From unlikely start-up to major scientific organisation: Entering our tenth year at DeepMind

Since we started DeepMind nearly 10 years ago, our mission has been to unlock answers to the world’s biggest questions by understanding and recreating intelligence itself.

As we approach the end of 2019, we’ve come a long way in building the organisation we need to achieve this long-term mission - from our environment for research, to our collaborations with other Alphabet companies, to our increasingly interdisciplinary and diverse team.

Pioneering research, growing impact

A mission this ambitious requires pioneering research on many fronts over many years. We’ve been privileged to make some significant advances over the past twelve months, from winning a major international contest to predict the shapes of proteins - the building blocks of life - at the end of 2018, to developing AI agents that cooperate with each other and people in our Capture the Flag paper published by Science, to our latest work on mastering the complex strategy game StarCraft II, which led to a Nature cover article last month.

Beyond the moments that capture widespread imagination, we have a broad programme of fundamental research with hundreds of papers published each year. Many are substantial and exciting contributions to their domains, such as MuZero, which can not only master chess, Go, Shogi but also extend to Atari, without knowing the rules first. We’ve developed new ways to open up and explain our work, such as our dedicated AI safety research blog, thematic summaries of research concepts such as unsupervised learning, and our many open source projects, which bring new frameworks and environments to the field.

As our research matures, we’ve been finding more opportunities to partner with others for social and commercial impact, often with our colleagues across Alphabet. This year, we demonstrated how AI could predict potentially fatal patient deterioration two days before existing tests, used machine learning to accelerate ecological research in the Serengeti, collaborated with Waymo on evolutionary selection to train more capable self-driving cars, learned how to boost the value of Google’s wind energy, helped Play Store users discover more relevant apps, and more.

Entering our next phase

As I discussed with Wired in the summer, this year feels like the start of a new phase for DeepMind as an established scientific organisation. We’re no longer the tiny start-up working in a very unfashionable area, as was the case in 2010. Fuelled by breakthroughs like DQN and AlphaGo, and many exciting advances from colleagues at other labs, AI is once again one of the most vibrant areas of scientific research.

Over the past year, we’ve also been formalising a leadership team with the seasoned experience and skills for our second decade. We want to ensure DeepMind continues to be the best place in the world for fundamental breakthroughs in AI, and that we conduct this work thoughtfully and responsibly. Much of this work is led by Chief Operating Officer Lila Ibrahim and VP of Research Koray Kavukcuoglu.

Lila joined DeepMind in April 2018 following a distinguished, global career including a long stint at Intel where she started as an engineer and took on management roles including Chief of Staff to the CEO and Chairman, and then leadership roles at investment firm Kleiner Perkins and education startup Coursera. Koray is one of the foremost research scientists in AI and deep learning, and a key contributor to many of DeepMind’s biggest breakthroughs, from the DQN system that mastered Atari in 2013 to the research and deployment of WaveNet, which improves the experience of Google users around the world.

As we enter this next phase, Mustafa Suleyman is leaving DeepMind. I founded DeepMind back in 2010 along with Shane Legg (our Chief Scientist) and Mustafa. As a serial entrepreneur, Mustafa played a key role over the past decade helping to get DeepMind off the ground, and launched a series of innovative collaborations with Google to reduce energy consumption in data centres, improve Android battery performance, optimise Google Play, and find ways to improve the lives of patients, nurses and doctors alike. Mustafa leaves DeepMind having helped set us up for long-term success, and I’m looking forward to what he’ll achieve in the years ahead as he joins Google in a new role.

Many grand challenges remain

Of course, there’s a long way left to go for DeepMind and for the AI field overall, and many grand challenges remain. Right back to our origins blending neuroscience with machine learning, we’ve found that breakthroughs happen faster when different disciplines come together. In our offices across the world - soon to include our incredible new purpose-built HQ in London - we recruit and develop brilliant people with backgrounds in research and engineering, program management, games design, operations, ethics and safety research and beyond. You can get to know just a fraction of our team by listening to the DeepMind podcast. Increasing the diversity of our workforce - and doing what we can to improve access to science among under-represented groups - remains a personal and organisational priority.

Thank you to the hundreds of amazing colleagues who have made DeepMind what it is today, and to all those at Alphabet and in the wider AI community for your long-term support and collaboration. I can’t wait to show you what we have coming up... all I can say for now is, keep your eyes peeled for some very exciting advances in 2020!