Introduction
The aim of this paper is to investigate cases of where inequality is happening in algorithm reliant technologies by the implementation of bias within its inputs, method and perception of output, concluding with alternative pursuits which address the issue. Firstly, psychological analysis of perceptions surrounding bias will be obtained to provide scientific reasoning as to why bias exists in our society allowing for its understanding in programmatic technologies such as machine learning. This will be followed by historical and current cases that have been highlighted academically to be a promotion of inequality within technology. Finally, analysis of pursuits to address the matter will be brought forward to propose possible solutions or differing methods in the case of biased algorithms.
Algorithms are “detailed instructions defining a computational process, which begins with an arbitrary input, and with instructions aimed at obtaining a result (or output) which is fully determined by the input.” (Uspenskii, 2019). Simply put, algorithms are a recipe; you need certain recipes to make certain foods. Yes, recipes can be tweaked to still produce good food, but if you miss a key ingredient or mix the ingredients in the wrong order, you often produce something that is rather unpalatable. Sometimes recipes can be followed perfectly and you might think the dish tastes delicious, but some people may not have the same pallet as you and may have a differing opinion of the result. The recipe (method or design), ingredients (input) and people in your restaurant (population or users) are all considerations that have to be taken into account if an algorithm is to be free of bias and accessible by all. But biases can find their way into our technologies, such as machine learning software, leading to individual lives being changed dramatically; from being discriminated against by face tracking software (Buolamwini and Gebru, 2018), having your job striped from you for no reason or ability to question why (O’Neil 2018 pp. 1-13) or deciding if you deserve to be watched and stopped by the police depending upon cards that you were dealt from birth (Lum and Isaac, 2016). Algorithms are the reasons for these events taking place, but algorithms are not all to blame. All systems are designed and programmed by humans and so causality should fall to them.
“The big message for people is to understand that computers aren’t born with bias. We think of them as this pure calculation device, but then we feed them with all of this biased data,”
- Matt May, head of inclusive design at Adobe.
Although we are quickly developing artificial intelligence (AI) to think for ourselves such as machine learning from companies like Google, who have their DeepMind department researching and building algorithms to solve our data problems of the future (DeepMind, 2019), we still need to be cautious laying the foundations of these technologies. Due to the rapid advancements of computing power illustrated by Moore’s law where the number of transistors on integrated circuit chips doubles every two years (Moore, 2006), if we do not formulate methods for regulating the input data of bias into these technologies, the foundations which we are laying now could compromise the greater structure in the future when even more data is being processed at faster rates. We could quickly find ourselves relying upon massively powerful systems with hard coded inherent biases to help us in our day to day lives, making it a much harder task to fix. Wachter-Boettcher (2017) describes how although bias in technology such as algorithms is not the end of the world, to a minority user in the eyes of a design team, small instances of discrimination can build up to become deep routed mental illness (Wachter-Boettcher, 2017 pp. 72). Organisations such as the Institute of Electrical and Electronics Engineers (IEEE) are now promoting awareness of the issue and writing standardisation documentation to educate the individuals responsible for the algorithms that process our data (Standards.ieee.org, 2019). But without the outcry of consumers who use the algorithms which promote inequality within the greater population, algorithms could continue along its current path of discrimination. Ansgar Koene a member of the IEE states that, “We’ve been trained to believe that humans are not neutral but assume computers just run through a series of processes so there is no reason to suspect the programs are biased, even though they’ve been programmed by people,” (Pretz, 2017) Koene alludes to going beyond the algorithm itself and focusing on the forces behind making these algorithms what they are. Wachter-Boettcher (2017 pp. 149) adds to this stating, “every digital product bears the fingerprints of its creators.”. Digital products should always be accountable by those who set the variables for the manipulation of our digital lives.
A key part of how an algorithm interacts with its users is its ability to be interpretable by those being affected by it. It’s okay to receive an output of an algorithm if it is to your favour such as how we profit from seeing content that the platform knows we will like on websites such as Facebook and YouTube. If it was not, individuals would want to know why this was happening. What separates a user having knowledge of why an algorithm outputs what it does, is all a result of its source being either open or closed. Open and closed refers to how the source code or, for our discussion, mathematical procedures in algorithms, formulate the input data to produce an output for a user. It is commonly acknowledged that sources should remain closed, commonly called a black box, if a system wishes to be as un-comprisable as possible in terms of its security (GOV.UK, 2019, Anderson, 2002), but if a system remains closed, transparency and accountability of the creator is not possible. The actual users themselves have no knowledge of the process for which the technology such as the algorithm, came to a decision.
For the case of algorithms it doesn't entirely matter if the source is open or closed, problems can still arise; teachers in Washington DC are losing their jobs based upon multiple withheld factors formulated by a closed source algorithm, giving them no ability to question the generation of their scores, however colleges and universities all over the US are trying to game an open source algorithm to compete against one another to be the highest ranked, all based upon the publicly available set of variables used by the algorithm; ultimately leading institutions to produce false positives to appease said algorithm (O’Neil, 2017 pp. 1-13, 50-68). Simply put: close algorithms provide no reasoning to their output and yet open algorithms promote gaming of systems to encourage proxies in said system. These flaws in the algorithmic models are combined with the growing public concern to data protection such as how big data companies such as Facebook who in partnership with Cambridge Analytica leaked data and ultimately broke GDPR legislation (Hern and Pegg, 2018). Academic interests have picked up the pieces where commercial businesses have failed by acting as the force against faceless algorithms that are designed to maximise profit at the expense of its users.
Judith Donath of Harvard Berkman Klein Centre for Internet & Society states, “The algorithm should not be the new authority; the goal should be to help people question authority.” (Rainie and Anderson, 2017) As previously mentioned, guidelines from recognised corporations such as the IEEE are starting to be published and recognised (Standards.ieee.org, 2019) but the lids on black box algorithms still remain shut with only improvements to data transparency and not the use of the data itself taking place in the example of Facebook's data breach (Facebook, 2018). Recent patent filings actually point to Facebook's algorithm becoming more intrusive to peoples lives with the platform wanting to find out who you share a room with (Rodgers, 2018) there is also scrutiny arising that content filtering algorithms used by the platform promote poor mental health and in some cases suicide (Crawford, 2019). Although academic opinion of the technology’s future is almost split down the middle (Rainie and Anderson, 2017), with such damaging cases ultimately affecting the population it can be hard to justify their use within future technologies, however algorithmic machine learning systems are not all negative.
Although human brains are finite, they are still 99% faster than the worlds largest supercomputers (Riken.jp, 2013). What computers do have over the brain is storage. One main advantage that algorithmic decision making has over human decision making is the amount of data each one can process instantaneously. Banking is an example of where data driven industries have thrived because of the implementations of data processing algorithms for example, allowing banks to measure investment risks more accurately and ultimately produce loans at a higher rate (Varetto, 1998). Banking has also recently seen the implementation of facial recognition software to provide a greater level of security to their clients accounts while at the same time improving said clients experiences with the banks products (Szczodrak and Czyzewski, 2017). Another area where algorithms thrive over traditional methods is within the medical industry. Algorithms provide medical workers with in depth analysis into patients possible needs and outcomes based upon historical and predictive data, allowing for more accurate diagnoses, quicker cycles of patients, higher success rates and cheaper functionality overall. This data is also shared amongst the medical research community where cures and treatments for conditions can be achieved at a faster and cheaper rate (Dey and Rautaray, 2014).