Sudden and uncontrollable violence makes earthquakes one of the most frightening natural disasters produced on earth. Mexicans were reminded of the fact on September 19, 2017 when an earthquake measuring 7.1° on Richter scale levelled 44 buildings in Mexico City (Jacobo, Windsor, Nestel). Movements so overwhelming that one wonders if it's a humble reminder from Tepeyóllotl: the prehispanic Aztec god of mountains, earthquakes and jaguars. His name translates to “heart of the mountain”, referring to the sound the earth makes as it quakes. A sound so frightening that ancient Aztecs related it to the jaguar, one of the most deadly predators of the jungle. (Oliver p.109)
Representation of Tepeyóllotl (Oliver p.103)
During the following days a blanket of dense silence covered the city and its surroundings. In part because rescuers relied on their hearing to locate people trapped between the rubble. Residents also noticed that bells stopped ringing. The seismic activity compromised the structural integrity of many church towers. Through the symbols, noise and silence of this devastating earthquake, we explore the complex nature of notification systems and their relation to society. It became clear how the evolution of communication technology has augmented our capabilities, transforming the way communities organize.
The purpose of notification systems is to deliver a message to a set of recipients given a certain condition is met. The trigger that initiates the transmission and the medium through which the message is delivered gives us different abilities to communicate information (Santiago). For instance: the seismic alarm system depends on an array of sensors in order to trigger the notification before the waves hit densely populated areas. When the intensity of the wave surpasses a set threshold, it triggers a set of speakers around the city to blast an ominous warning siren. Residents are suddenly notified that the building they are in might collapse in the following minute, urging them to evacuate and get to safety. These traumatic events have d developed a collective PTSD triggered by any sound that's remotely similar. Day or night, holiday or working day, at any moment Tepeyóllotl might make his presence felt. This life saving notification system is made possible only through the technology that powers the sensors and sends information in real time; capabilities that were impossible before 1991 (Allen). Unfortunately, this system wasn't available back in 1985 when a violent earthquake of 8.1°on Richter scale hit Mexico City that collapsed more than 400 buildings, killing about 10,000 people (“Mexico City earthquake of 1985”). Through this example, we see how triggers, messages, mediums and recipients work together to notify and organize communities with the goal of taking action. In this case, taking action means running out of a collapsing building in order to save a life.
The ruins of Hotel Regis in Mexico City after the 1985 earthquake. (Derrick Ceyrac / AFP / Getty Images)
Tepoztlán was one of the towns that was most heavily affected by the earthquake. Located about one hour from the capital this quaint and mystical town transports you back in time. It enjoys one of the best climates in the world, with narrow stone streets and historical landmarks. The town is guarded by a tall range of chiselled mountains called El Tepozteco. An archaeological site with a couple of pre-hispanic temples adorn the top of the range, inviting tourists and pilgrims as far as Guatemala since pre-hispanic times. The structures and bridges to get up the mountain were so damaged that access was prohibited for months after the earthquake. The earthquake also damaged the churches located within the town. This forced residents to close the facilities and adapt makeshift prayer spaces while their structural integrity was restored. A town that was once vibrant was now still and quiet. Void of tourists and bell sounds, the residents were overtaken by a disorienting sensation. It is through this silence that we understood the profound relationship between the town and its bells.
The Ex-Convent Museum, also damaged, remained closed for the following months. During this time the museum's collaborators were tasked with documenting through interviews the feelings of the town residents during this dramatic moment. In this valuable recorded oral history, the sentiment that predominated was the emptiness, sadness, and the disorienting feeling of warped time caused by the absence of bell notifications. It became evident that the bells were much more to the town than metallic artifacts. They have an active role in their community as an agent that validates ceremonies, notifies the town of important events and keeps track of time (Mariano). Now, residents would rely on Facebook posts and word of mouth. The church bells are one of the driving forces of the town, providing punctuation and rhythm to everyday life.
When the town bells stopped ringing, other bells took over: cell phone notifications. In a way mobile devices replaced certain functions that the bell towers previously held in the town of Tepoztlan. Residents of the town became more active in social media, creating groups with fellow community members in order to exchange information. People started relying more on their phones in order to know the time. This mobile device that comfortably fits in the palm of your hand is the most powerful tool at connecting and communicating with people all over the world in the blink of an eye. In my opinion, it enables the most revolutionary features for notification systems: feedback and optimization (“What is Customer Experience Optimization? Up-to-date Guide.”). Mobile devices allow for a more complex and customized style of communication with its users. Each device is matched to a specific person, and we can track that person's behavior through the use of their phone. We used to think that all of this information created as a byproduct was some sort of waste, a type of data exhaust (“How to Turn ‘Data Exhaust’ into a Competitive Edge”). We are now learning that it is extremely valuable given that it is used to analyze, categorize and predict user behaviour. By providing feedback about that person's behaviour, the notification system can learn and improve how and when it sends its notifications in order to maximize the probability that the user takes a specific action. For the first time in human history, we have developed a digital information infrastructure capable of tracking large scale user behaviour, and subjecting it to statistically-powered hypothesis validation methodologies based on large scale data recollection. In essence, we are building the most effective way to drive people to take action. The question is: Who benefits?
