Google CEO Sundar Pichai. Google Research will remain a separate division and will focus on computer science in areas such as algorithms and theory, privacy and security, quantum computing, health, climate and sustainability, and responsible AI.
Meanwhile, as part of the formation of Google DeepMind, Google said it will create a new scientific committee to oversee the department's research progress and direction, which will be led by Koray Kavukcuoglu, vice president of research at DeepMind. Eli Collins, vice president of products at Google Research, will join Google DeepMind as vice president of products, while Zoubin Ghahramani, head of Google Brain, will be a member of Google's DeepMind research leadership team, reporting to Kavukcuoglu.
Google CEO Sundar Pichai said in a blog post: "Integrating talent into a dedicated team, backed by Google's computing resources, will greatly accelerate progress in artificial intelligence."
Pichai sent an internal letter to Google employees
"We have been an AI-first company since 2016 and believe that AI is the most important way to achieve our mission. Since then, we've used AI to make improvements to many core products, from search, YouTube, Gmail to Pixel phones, and more. We've helped businesses and developers harness the power of AI with Google Cloud, and we've shown its potential to solve social issues like health and climate change.
Along the way, we've been fortunate to have two world-class research teams that are leading the industry forward with fundamental breakthroughs that usher in a new era of AI.
AI is now advancing faster than ever before. To ensure a bold, responsible development of AI in general, we are creating a division to help us build more powerful systems more safely and responsibly.
The team, called Google DeepMind, will bring together two leading research teams in AI: Google Research's Brain team and DeepMind. Their achievements in AI over the past decade include AlphaGo, Transformers, word2vec, WaveNet, AlphaFold, sequence-to-sequence models, distillation, deep reinforcement learning, and distributed systems and software frameworks for expressing, training, and deploying large-scale ML, such as TensorFlow and JAX.
Combining these talents into a dedicated team, backed by Google's computing resources, will greatly accelerate our progress in artificial intelligence.
As CEO of Google DeepMind, Demis Hassabis will lead the development of the most capable and responsible general artificial intelligence systems – which will help drive research into our next generation of products and services.
Jeff Dean will report to me in a senior role as Google's chief scientist. In this capacity, he will serve as chief scientist for Google Research and Google DeepMind. Together with Demis, Jeff will help determine the future direction of our AI research and lead our most critical and strategic technology projects related to AI, beginning with a series of robust, multimodal AI models.
Google Research will continue its important work leading fundamental advances in computer science in areas such as algorithms and theory, privacy and security, quantum computing, health, climate and sustainability, and responsible artificial intelligence, and will report to James Manyika.
I am very excited about the next phase of our journey as we progress in our mission and we will help people reach their potential through increasingly powerful and responsible AI."
Hassabis: Google DeepMind is committed to the next wave of world-changing breakthroughs
The following is the full text of Demis Hassabis' internal open letter.
When Shane Legg and I started DeepMind in 2010, many people thought of general artificial intelligence as a far-fetched sci-fi technology that would take decades to become a reality.
Right now, we live in an era of exponential advances in AI research and technology. In the coming years, AI and eventually AGI have the potential to drive one of the greatest social, economic, and scientific transformations in history.
That's why today Sundar announced the merger of DeepMind and Google Brain into Google DeepMind. Our talents and efforts will accelerate our march towards a world where "AI helps solve humanity's greatest challenges." I'm excited to lead Google DeepMind and build it with all of you.
I believe that the establishment of Google DeepMind will bring us closer to the desired future. Building more robust and versatile AI safely and responsibly requires us to solve some of the toughest scientific and engineering challenges of our time. As a result, we need to work more quickly, improve collaboration and execution, and streamline decision-making to focus on achieving maximum impact.
Through Google DeepMind, we will bring together world-class talent and computing power, infrastructure and resources in AI to enable the next generation of AI breakthroughs and products in Google and Alphabet in a bold and responsible way. The advances in Google Brain and DeepMind have laid a lot of foundation for the current AI industry, from deep reinforcement learning to transformers. In the future, Google DeepMind will be committed to making the next wave of breakthroughs that will change the world.
Sundar Pichai, Jeff Dean, James Manyika and I have built a great relationship in recent months together. I very much look forward to working closely with Eli Collins and Zoubin Ghahramani." In addition, we have created a new scientific committee for Google DeepMind to oversee the progress and direction of its research. In the coming days, Jeff Dean, Koray, Zoubin, Shane and I will work together to determine the composition of the committee.
Finally, I am excited to embark on this new journey with all of you and look forward to meeting you.
DeepMind Timeline
DeepMind is a London-based AI startup founded in 2010 by renowned AI researcher and game designer Demis Hassabis and others dedicated to combining the most advanced technologies in machine learning and systems neuroscience to build powerful general-purpose learning algorithms. The initial results were mainly used in the fields of simulation and game development.
In 2014, Google acquired DeepMind for $600 million. Since then, DeepMind has received a steady stream of financial support from Google's parent company Alphabet and has made a series of remarkable AI breakthroughs.
In March 2016, AlphaGo, an AI program developed by DeepMind, defeated South Korean Go champion Lee Sedol 4:1, becoming a milestone in the field of AI in recent years.
In October 2017, DeepMind released the latest version of AlphaGo
AlphaGo Zero
, ravaged AlphaGo 100-0 after 3 days of self-training.
In December 2018, DeepMind launched AlphaZero, which not only conquered Go, but also achieved performance beyond humans in complex games such as shogi and chess.
In the same month, DeepMind launched AlphaFold, which predicts the 3D structure of proteins based on genetic sequences. It has since been launched and open sourced
AlphaFold 2
, the AlphaFold dataset, and more.
In December 2020, DeepMind was released
MuZero
, using only pixels and game scores as inputs to reach human level in Atari video games.
In February 2022, DeepMind released a new model based on Transformer
AlphaCode
, which enables large-scale code generation.
In May 2022, DeepMind released the "generalist" AI agent
Cat
, with multi-modal, multi-tasking, multi-embodied characteristics, you can play Atari games, output subtitles to pictures, chat, stack blocks with robotic arms, and decide whether to output text, joint torque, button presses or other tokens according to the context.
In October 2022, DeepMind launched AlphaTensor, which uses reinforcement learning to discover matrix multiplication algorithms. It is the first AI system that can be used to discover novel, efficient, and provably correct algorithms for fundamental tasks such as matrix multiplication.
What major breakthroughs the new Google DeepMind will make in the future makes us look forward to it even more.
Reference Links:
https://www.deepmind.com/blog/announcing-google-deepmind
https://blog.google/technology/ai/april-ai-update/
https://twitter.com/demishassabis/status/1649097532772069377