DeepMind’s efforts to achieve artificial general intelligence have won the firm both plaudits and critics since it was founded in 2010. The firm’s research into deep learning techniques convinced the search engine giant Google to spend £400 million on the company in 2014, but it has since incurred heavy losses and while its scientific discoveries have earned acclaim, DeepMind has also been rebuked for its laissez faire approach to data privacy and security practices.
Here’s our timeline of DeepMind’s short but eventful history.
September 2010 – DeepMind founded
DeepMind was founded in London by machine learning researcher Shane Legg and childhood friends Demis Hassabis and former consultant Mustafa Suleyman. The cofounders all met at University College London, where Legg was a research associate and Hassibis was studying for a PhD in cognitive neuroscience.
The trio declared a grand ambition for their new company: “To solve intelligence and then to use that to solve everything else.”
They initially pursued this lofty goal through video games. A 16-year-old Hasabis had co-developed the hit simulation game Theme Park, and at 22 was running his own games studio. He combined this experience with his neuroscience PhD to create AI programmes that could master video games. One of these systems taught itself how to play 49 different Atari games, including Pong and Space Invaders, just by viewing the score and pixels on the screen.
December – DeepMind unveils deep reinforcement learning
These experiments with video games led DeepMind to focus on an AI technique called deep reinforcement learning, which combines the pattern recognition of deep learning with the reward signals for completing tasks achieved through reinforcement learning.
DeepMind announced the technique in a research paper about its Atari trials, which it called “the first deep learning model to successfully learn control policies directly from high-dimensional sensory input using reinforcement learning.”
The technique was responsible for DeepMind’s most impressive achievements, but the company’s relentless focus on the technique has been questioned by some AI experts. In August 2019, Gary Marcus, the founder of Robust.AI and a professor of psychology and neural science at NYU, noted in Wired that the company was still yet to find a large-scale commercial application of deep reinforcement learning.
“Ten years from now we will conclude that deep reinforcement learning was overrated in the late 2010s, and that many other important research avenues were neglected,” he wrote. “Every dollar invested in reinforcement learning is a dollar not invested somewhere else, at a time when, for example, insights from the human cognitive sciences might yield valuable clues.”
January – Google buys DeepMind
Google made DeepMind one of its biggest-ever European acquisitions when it splashed out £400 million on the London-based startup. Google agreed to establish an AI ethics board as part of the deal, but the members and workings of the board have never been made clear.
October – AlphaGo takes on Go champions
A DeepMind-created system became the first AI to beat a professional Go player when AlphaGo routed European champion Fan Hu by a score of five to zero. Later that year, the system defeated Ke Jie, the world’s number one player of the ancient and highly complex board game.
September – DeepMind signs information sharing agreement with Royal Free hospital
DeepMind began its controversial relationship with the Royal Free hospital in London when the two organisations signed a deal that gave the Google subsidiary access to healthcare data on 1.6 million patients. DeepMind later announced that the partnership had yielded an app called Streams that would help clinicians monitor patients for early signs of kidney disease.
February – DeepMind Health founded
DeepMind turned its ambition to use AI to improve healthcare into a separate division of the company: DeepMind Health. Suleyman, whose mother was an NHS nurse, who chosen to lead the unit.
Suleyman went on to sign further NHS deals with Taunton & Somerset Foundation Trust, Yeovil District Hospital, University College London Hospital, Imperial College Healthcare and Moorfields Eye Hospital to apply AI to various medical challenges.
July – ICO rules that NHS illegally handed DeepMind data
The Information Commissioner’s Office (ICO), the UK’s data regulator, ruled that the Royal Free “failed” to comply with data protection rules when it provided DeepMind with patient data as it didn’t properly inform patients about how their details would be used.
The Royal Free accepted the findings and was not fined. The Trust announced that it had started to address the concerns.
February – DeepMind announces collaboration with the U.S. Department of Veterans Affairs
DeepMind revealed it had attracted a major client in the US when it announced that it was teaming up with the US Department of Veterans Affairs to predict patient deterioration by analysing patterns in medical records.
The project also involves researching ways to improve the algorithms DeepMind uses to detect acute kidney injury.
November – DeepMind announces that its health division would be absorbed into Google Health
Privacy campaigners raised alarm when DeepMind announced that its healthcare subsidiary was being absorbed into Google. The arrangement meant that the group would no longer operate as an independent unit but instead merge with the newly-formed Google Health team, led by former Geisinger CEO David Feinberg.
Critics argued that the shift betrayed DeepMind’s promise never to share data with its parent company. DeepMind claimed that all patient data would remain separate from Google services and projects.
December – AlphaFold wins protein-folding competition
DeepMind made its biggest scientific breakthrough yet when its AlphaFold system won a competition to predict the 3D shapes of proteins based on their genetic codes. The victory suggested that AI could help understand the protein-folding puzzle that plays a key role in the development of new drugs.
“This is a lighthouse project, our first major investment in terms of people and resources into a fundamental, very important, real-world scientific problem,” DeepMind CEO Demis Hassabis told the Guardian.
January – AlphaStar triumphs at Starcraft
DeepMind continued its long history of applying AI to video games by introducing AlphaStar, a programme that can play strategy game StarCraft II. The system went on to defeat some of the world’s best StarCraft II players.
August – Mustafa Suleyman placed on leave
DeepMind announced that Mustafa Suleyman, the company’s cofounder and head of applied artificial intelligence, was leaving the company for an indefinite period that the company said would likely end later the same year. DeepMind claimed that the decision was mutual and not related to his performance, but rumours spread that his departure was related to the company’s various healthcare controversies.
Read next: Google DeepMind loses its cofounder Mustafa Suleyman indefinitely
September – Google completes takeover of DeepMind health division
On September 18, Dr Dominic King, the UK site lead at Google Health, announced in a blogpost that Google had completed its takeover of DeepMind’s health division.
“It’s clear that a transition like this takes time,” he wrote. “Health data is sensitive, and we gave proper time and care to make sure that we had the full consent and cooperation of our partners. This included giving them the time to ask questions and fully understand our plans and to choose whether to continue our partnerships. As has always been the case, our partners are in full control of all patient data and we will only use patient data to help improve care, under their oversight and instructions.”
The Royal Free, University College London Hospitals, Imperial College Healthcare, Moorfields Eye Hospital, Taunton & Somerset, and University College London Hospitals NHS Foundation Trust all went on to release statements confirming that their contractual arrangements had been moved to Google.