The new digital notification systems have two additional components that enable the bi-directional relationship between the notification and its recipients. First, there is a piece of infrastructure that tracks, stores and shares information about users. This information is utilized to profile you based on your behavioural similarity to other users on the platform. By using statistical and machine-learning methods, we can infer additional attributes about your personality, providing more information about your user profile. With this data enrichment and augmentation processes we can build more precise models that predict your future behaviour (Park).
Inferring additional information through statistical methods represents an interesting debate regarding online privacy. Is it ok that Facebook knows my political or sexual orientation even though I haven't explicitly shared this information with the platform? What happens if there are more nefarious motives behind the people that capture and control this information? The same technology used to sell products more effectively through Instagram is fueling mass surveillance of autocratic governments. As users become more aware about how their information is being used to profit and monitor, online privacy has become so relevant that it is the focus of Apple's new ad campaign (Hollister).
Second, you need a predictive machine learning model that analyzes, learns and predicts user behaviour given the set of features available. This statistical algorithm is trained to maximize for a single variable given the business context. Some product brands focus on increasing customer lifetime value while media outlets optimize for engagement and watch time. Then, it will analyze all available features in relation to the given goal in order to identify the importance of each in order to predict future user behaviour (Santhanam). With this statistical prediction, the algorithm can modify the trigger, message and medium according to each individual in order to maximize the probability involving that user taking a key desired action. These statistical methods can go even further, modifying the information fed to a user in order to increase the likelihood that they will behave more in line to the model, not only predicting but effectively causing future actions. The implications that these nudges have on free will are certainly a matter of critical debate. Are we buying a product because we want it, or because it was suggested?
This technology has successfully been deployed by companies like Amazon to generate many billions in revenue. It is this prediction of future behaviour that is being commoditized through online advertising. We are essentially buying the probability of a person taking a desired action (Naughton). In my opinion, this is the most important evolution that advertising has seen in the past thirty years. We have become so successful at measuring, predicting and monetizing human behaviour that digital advertising has built some of the most profitable businesses of the modern age like Facebook, Google, Twitter and Amazon. We have moved from an industry that captures, packages and sells attention in the form of ad impressions, to an industry that sells predictions of future behaviour.
The implementation of these modern notification systems hasn’t been all peaches and cream. Rather, there has been some blowback from governments, regulators, users and software manufacturers alike. We have witnessed second and third order consequences of this technology that has had harmful effects on society. A common malady often arises given the reductionist approach of behavioural statistical models that don’t take into account additional factors. The human psyche cannot be completely reduced to numerical models. Take for instance the adverse effects YouTube video recommendation models have had on the contemporary political landscape.
Every time you finish a YouTube video, a semi-transparent timer appears on the playback window notifying you that the next recommended video will start soon. This is one of the most crucial mechanisms on the largest video platform, given that these types of video plays account for 70% of the content consumed. This means that the greater part of the content being watched on the platform relies on recommendations, and not on the user's explicit desire to watch those specific videos. The video that will be played next has been selected out of millions of possible files. It has the highest statistical probability that it will maximize your watch time on the platform. Some of the most brilliant experts on artificial intelligence have created such an incredible recommendation system, that some users experience a suspension of time and space, only to be drawn in by each new video. They call this the rabbit hole. Because the explicit goal is to increase the watch time per session at any cost, the algorithm will favour more controversial content given it creates more engagement. This is the same mechanism that spreads Fake News faster than real news. YouTube users end up consuming videos about conspiracies, radical political views and controversial themes. The bluntness of this recommendation algorithm has resulted in greater political radicalization, mistrust of institutions and the spread of conspiracies such as QAnon. Amplifying harmful content just because it meets our quantifiable goals seems counter-productive. Without wider considerations on how optimizing for watch time can have an effect on political views, mental health, and culture, we run the risk of deploying these technologies for the worse (Roose).
Another critical aspect of these predictions is what happens if we get it wrong? Remember that we are dealing in the realm of statistical probabilities that operate on a margin of error. Real errors have real consequences. Imagine a government agency collaborates with Facebook in order to try and identify terrorists or terrorist supporters. The answer these tools give us are expressed as: people in this group have similar characteristics and are twice as likely to be terrorists compared to the general population. What happens if an innocent user is classified as a false positive, a potential terrorist. What happens if we aren't able to detect a terrorist in time and a bomb goes off? Or what happens if a facial recognition software identifies innocent people as criminals? (Porter) It is my opinion that the impact of these decisions span way beyond the capacity of data scientists. We need to involve a wider variety of stakeholders in the design and implement more inclusive and thoughtful predictive notification systems.
