Death By Relevance
The Danger of Monoculture
The Internet is arguably the greatest invention in history. A global network of computers that has eclipsed the human brain in speed of calculation and capacity to store data. A system so vast it has effectively become a universe unto itself, immersive and complete.
Empowered with instant access to information, people in developed nations spend more than 7000 hours each year on the internet, the world’s information at their fingertips. But the words we’ve adopted to describe this activity, searching, surfing, exploring, are illusory. When we’re on the internet, we’re not traveling anywhere. We never move at all.
208.184.43.213
This was my IP address while I wrote most of this essay. During my research, I looked at articles, images and comments from people in over 27 countries spanning four decades, but I never traveled an inch. Everything I saw came from somewhere else, delivered to my computer by a search engine.
The phrase Search Engine conjures the image of a futuristic device capable of propulsion and locomotion. That was intentional — the term was coined by marketing executives selling computers. But the technology behind this doesn’t “search” in the same sense that a human being does when exploring a physical environment, subject to random interruptions, accidental encounters, and unintended juxtapositions of people, places, and events.
A search engine uses a string of data to rearrange the order of larger stores of data organized in massive, indexed tables. Essentially, it sorts spreadsheets. It doesn’t travel into the unknown. It always knows what it’s looking for because it has to. It never notices things outside the scope of a request because it can’t.
We don’t surf the internet. We sort it. We filter it. Like a huge warehouse filled with sand that we rearrange with a rake while we tell ourselves we’ve gone to the beach.
Q & A
Think about the phrase “Search Results.” There is no promise of magic in those words. They describe a mechanical output in response to a mechanical input. Because the human brain is wired to find meaning in patterns, without the technical understanding of what’s really happening inside a computer, we created a metaphor to frame this interaction in more understandable and relatable terms.
We ask a search engine a question, and we get an answer in return.
But entering information in the form of a sentence isn’t necessary for a search engine. “How much time does the average person spend on the internet?” delivers the same page of results on Google as, “average time person internet” or, “person time average internet.” The search engine does not answer us. It receives a string of data and delivers a result.
This distinction is subtle but important. Regardless of how we enter an input, our cultural awareness (“ask Google”) essentially defines this process as asking a question. By association we accept the results as an answer. This definition implies the search engine’s output is the result of a thought process, as if it considered our question and gave us what it thinks we asked for.
That’s not what’s happening at all.
The Algorithm
TF-IDF stands for, Term Frequency–Inverse Document Frequency. It was the first, and remains one of the most common, algorithms for filtering databases on the Internet. It is foundational to the mechanisms that present every single thing you see online.
The algorithm analyzes a document (or string of text) for keywords based on placement and frequency and parses them out.
Then it basically inverts itself to remove words that have extremely high frequency. These are often prepositions, definite articles and conjunctions.
This generally leaves nouns, verbs, and adjectives which are then ranked by prevalence and placement within the original document.
The resulting list becomes the basis for a search query or an index ranking in a database.
Modern search algorithms have become more complex over the years as engineers attempt to make them faster and more accurate, but the underlying mechanics of TF-IDF are still present in all of them.
The core function of discovery on the Internet is to match like things with like things.
The Search Within
Today, every day, we engage with far more sophisticated applications than a search engine. Social feeds, shopping sites, streaming platforms, news publishers, instant messengers. Applications that track our location, physical activity, and even our mood generate a constant source of data collected by browsers, televisions, automobiles, mobile phones, IOT devices, wearables and voice-actuated appliances.
Today, not just your search history, but your entire digital history, every action you take in response to a result, all of your behavior that can be tracked, is collected.
In a mind-bending twist, this constantly evolving record of your activity online has become a search string in and of itself. It becomes a frame for each new input in an attempt to further refine your results, and acts as a filter on your entire online experience, informing every new request with the limitations of your past.
- Every search result is affected by what you have already seen.
- All your social feeds are tuned to amplify the content and people you interact with and suppress what you ignore
- All the people you are prompted to connect to are screened through a list of those you already know
- All the ads you see are targeted by what you have already bought, seen, listened to, researched, or where you’ve physically been
This constantly sharpening lens of data creates an ever-tightening aperture on your view of the online world. The narrowing of this window is called: relevance.
Mirror, Mirror…
The mechanisms driving increased relevance in content delivery online are incredibly powerful, but anyone who’s been stalked by advertisements for pair of shoes they just bought knows that in practice their sophistication sometimes leaves a lot to be desired.
And this annoying phenomena is troubling if we consider it more deeply.
The underlying assumption of a relevance algorithm is that your observable behavior is the evidence of all you desire, and all you ever intend. The result surrounds you with the echo of all your previous actions, creating a virtual hall of mirrors.
There are many ways this technology can be extremely useful. Algorithms used to harness data and help machines learn from human behavior now power self-driving cars, increase energy efficiency, and help physicians diagnose rare diseases.
But not every activity benefits from being refined to pinpoint accuracy. When relevance algorithms become the default mechanism for delivering news and information, or defining the community of people we engage with, they present severe risks to our ability to function as a society.
