Google is a creepy organisation.
I don’t know whether it’s their satellite-imaging program, their 1000-acre data farms, or maybe their army of Terminator-style military robots; all I know is, Google creeps me out.
While all of these initiatives are points for concern, there is one area of Google’s business that goes above and beyond when it comes to unsettlingly activity — the mysteriously named Google DeepMind.
Described by one investor as a “Manhattan project” for artificial intelligence, DeepMind is one of those fascinating, yet weirdly unnerving projects that seems to thrive in secrecy. Everything about it screams sci-fi cult. From its oddly Orwellian name through to its ambiguous mission statement (“to solve intelligence”), DeepMind can’t help but come across as a bit sinister.
So what do we actually know about this mysterious project?
For starters we know that DeepMind started life as a London-based artificial intelligence firm. Founded in 2011, DeepMind was the brainchild of Denis Hasssabis, a former chess prodigy, video game designer, and advanced neuroscientist.
Then in 2014, Google acquired the company for an undisclosed sum (estimated at around £450M). Since then, the search giant has remained uncharacteristically quiet about DeepMind’s work. Even the DeepMind website, which contains a mere 176 words, tells us virtually nothing about what the business does.
While it’s clear that Google doesn’t want people asking questions about what goes on at DeepMind HQ, what we do know for sure is that this deal is a very very big deal. Not only was the search giant willing to pay £450 million for the business — fighting off a rival bid from Facebook — it has also subsequently thrown everything it’s got at the project (both in terms of investment and manpower).
Now, one year on from the initial investment, Google is at long last starting to lift the lid on project DeepMind. And I must say… It looks awesome.
According to Google, at the heart of the project lies a focus on “deep reinforcement learning”. Unlike traditional artificial intelligence systems, which merely follow predefined patterns, DeepMind’s system is not pre-programmed. Instead, it actually learns from its experiences using nothing more than raw pixels as input.
In one of the firm’s earliest experiments, DeepMind AI taught itself to play the classic Atari video game Space Invaders. Provided with nothing more than an objective (to rack up as many points as possible), DeepMind was able to master all of the game’s rules, controls and subtleties within a single day.
Since this initial test, DeepMind has subsequently taught itself to play a total of 49 classic Atari video games, often reaching a level of skill that most human players would be unable to best. The system is now moving onto three-dimensional games including sci-fi classic, Doom.
But it’s not just video games that the DeepMind team has been working on. In fact, Google has described this as little more than a test phase, ultimately hoping to put the firm’s research to far greater use. Already, the search giant is looking at ways to implement DeepMind within its controversial ‘Google Brain’ initiative, which sets out to build a functioning human brain from more than 16,000 integrated computers.
But it doesn’t stop there. According to an announcement last month, Google is now calling upon DeepMind to help solve one of the most pressing challenges of all: finding a cure for cancer.
Yes, this March the AI unit began work with Google’s Life Sciences division, which has been mapping cancer cells in an effort to better understand the disease. Following this, Google has entrusted DeepMind with the task of generating a solution to effectively “turn off” these cells, relying on the firm’s recent research into intelligent nanotechnology.
While all this may sound like science fiction, the truth is that these things are already well underway — admittedly behind the locked doors of Google HQ. While it’s unclear how long these projects will remain shrouded in mystery, one thing is for sure: if you want to understand the future of tech, DeepMind is a good place to start.
Photo credit: DeepMind