Artificial Intelligence has been a hot topic over the past few years. The major players in the market have been making some significant leaps in increasing the accessibility and opportunities that AI presents. One of the more intriguing factions of AI is Google’s subsidiary “DeepMind”. DeepMind was originally an AI start-up founded by Demis Hassabis which was eventually bought by Google for $500,000,000. Now Google DeepMind, the department is set out to “solve intelligence” by setting up general purpose algorithms that are inspired by the human brain. To decode that phrase used repeatedly by DeepMind, they are essentially producing AI that learns and optimizes through experience, much like how the human brain works. The AI can improve upon experience, without instruction, eventually giving it the ability to predict and foresee specific outcomes.
Developing DeepMind
DeepMind developed its AI through beta testing it on popular Atari video games such as Space Invaders, Pong, Brick Breaker and Pac-Man. The AI is not instructed controls or rules to the game; all it is told is to maximize the score. The algorithm can learn how the game works by repeatedly running simulations, eventually optimizing itself to predict outcomes and master the game. This is the first AI to be self-taught without prior instructions. Google calls it “Deep Reinforcement Learning”. This type of experiment is similar to others, such as IBM’s Deep Blue, but never without instruction or with the ability to learn from experience. DeepMind has gone on to test this algorithm in 3D games such as Doom and has its sights set on more modern, complex video games.
AlphaGO
Although this was a great feat, DeepMind wanted to see how the technology would fare against a worthy human counterpart. Enter AlphaGO. AlphaGO is AI optimized to master the game “Go”, an abstract Chinese board game that was developed over 3,000 years ago. The game is played one on one with the goal of occupying more territory than your opponent. This game was particularly interesting to Google as it works mostly on intuition. The game cannot be played mathematically like Chess as there are 10 to the power of 170 possible outcomes (that works out to more results than there are atoms in the known universe!). Lee Sedol, a legend of the game, was to face-off against AlphaGO. Throughout the game, AlphaGO made some peculiar decisions that wouldn’t typically be played by the human mind. This was extremely powerful as the algorithm was being creative and intuitive in those moments. AlphaGO would go on to beat Lee 4-1 and impress the media with this being DeepMind’s first big splash in the general public. There was a documentary developed on AlphaGO which you can find on most streaming services. DeepMind has recently introduced AlphaZero, an updated version that runs more effectively and on fewer processors. It now has the ability for one agent to teach itself Go and Chess simultaneously.
Real-Life Implications of DeepMind
Although mastering the game of Go won’t solve the world’s problems, DeepMind has its hopes set on impacting real-world issues, DeepMind recently launched AlphaFold which is the same AI used with AlphaGO, but in this case, it can teach itself to predict how certain proteins will fold. This is an essential learning curve in understanding protein structures which in turn can allow for more straightforward diagnosis of diseases such as Alzheimer’s. Also, DeepMind looks to use its technology to tackle climate change. DeepMind has recently been testing their algorithms in data centers and industrial buildings to find the optimal operating system to lower general energy consumption. Through their tests at Google data centres, they have been able to minimize energy consumption up to 40% lower than human optimization. They see this number increasing with more data and opportunities the algorithm utilizes.
Of course, with all of this exciting innovation, society has its critics. For one, Google has been criticized for putting such a significant investment into something that hasn’t had real-life implications yet. Mastering Atari video games and old Chinese board games are fascinating, but will their promises of helping healthcare and climate change be played out? Although they are in the beta phase, investors are becoming increasingly impatient with such a substantial investment by the company. Also, there is an ever-growing cloud over AI with the worry about the unknown outcome. How far is too far? As AI becomes more efficient and productive, it will begin to replace jobs typically occupied by human beings. The silver lining here is in the AI’s reliance on data. Jobs that require intuition, creativity and forward thinking are more likely to be safe from AI replacement. Experts believe careers in the arts, marketing, and psychology will thrive in the AI period.
Arguably the most significant concern in AI is how long until the technology realizes that the only thing standing in the way of their optimization is us? Eventually, we will be the ones slowing down AI, and the failsafe may not be strong enough to withhold the inevitability of their takeover. AlphaGO has already proved smarter than some of the greatest minds in humanity. How long until it outsmarts us completely? Google has recently put resources into an ethics board, but my guess is AI won’t be using ethics when trying to optimize to its full potential.