Understanding Structures of Oppression

Structures of oppression or systemic oppression are defined as “historical and organised patterns of mistreatment”. It is any system which works to alienate certain members of society from access to resources, privileges and opportunities that would otherwise be available to the rest of society. Such systems, while clear in their aim, are amorphous in the way they manifest themselves. Let us take the example of sexism. Having a wage gap for employees based on their gender, judging someone for not appearing ‘traditionally feminine’ or making inappropriate jokes about the rightful place of a woman are all examples of sexist behaviour. While the goal of each instance remains the same, i.e., the reinforcement of the belief that women are inferior, the method vastly differs based on the situation. Similarly, every structure of oppression hinders different members of society based on the social context it presents itself in. This also influences the level and type of harm inflicted on its victims.  

Below is a short list of different types of systemic oppression (in alphabetical order). It is important to note that this list is non-exhaustive. Furthermore, while each structure is explained separately, multiple systems of oppression often overlap and influence each other to create convoluted webs of multidimensional discriminatory experiences. In other words, many types of oppression often come together to create different types of discriminatory experiences. Therefore, just because they appear individually does not mean they are separate from each other as will be illustrated in the case study. All oppression is connected.  

Case Study: Over-policing in Skid Row due to predictive policing algorithms 

PredPol is a predictive policing algorithm that was used in Los Angeles to predict future crime. To the LAPD and PredPol’s developers, PredPol was an opportunity to conduct proactive policing and patrol ‘hot spots’ (areas with the highest expected crime rate) within the city. But to the residents of the communities situated in and around the hot spots? PredPol was nothing but a data-driven initiative to “contain, control and criminalise” the communities based on the highly criticised ‘broken windows policing’ strategy.  

Skid Row is an area that lies in the middle of downtown Los Angeles. By its residents, Skid Row is described as “a vibrant community of poor and predominantly Black, migrant, Indigenous, and disabled people helping each other survive”. By the LAPD and PredPol? A hotspot with guaranteed trouble.  

What came first: the chicken or the egg? Similarly, what creates the atmosphere of supposed danger around Skid Row? Actual crime or predicted crime?  

PredPol’s crime predictions were based on three elements: crime type, crime location and crime timestamp. This data is generated from a mix of police-filed crime reports as well as 911 calls, both of which could be unsubstantiated. The issue is that PredPol continuously collects faulty data and automates the same results. When police patrols target certain areas, they also collect more data in the form of crime reports and arrests. This very data is then fed into the PredPol algorithm whose consequent predictions lead them back to the same area, leading to the community feeling overpoliced.  

The problem with predictive policing algorithms like PredPol is that they are based on biased policing strategies that tend to over-criminalise certain areas, thereby creating an atmosphere of distrust in the communities that law enforcers are also supposed to serve and protect. As a result, ordinary residents may be treated as suspects based on vague suspect descriptions, leading to excessive stops, racial profiling, and violence or abuse. Incidents have included the use of excessive force and even sexual misconduct during routine checks. These practices disproportionately affect Black and brown individuals, disabled people, and other marginalised groups, often with tragic consequences. 

The problems above are not one dimensional. They are the product of multiple years of systemic oppression which places its residents on the losing side of a power imbalance. This makes Skid Row vulnerable, therefore an easy target for initiatives such as PredPol whose use of data and algorithms ensure that SkidRow residents remain punching bags for occurrences such as unprecedented stop and frisks that are driven by multiple oppressive structures.  

Systemic racism plays a key role in the disproportionate targeting of Black and brown individuals during police stops. Gender-based violence, particularly during suspicion checks, reflects broader patterns of sexism within law enforcement practices. Similarly, ableism and saneism contribute to the fear and stigmatisation of disabled and mentally ill individuals, often escalating into deadly use of force. Classism is evident in the treatment of areas like Skid Row, which has long been treated as a containment zone for the urban poor. Capitalist interests further exacerbate this by promoting the criminalisation and displacement of unhoused populations to make way for profitable redevelopment in the Los Angeles city centre. 

However, these systems do not act alone. For example, it is the combination of capitalism, classism and racism which makes it difficult for people from the community to invest into their own land without having their storefronts claimed to be a front for gang activity by the police and becoming a hotspot for patrols, stops, arrests and even unjustified killings.  

It is the influence of multiple systems of oppression that creates the data fed to predictive policing algorithms which reinforce the one-sided story of Skid Row. 

Moving towards detangling

Skid Row is not the first and it will certainly not be the last community that is a victim of systemic oppression. Predictive policing algorithms are not the only technologies to fall prey and become predators at the hands of structures of oppression. Hiring algorithms that skip over certain marginalised individuals reproduce racism, sexismand ableism. Neuro-prosthetic companies which supply sensory implants to low-income individuals also subject its users to dependency on their devices, thereby pursuing ableism. By forcing them to buy new parts with each upgrade, companies force the poor to fall victim to capitalism. Dismantled policies about online speech conduct in tech-companies which consequently allow online hate speech to flourish, disproportionately target members of the LGBTQIA+ community, codify cisnormativity and heteronormativity.  

All these examples show that the social context through which we observe systemic oppression changes not only the impact of the type of oppression but also how we view a certain issue. Thus, the very nature of the problems we try to solve are intersectional. When we look at oppression as a multi-dimensional problem that is context dependent, we realise two things. Firstly, there is no ‘one (algorithmic) solution’ to solve the problem of systemic oppression. Secondly, due to the sheer size and nature of the issue, the pursuit of a solution cannot entirely be the responsibility of one domain.  

As people who increasingly work with AI, it is imperative to assess the impact of the technology we build. This means thoroughly assessing the pros and cons of the models we design. But we cannot and should not do this alone. It is important to consult domain experts to understand the harms the AI model can perpetrate and brainstorm collectively on how this can be mitigated. When we design our technology and collaborate with the intention of co-liberation, we can collectively move towards detangling the web of oppression. 

*All sources used in the article are linked below 

Actionable Recommendations 

• Thoroughly examine the role your AI technology will play and ground it in societal context to ensure a realistic representation and understanding of its impact. It is also useful to redefine concepts you aim to achieve (e.g., fairness, transparency, accountability) with power and social context in mind. This allows you to align your priorities and remind yourself of the greater purpose of your AI product.  
• Try to incorporate viewpoints from various disciplines to gain multiple perspectives in understanding the problem and setting goals. Insist on iterative interdisciplinary collaboration throughout the AI lifecycle.  

Next Article ->