DeepMind is a British AI startup which was relatively unknown until it was bought by Google for around $600 million in 2014. Since then DeepMind has continued to refine its neural-network driven technology which has broken new frontiers with machine learning, particularly deep learning.
Perhaps DeepMind’s most famous accomplishment so far is being the brains behind AlphaGo, the first computer program to beat a professional human player of the board game Go. AlphaGo was developed by feeding DeepMind’s machine learning algorithms with 30 million moves from historical tournament data, and then having it play against itself and learn from each defeat or victory.
DeepMind’s work is based on a solid grounding in neuroscience. Two of the founders – Demis Hassabis and Shane Leg – met while undertaking research at the UCL’s computational neuroscience unit, and Hassabis has a PhD in the subject. This has underpinned their strategy of developing AI by teaching computers to mimic the thought processes of our own brains, in particular how we use information to make decisions and learn from our mistakes.
DeepMind has found itself rather adept at playing games. Arcade games from the 1970s and 1980s have been another area of focus. The AI became proficient at playing the Atari game Breakout. Without knowing anything about the game, it was simply told to score as highly as it could, and after two hours’ training it became an expert. The cognitive processes which the AI goes through are said to be very like those a human who had never seen the game would use to understand and attempt to master it.
It’s unlikely that Google thinks there is a great deal of money to be made in the short-term from teaching computers to play dated arcade games. But where it undoubtedly does see a business case is the project’s usefulness in advancing the fields of machine learning and simulated neural networks – developing machines with more human-like thought processes, with the capacity to carry out jobs which previously would have required trained humans.
Another DeepMind project involves a collaboration with London’s Moorfields Eye Hospital. DeepMind has been given access to one million images from historical eye scans, along with associated (anonymized) patient data. It is training itself to read the scans and spot early signs which may indicate the onset of degenerative eye conditions. The hope is that eventually the AI will become proficient enough to spot warning signs far earlier than a human could.
DeepMind has also been implemented across numerous other healthcare projects, such as a collaboration with UCL’s radiotherapy department to reduce the amount of time it takes to plan treatments. The group has said that by unleashing machine learning on the process of mapping a patient’s head and neck area, the time taken to create treatment plans for these complicated procedures could be reduced from four hours to around one hour.
Another project has been developing an “early warning” app known as Stream, which will go into use in the NHS (National Health Service) next year. Stream analyzes patient data and delivers cellphone alerts directly to a doctor or nurse when urgent intervention may be required. Initially it is being trained to detect signs of acute kidney injury (AKI) but could potentially learn to spot many other conditions.
Google is not seeing any direct profits from these healthcare partnership, which may initially seem strange considering it spent more than half a billion dollars on the tools to do these jobs. But all Google is really doing is staying true to its core belief that knowledge – or data – is the real prize. The experience and learning that DeepMind will acquire from analyzing these medical datasets is a reward in its own right, and will help to further refine its cognitive processes, enabling it to tackle ever more complex challenges.
DeepMind is also spearheading Google’s involvement with the Partnership on AI to Benefit People and Society – a non-profit which it co-founded alongside Facebook, Amazon, IBM, Apple and Microsoft to research using AI to drive social change as well as examining the ethical implications of doing so.
There is at least one known utilization of DeepMind technology, however, where it’s clear Google does hope to see short-term financial gain. The AI has been given the job of managing power consumption at its massive data centres, where Google claims it has reduced the energy used for cooling by an impressive 40%.
This has been done by feeding it historical sensor data on the operation of the cooling systems, which dynamically adjust according to demand as user activity changes across the group’s services – web search, cloud storage, Gmail, YouTube videos, mobile app marketplace, and everything else.
The AI has been able to more accurately predict how the temperature in each part of the building will be affected by these spikes in use, and more efficiently regulate the distribution of power to cooling machinery. After the successful trials, plans are in place to have the technology operational in its 16 main data centres soon.
In striking a balance between commercial opportunity and learning opportunity, DeepMind has demonstrated that it is taking a long-term approach to the development of its AI. In fact, not too different from the way a human parent or teacher might approach overseeing the intellectual development of a child. Perhaps we will find out that giving infant AIs time to play, and to find their feet, before putting them out into the world and expecting them to earn a living will lead to the development of more healthy, rounded adult AIs.