AI darling and Alphabet subsidiary DeepMind mounted its total costs and losses to around £470.2 million through 2018, a stark figure against a backdrop of roughly £1.04 billion in debt to Alphabet and total revenue of around £102.8 million.
On a good note, though, that revenue is close to double what it made back in 2017. Even so, with its debt to Alphabet coming due and staff costs just about doubling year on year, hitting £398 million, DeepMind’s tendency to prioritize innovation and pushing boundaries over turning a profit is starting to become a real problem.
DeepMind defended its work in a statement, with a representative pointing to work that’s already beginning to make an impact and saying that the team has a “long-term mission” with a focus on improving AI technology overall and applying the tech for a range of beneficial use cases. The company already has a good amount of AI research that’s beneficial to society at large under its belt, such as its work in the healthcare world under the DeepMind Health banner.
Good will may not pay the bills, but the company’s advances have also benefited Google in a number of ways. DeepMind has a special team that works directly with Google on using AI innovation to improve core products, and these efforts have shown themselves in things like Android battery savings. DeepMind has also worked with other Alphabet companies in meaningful ways, such as helping Waymo’s self-driving cars to better detect pedestrians and predict happenings on the road.
DeepMind has perhaps made the most headlines for its more useless, yet awe-inspiring stunts. The company created AlphaGO, for example, an AI program that mastered the ancient Chinese board game that eluded AI scientists for decades and was long thought impossible for computers to play.
It also created an AI that can play famous strategy game StarCraft 2, along with an AI program that successfully taught itself to play first-person shooter Quake 3, and another that was given a humanlike 3D model and the goal of walking, and taught itself to do so with no help.
All of this seemingly aimless AI experimentation plays to a single goal, according to a recent interview that co-founder Demis Hassabis had with Wired, and that goal is to “solve intelligence”. Essentially, Hassabis and his team are trying to decode the basics of artificial intelligence in order to find new ways to approach the technology, which will have the end result of opening up new use cases, quite possibly an endless stream of them.
DeepMind, which emerged as a startup back in 2010, has become a force to be reckoned with in the world of AI research. Google rallied its resources behind DeepMind back in 2014, officially stationing the company as an arm of Alphabet after the larger company’s 2015 restructure. Since then, DeepMind has been able to secure top talent in the field, and will be moving into a brand new office in 2020 after having spent a few years in a King’s Cross office owned by Google. This move is being touted as a genesis of sorts for DeepMind, and only time will tell if that’s for real, or if more of the same is going to be happening under the new roof.