This article contains spoilers for Rumu.
Computers being able to beat humans at chess is now old news. Chess has a rigid set of rules with clear good and bad moves for every situation, so in that sense it was always a matter of time. But in recent years, due in large part to something called deep reinforcement learning, computers have been able to best humans at some of the most complicated games ever invented – which many considered, only a few years ago, impossible for a machine.
Deep reinforcement learning is a catch-all term for some of the latest algorithms used to teach AI how to play games. It has gone from mastering Atari games to defeating a Go champion and even to besting some of the most talented pro Dota players at last year’s International. In the case of Go, the winning turn was an unprecedented and original move. It shocked everyone in attendance and was such an upset it resulted in the challenger, Lee Sedol, having to literally leave the room.
A new Go strategy, or just straight-up psychological warfare?
OpenAI first ever to defeat world's best players in competitive eSports. Vastly more complex than traditional board games like chess & Go.
— Elon Musk (@elonmusk) August 12, 2017
These huge advancements in AI are of course not exclusive to the realm of video games. Personal electronic assistants and voice activated, smart appliances are becoming ever more sophisticated. As the technology grows and becomes cheaper there are many people who will inevitably find themselves made redundant by machines. The subject is becoming more and more mainstream with even local politicians, such as Alan Mak, utilising the subject (no matter how uninformed they are) to push a narrative. The conversation around AI and its representation in all forms of media, be it films, games or books is ever growing and is damn near inescapable. Advances in motion capture and facial recognition, even at a consumer level has reached the point that last year we even saw a “virtual YouTuber” rise to prominence. The subject of AI is everywhere.
This is just the start to the AI quagmire we are entering. One issue that is still not often addressed is how people feel around machines. If machines looked like the above, segmented expressionless terminators, what are the chances you will comfortably share your home with them? Take last year's viral AI, Sophia, the first machine to be granted citizenship. While Sophia is an undeniable marvel of technology she, much like Henry Cavill’s human Shrek face, fall into the uncanny valley. If nothing else Sophia showed that the question is not can we make machines recognise and emulate emotions – but when should we?
Which leads us nicely into the debut title of studio Robot House. Rumu is an adventure game where you play the eponymous roomba who is only able to love or not-love. As a sentient hoover, your one objective is to clean. Why exactly a roomba would need to feel and express love or any emotion at all is one of the questions the game explores, as it showcases a range of future hurdles that will need to be overcome in the relationships humans have with machines. The other main character is a 'smart' home system named Sabrina, who helps direct Rumu to the next mess and provides companionship in the otherwise empty home of David and Cecily.
Your time in Rumu mainly consists of puzzle solving/password finding, routing power to other smart machines and, of course, cleaning. While things are clearly not as they seem from the offset, Rumu keeps its cards close to its chest, constantly keeping the player guessing as to what is going on. Are you dealing with a Skynet situation here? Or is there an incoming Roy Batty revelation?
As you explore the game you discover the existence of Sabrina, your creator’s daughter who shares the same name as the home AI system. As Sabrina grew up around the seemingly sentient machines, she rejected their existence and went to a great deal of effort to hide from her parent’s creations. Sabrina’s bedroom and even her life goals are a clear rejection of her parent’s way of life, until her revulsion for the machines became too much and eventually she ran away from home.
In response to this, David and Cecily threw themselves back into their work, trying to create a machine that their daughter could accept, one that she could connect and live with: hence, emotional appliances. Rumu’s rounded and “cute” design would not be out of place in a Disney movie, the masters of anthropomorphic characters. Humanising machines is not exclusive to video games: call a speaker Alexa, give it a female voice, and many people will speak to it as they would a human. Saying please, thank you and even going so far as to apologise for shouting. I’ve personally had a slightly different experience with Google Home; using an iconic brand name is a great way to dehumanise something. Even the aforementioned “virtual YouTuber”, while following the storyline of being a sentient super AI, is designed to be appealing to certain people.
An interesting thing about the concept of imbuing AI with human-like emotion is that it could help mitigate the problems inherent with goal-assigned AI. In Life 3.0, Max Tegmark goes into great detail about the potential dangers of assigning an intelligent machine a singular goal. For example: Rumu is an intelligent roomba with the direct goal of “clean” and, while considering the potential obstacles it may face, a plausible conclusion to draw would be that the biggest issue is the possibility of deactivation. The roomba would thus create the subgoal of avoiding deactivation, and a possible further subgoal of arming itself to help accomplish this. If you were then recognised as someone who may seek to deactivate it, you'd be stuck in a house with a hostile roomba. You don’t need to have watched Black Mirror to see where that’s going.
Tegmark goes on to state however, that we humans have been able to supersede our genetical goal (replication) thanks to a myriad of rules our evolution has created to help procreate: our feelings of hunger, love, pain, fear etc. These feelings can now have an adverse impact on creating sequels. “We're loyal only to our feelings.” While Tegmark does not specifically state so, and in fact repeatedly warns against anthropomorphising AI, implementing human-like emotions could be a tool for keeping human and machine aligned. Yoko Taro’s NieR: Automata did this very thing: androids are infused with an all-encompassing love and longing for humanity, and on a smaller scale Rumu too is built around this concept. You encounter other emotionally endowed appliances, some of which can 'feel' only one or two emotions.
But this raises another concern, much like how humans are no longer bound by the biological goal of reproduction. If granted emotion, perhaps machines could learn and grow to find their assigned goals banal and uninspiring. Which would leave us, as in the knife-wielding roomba example, ever-so-slightly fucked. Current industry consensus seems to be that understanding and teaching AI intrinsic human values — which is about as easy as it sounds — is the safest route to take.
As you play Rumu and become more invested in the characters, your personal position and emotions parallel with Rumu’s. You start to question if the roomba is sharing the same feelings of doubt and fear as you, a feeling smartly backed up by the evolving dialogue choices. David and Cecily’s belief as engineers is that emotions can be specifically cut and assigned individually. But maybe they are wrong. The other appliances, and Rumu itself, are limited not in what they are able to feel, but what they are able to express. Giulio Tononi’s Integrated Information Theory (IIT) argues that consciousness is a singular thing, and experience cannot be separated-out or split into individual components. Perhaps emotions work the same way.
The conversation around machines and our relationship with them is only going to grow and may come to dominate many spheres, not least entertainment and politics. According to experts the next apparent step will be the possible enforcement of driverless cars: with around 1.25 million deaths per year attributed to road accidents, there’s no question as to why. The conversation is also moving onto military issues and the potential dangers of terrorism. Professor Stuart Russell (one of the world's leading authorities on AI) and the Future of Life Institute also released a short film showcasing the very real dangers of autonomous weapons. Constructing an informed opinion is becoming ever-more important.
These are weighty questions raised by what, at first glance, seems to be a rudimentary point-and-click puzzle game. But Rumu’s beautiful art direction, moving soundtrack and Allegra Clark’s evocative performance as the Sabrina AI makes this a singular experience. Instead of wondering whether machines dream, Rumu asks if they should.