Protection from Surprise
Confirmation bias is the human tendency to filter the world for information that supports our existing views. It’s a useful mechanism when it reduces the effort of questioning all our underlying assumptions when we’re focused on fulfilling basic, immediate needs. But recent events strongly suggest this trait, combined with the power of relevance algorithms, can amplify confirmation bias to a dangerous degree.
This experiment by the WSJ revealing the information bubbles on Facebook during the recent US election provides an example of how relevance algorithms affected voter’s exposure to political news. Similarly, a company called Cambridge Analytica engaged in the highly controversial practice of aggressively targeting individuals with a combination of fabricated news stories and political ads in an attempt to remove any information outside of the individual’s discerned bias.
In an effort to refine discovery, relevance algorithms are killing exploration.
And there is now good reason to believe this phenomena may be fostering an intolerance for discovery, and possibly having a negative effect on our ability to feel empathy. These factors, combined with the disinhibiting effect the Internet has on human civility, have created a psychological ecosystem with disturbing potential.
Monocultures
Our natural environment has evolved over billions of years into a vast ecology of interdependent systems. Animals, plants, and microbes constantly evolve and adapt to change, shifting and refreshing the environment. Their inherent diversity makes them flexible, resilient, and able to continually rebalance. When we reduce diversity in natural systems, that flexibility and resilience is changed.
Modern agriculture has used science and industrial mechanization to radically increase food production, and simultaneously reduced diversity in the ecosystem. Immense swaths of land are now strictly cultivated, and not just to a single type of plant — an entire crop can be limited to a single variant of DNA.
Large, homogeneous crops enable parasites — bacteria, viruses, fungi and insects — to specialize on one specific host, increasing the chance they will mutate into even more pathogenic forms. To improve efficiency gains (e.g. sharing the costs of chemical spraying and seed storage) farmers tend to choose the same crop cultivated by neighboring farms, further amplifying the risk of virulence.
Biologists have proven that once certain risk thresholds are passed the transition to mass extinction can be dangerously abrupt.
Relevance algorithms are machines that eliminate diversity by design, and there are compelling parallels between the effects of monoculture on the biology of an ecosystem, and reduced variety of stimulus on the neurology of the human mind.
Experiments in sensory deprivation have shown that when the brain is deprived of information, it goes into a kind of overdrive trying to create its own stimulus. The longer an individual is without stimulus, the greater the potential for negative effects, including hallucination and psychosis. These effects can damage the human psyche so severely that sensory deprivation has been recognized as an extreme form of torture.
Even lesser degrees of reduced stimulus can have harmful effects on the mind. The Ganzfield Effect is a phenomena that occurs when human beings are subjected to uniform stimulus for a long period of time. Individuals subjected to a simple lack of variation in stimulus can show significant cognitive impairment. Most disturbingly, specific symptoms of this effect can include hallucinations in the form of passionate beliefs based on fragmentary information or repeated suggestion.
It is impossible to attribute the violent behavior of any individual purely to their online experience, but it would also be irresponsible not to seriously consider the psychological effects of relevance targeting, and to explore alternatives when we structure media and content delivery systems.
Faith Will Not Save You
While the Internet has had many powerful, positive effects on the world, it has also created problems. And this is one problem it doesn’t yet have the power to solve.
The Internet isn’t good at creating diverse exposure to ideas that challenge existing biases because, at its core, that’s not what it was programmed to do.
Varied exposure to other cultures, new ideas and concepts, is an inherent need for the psychological health of individuals and societies. Living in a culturally and intellectually diverse environment strengthens our ability to formulate belief systems that are reasoned, tolerant, and compassionate.
If we accept the idea the Internet is a necessary fixture of our environment, one we rely on to orient ourselves in the modern world, then we need to develop a keen and conscious awareness of its benefits and dangers.
We need to train our behavior online, to prevent the the Internet from training us.
The Internet was built to make the delivery of information easier and faster, but delivery is not exploration. A fundamental behavior of all living things, exploration takes effort and desire. Reaching beyond what you know, beyond where you inhabit, requires commitment and most importantly, curiosity.
Until we can find a way to program the Internet to become a better tool for discovery, we have to make a conscious effort to program ourselves to be more human.
- Deliberately spend time considering information outside your recommended newsfeeds
- Use the incognito mode of your browser and see how it changes your news and search experience
- Change browsers and search engines entirely, and observe the differences in your experience
- Force yourself to read and consider alternative opinions, even ones you viscerally disagree with
- Try and project yourself into the physical and economic circumstances of someone with different perspectives
- Consider the motivations of authors, who may not be aligned with the needs of their intended audience
- Consciously exercise politeness when interacting with people you do not agree with
- Before you post or Tweet pretend you’re having dinner in each other’s houses, or speaking face to face
- Diversify your conversations with online friends — share ideas and information about subjects besides just politics
- If you share things that make you alarmed or angry, share just as many things that inspire and delight you
- Meet new people outside your existing network
- Join a meetup for something unusual
- Travel somewhere you know nothing about
- Explore new kinds of media
- Intentionally listen to a kind of music you do not like for a whole day
- Go to a movie you don’t think you’ll like with someone who does
- Read a book
- Deliberately put down your phone and stare out a window, or just observe the world around you, for at least five minutes a day
Stuart Hynson Culpepper is a director of media and software in NYC, he works at the intersection of media, digital interface and community.
