Introduction: The Illusion of Neutrality
Two individuals who are equally skilled, inquisitive, and ambitious open the same app, type the same search term, or post an almost identical video. One receives traction, accolades, and offers. The other? Nothing! Or worse, abuse.
She begins to question whether it's bad luck. Perhaps she used the wrong hashtags. Perhaps her resume didn't read well. Perhaps, it's her tone. Her face. Her self.
We assume that technology is the great equalizer. It is neutral and unemotional. But we often forget, it is data-driven. Thus, it frequently represents the same inequalities we have built, endured, or hoped to leave behind. Algorithms are only constructed by people. Thus, they absorb our assumptions and systemic biases. They can do all this quietly and at scale because they are hidden in the background.
So when the internet begins to feel a bit more unjust, particularly for women and minorities, it's because the algorithm may be kind of sexist, like us.
Algorithms are Trained on Biased Data
An algorithm is only as good as what it's been trained on. If the data with which it's trained reinforces the biases of society, then the algorithm learns to replicate them. In 2018, MIT did a study that determined facial recognition algorithms were incorrect 34.7% of the time when identifying dark-skinned women and 0.8% of the time when identifying light-skinned men. Why? Because the datasets upon which the algorithms were trained were whiter and more male.
Search engines, too, reproduce stereotypes. A quick search for terms like CEO or nurse shows disproportionate gender representation. You can't blame AI! The results are nothing but learned behavior from patterns in data shared with AI; this is the same data that we have generated over time. However, it is important to note that this reinforces existing norms instead of challenging them.
Even translation software has been found to gender-stereotype occupations. Translate "he is a nurse" and "she is a doctor" from English to a sexless language like Turkish, and vice versa, you'll often find the genders swapped. High time we start to reflect!
Gender Bias in Social Media Algorithms
Some female creators either feel invisible or overexposed because social media platforms promote content with prominent aesthetics or behaviors. This eventually neglects creators who don’t fit those norms.
A report by Mozilla published in 2021 discovered that the YouTube algorithm pushes misogynistic content to female audiences even though they have never searched for it. Nevertheless, creators who speak about gender topics have their videos demonetized or downranked. The YouTube algorithm is not inherently sexist, but it becomes sexist due to user behavior and feedback loops.
TikTok has also been blamed for demoting posts from ‘less desirable’ users, which disproportionately affects women, the specially abled, and creators of color. Women who make feminist or political comments are far more likely to be harassed, which the algorithm fails to filter out.
This builds a lethal double bind: be palatable or be unseen.
Biased AI in Hiring and Professional Tools
Technology should make hiring and productivity easier, but it has the effect of filtering out candidates who do not fit the guidelines of the ‘perfect’ employee. Amazon scrapped an AI recruitment tool in 2018 after it persistently demoted resumes with the word "women's".
Other AIs employ algorithms favoring candidates from past hiring records, biasing men toward promotion to tech and managerial roles. All the data fed is backed by human behaviour. Hence, it is sadly shocking that AI screening sometimes excludes resumes of qualified women by misreading gaps in work history (e.g., maternity leaves) as red flags.
Algorithms and Online Safety for Women
Content moderation is another area where women are affected by algorithms. Posts that are genuine abuse or harassment are bypassed unless they contain specific keywords. This leaves women, particularly women of color or LGBTQ+ identities, open to serious online abuse.
In addition, the sites over-correct at times, banning or shadow-banning activists and educators simply talking about sexual health or gender.
Automated techniques can't always distinguish harm from education, so subtlety is lost.
A failure of cultural context in moderation software may flag slang or activism as ‘offensive’, but not recognize actually toxic content. The consequence? Harm goes up, advocacy is silenced.
The safety concern is systemic. When women's speech online is defended less regularly, it sets the tone for the whole digital conversation.
Reclaiming the Code: What Can Be Done?
Awareness is where it starts! Some technology companies are already addressing the problem. Varied development teams, open algorithm creation, and multiple datasets can lead to more balanced outcomes.
We can also demand algorithmic audits, external scrutiny of how systems impact different groups. Legislation like the EU's Digital Services Act is a step in the right direction. But there must be pressure from citizens, too. Technology won't change unless people push it to change.
Teachers, artists, and scientists are coming together for more responsible AI. From community-developed datasets and machine learning algorithms to focusing on fairness, all work together to shape AI better. Initiatives like Google's "AI Principles" and tools like IBM's Fairness 360 are on this track, but regulation must continue.
More diverse voices in tech and more people questioning how algorithms work can actually help us reshape the internet. Calling out algorithm bias isn’t being paranoid. It’s how we build a better digital future.
Conclusion
Algorithms mirror the world. And if the world is racist, sexist, ageist, then so are the ones we design. But that also makes us powerful. We can resist the code, question the outcome, and demand that technology benefits all, not just the default.
Tech bias isn't a bug. It's a human error that is replicated by AI through fed data. If we wish to change the algorithm's sexist behaviour, we need to change our behavior and rewrite the data being fed to it one code at a time.