At its core, notification systems are designed to transmit information. They can be the trigger of our collective trauma, designed to warn us of imminent danger; or have an active role in the community as a symbol of time and the keeper of sanctity; or take on a more proactive role as a predictive behaviour mechanism. Life used to be simpler when our notification systems didn’t have extensive Terms and Conditions attached to it. Bells just rung. But with these new powers, come greater responsibilities. The reality is that these digital communications systems are so new that we currently have no idea on how to best integrate them in society. We need accurate recommendation systems to help us navigate the petabytes of ever-growing information. Netflix would be a lot less useful without a recommendation system. More relevant content means a better user experience. Given how perversely the incentives are aligned between platforms, users and brands, the government has to regulate and enforce how this information is used. How can we have relevant and personalized information without harmful exploitation? Users have to be more conscious and demanding of governments and corporations on how their data is accessed and used. Some advances in blockchain information management and government data regulations hold some promise, but it is just the start. Privacy should also be a universal right, not a luxury. Currently only iOS users are better protected than Android users through their new opt-in tracking feature. There should be some type of incentive for people to share their data consensually given that only 4% of users do so willingly when asked to opt-in (Lovejoy). As economic interests, privacy rights, and consumers enter into conflicts, regulators, companies and citizens are asked to solve new kinds of complex problems. Only through trial, error and patience shall we strike a productive balance amongst stakeholders. It is expected that we will break, and fix, many things along the way.
Works Cited
Allen, R. M. (2019, January 11). Lessons from Mexico's Earthquake Early Warning System. Eos. https://eos.org/features/lessons-from-mexicos-earthquake-early-warning-system.
Encyclopedia Britannica, inc. (2008). Mexico City earthquake of 1985. Encyclopedia Britannica. https://www.britannica.com/event/Mexico-City-earthquake-of-1985.
Guilhem, O. (1998). Tepeyóllotl, "Heart of the Mountain" and "Lord of the Eco" , the Jaguar God of Ancient Mexicans. Náhuatl Culture Studies.
Hollister, S. (2021, May 20). Apple's new privacy ad imagines Thanos-snapping your stalkers out of existence. The Verge. https://www.theverge.com/2021/5/20/22446220/apple-privacy-ad-video-ad-tracking-transparency.
Jacobo, J., Winsor, M., & Nestel, M. L. (2017, September 20). Over 230 dead in Mexico quake as rescuers desperately search for survivors. ABC News. https://abcnews.go.com/International/200-dead-magnitude-71-earthquake-strikes-mexico/story?id=49958650.
Lovejoy, B. (2021, May 7). Opting in to App Tracking? Only 4% of Americans are doing so. 9to5Mac. https://9to5mac.com/2021/05/07/opting-in-to-app-tracking/.
Mariano, I. (2018, September 23). Callaron las campanas en Tepoztlán tras el 19S. El Sol de Cuernavaca | Noticias Locales, Policiacas, sobre México, Morelos y el Mundo. https://www.elsoldecuernavaca.com.mx/local/callaron-las-campanas-en-tepoztlan-tras-el-19s-2016567.html.
Naughton, J. (2019, January 20). 'The goal is to automate us': welcome to the age of surveillance capitalism. The Guardian. https://www.theguardian.com/technology/2019/jan/20/shoshana-zuboff-age-of-surveillance-capitalism-google-facebook.
Santhanam, N. (2019, October 14). Explain your machine learning with feature importance. Medium. https://towardsdatascience.com/explain-your-machine-learning-with-feature-importance-774cd72abe.
Park, A. (2015, January 12). Here's Proof That Facebook Knows You Better Than Your Friends. Time. https://time.com/3663775/facebook-likes-personality/.
Porter, J. (2020, June 24). A black man was wrongfully arrested because of facial recognition. The Verge. https://www.theverge.com/2020/6/24/21301759/facial-recognition-detroit-police-wrongful-arrest-robert-williams-artificial-intelligence.
Roose, K. (2020). The Rabbit Hole. The Rabbit Hole Podcast. https://rabbitpodcast.com/.
Santiago, S. V. (2020, June 17). A Comprehensive Guide to Notification Design. Toptal Design Blog. https://www.toptal.com/designers/ux/notification-design.
Wharton University of Pennsylvania . (2018, March 1). How to Turn 'Data Exhaust' into a Competitive Edge. Knowledge@Wharton. https://knowledge.wharton.upenn.edu/article/turn-iot-data-exhaust-next-competitive-advantage/.
What is Customer Experience Optimization? Up-to-date Guide. VWO. (2021, May 28). https://vwo.com/customer-experience-optimization/.