Blog / Technology & society

Algorithmic Bias: The Dark Side of the Digital Age

Pakorn AJ Leesakul 6 min read
A hand reaching into a sculpture of orange and black data particles
Photo: Google DeepMind / Unsplash

Equality that is only half true

“Digital technology will help create equality in the world.” This statement is only half true. The true half is that data networks let us watch films, buy clothes or use all kinds of services without leaving home, and children in remote areas can reach good knowledge without distance standing in their way.

The other, untrue half is that this vast space of data can create enormous inequality if the systems that process and curate information for us carry biases of their own — what is called “algorithmic bias.” This is the dark side of the digital world’s light, one that most people cannot see.

Bias hidden in the system

Algorithmic bias comes in at least three forms.

  1. Bias from ranking. Results at the top of Google are seen far more than those on later pages. Even alphabetical order, which looks neutral, gives airlines whose names start with A an advantage over those starting with B.
  2. Bias from assuming everyone works the same way. Plagiarism checkers more often flag work by students whose first language is not English, because native speakers are better at rephrasing so their writing doesn’t match the original.
  3. Self-reinforcing bias. The crime-prediction program PredPol, in Oakland, California, recommended sending more police into communities of colour based on statistics of reported crime. The Human Rights Data Analysis Group pointed out that this is a cycle of discrimination: the more police are sent, the more “crime” they find, and the more the area is seen as more dangerous than it really is.

Conversely, having no data at all is also a problem. If insurance decisions are based on accident records that never captured minorities or marginalised people, those people become invisible — no record means no existence. Easing these biases therefore takes citizens’ critical literacy, government policy, and data activism from civil society, such as data volunteers.

The Daily Me and the digital class

Online crowds gather around shared interests, shaped both by our own choices and by algorithms that decide what to show. The result is a “Daily Me” world with ourselves at the centre. Mr. A, who follows Party A, only hears about Parties B, C and D when they relate to Party A. Ms. B, who follows only BNK48, gets to know politicians only when they talk about her favourite group.

Online spaces should be public spaces for debate, but in practice extreme leaders use them to measure their popularity and harvest support. Conversations are therefore shaped by the “volume” of voices backing a leader rather than by substance that comes from public participation.

Digital platforms have also created a new elite known as the “digital class”: people with enough knowledge to build and steer technology. So the question is not only whether people can access the tools, but one of power — whether technology truly gives people more opportunities in life. The good news is that consumers are learning faster and faster. When membership cards like All Member became widespread, it didn’t take long for people to start questioning how hard they were to cancel and what they meant for privacy.

How our feed is arranged

An algorithm is not some mysterious technology. It is a set of problem-solving steps that can be explained clearly — like when your partner is running late and you have to order food for them. You think it through step by step: she prefers fish to hamburgers, dinner shouldn’t be greasy, and it shouldn’t repeat her last meal. In the end, you land on the dish you expect her to like most.

Facebook does the same with the enormous amount of content created every day, through a ranking system in three parts. The first is inventory: all the content eligible to appear in the feed. The second is signals, such as how recent a post is, who posted it, and whether it is spam or clickbait. The last is predictions: estimating which posts we will like, comment on, share, hide or report. The system combines all of this into a relevance score and orders the feed by that score.

Relationships are ranked too. Friends whose posts we like often appear more often, so the feed gradually narrows until only the people and news sources we’re interested in remain. That sounds like a good service, but it is a double-edged sword. If we engage with one-sided or fake news — say, that lime soda cures cancer — the feed will serve us more of the same, until we believe what we see is the whole truth.

When we ask and AI answers

An AI chatbot screen with an “Ask anything” prompt
Photo: Zulfugar Karimov / Unsplash

Today, many people no longer search on Google or scroll through a feed; they type a question into an AI chatbot and get a ready-made answer back. Ranking bias has therefore intensified. Where we once saw several search results to choose from, today there is just one answer summarised by AI — and many people never open the sources at all.

This kind of AI learns from vast amounts of text on the internet, so it picks up the biases in that data too. Viewpoints that many people have written about, and languages with lots of data such as English, tend to be reflected more, while people with few records become even more invisible. AI can also “make up” information that sounds convincing but isn’t true (hallucination), and tends to go along with assumptions the asker builds into the question (sycophancy). A leading question therefore tends to get a leading answer, turning AI into a personal echo chamber that can talk back.

The way to cope is to ask open questions without building the answer you want into them, ask for sources and actually open them, ask again from another angle or have the AI make the case for the other side, and treat AI’s answer as a starting point for thinking, not the conclusion.

From feed to conspiracy theory

A man with his face in his hand, lit by the glow of his phone in the dark
Photo: Adrian Swancar / Unsplash

Why do we keep turning to social media? One answer is that when offline life is satisfying, people tend to use social media less; but when life is hard, social media steps in to meet the needs of social animals. These people are the most vulnerable to “manufactured truths,” because those truths offer both answers and a place to belong.

Conspiracy theories reason backwards from effect to cause. Sound logic says “it rained, so the traffic is bad,” but a conspiracy theory says “the traffic is bad, so it must have just rained” — even though traffic can be bad for other reasons. In other words, the answer is set first and reasons are found to support it. When such stories are exciting and fed to us again and again through the feed, belief slowly takes shape, like water wearing away stone.

The scarier question is: what if those who feed content into our feeds deliberately create the falsehoods themselves? The Cambridge Analytica case, in which Facebook users’ data was used without consent to micro-target political ads during the 2016 US presidential election, is the answer. Analysing data is not wrong; using data obtained without consent to spread hatred and propaganda is. Users were left with only “passive participation,” and the claim that social media “brings people together” came under heavy scrutiny.

Who arranges our feed?

The world has changed a great deal, but people still look for things to fill what is missing, from love to technology. What is different is that today algorithms decide for us what we should see — and algorithms can always be biased.

Since we have already paid for social media with our personal data, the remaining question is whether we let the feed steer us, or build a feed we choose ourselves. We can start by following sources that see things differently from us, checking before we share, and remembering that what the feed shows is not the whole picture of the world.

Photo credits

  1. A hand reaching into a data sculpture (cover): Google DeepMind / Unsplash
  2. An AI chatbot screen: Zulfugar Karimov / Unsplash
  3. A man lit by his phone: Adrian Swancar / Unsplash

All photos are free to use under the Unsplash License.

← All posts