An algorithm is a defined, ordered and finite set of simple operations that can be used to find the solution to a problem. Each algorithm begins in an initial state with a set of specific data, and during the process each step is clear and shows no ambiguity. If the algorithms are deployed in combination with data or signals from their environment to make decisions independently from them and learn from them, this is called artificial intelligence (AI). This mimics the thinking capabilities of humans (Ismail, 2018).
Today, AI is widely used in various sectors of our information society. Unfortunately, the algorithms often contain pre-programmed biases. This is mainly because data is used that is not representative of different population groups in society. If data from certain groups is not entered, then the system cannot recognize it. The creators who develop the systems can, by consciously or unconsciously entering certain data, influence the system, which can result in discrimination and racism (Chambers, 2021). Examples include that communities of color pay 30% more for auto insurance premiums than white communities with similar accident costs. In addition, it was found that in New York, black and Hispanic students were half as likely to be admitted to top schools as white and Asian students. Furthermore, at some universities, black students were labeled “high risk” up to four times more often than white students. Moreover, black defendants are as much as 77% more likely to receive a higher risk score than white defendants in the criminal justice system (Public Citizen, 2021).
To change this, there need to be transparency and accountability in artificial intelligence (AI) from companies and the government. It is important to track why certain decisions were made, this could be to gender, socioeconomic status, immigration status, ethnicity, nationality, sexuality, ability, and more. The focus should be more on the sociocultural content in which concepts such as ethics, privacy and fairness appear. Furthermore, the creators of the algorithms should have diverse backgrounds and the data used should be representative of all populations. It is extremely important that this change will happen, since artificial intelligence has the potential to have an even greater impact on our daily lives (Richters & Haasdijk, 2022).
References
Public Citizen (17 August 2021). Report: Algorithms Are Worsening Racism, Bias, Discrimination [Online]. Retrieved from https://www.citizen.org/news/report-algorithms-are-worsening-racism-bias-discrimination/
Richters, H. and Haasdijk, E. (2022). De inzet van betrouwbare en transparante AI binnen de overheid, Deloitte [Online]. Retrieved from https://www2.deloitte.com/nl/nl/pages/publieke-sector/articles/verantwoord-omgaan-met-ai-en-data.html
Ismail, K. (26 October, 2018). AI vs. Algorithms: What’s the Difference?, CMSWire [Online]. Retrieved from https://www.cmswire.com/information-management/ai-vs-algorithms-whats-the-difference/
Chambers, J. (22 December 2021). Can Algorithms be Racist?, Towards Data Science [Online]. Retrieved from https://towardsdatascience.com/can-algorithms-be-racist-6dddf8d69065
Thank you for writing this blog post! Unfortunately, there have been many cases of AI algorithms that have racist tendencies. It happened recently in the Netherlands with the “Toeslagenschandaal”. The tax authorities used an AI algortithm that turned out to be able to racial profiling and even violating human rights. In my opinion governments and other organizations who are guilty of creating such AI algorithms should face repercussions. These organizations are responsible for making sure that their algorithms dont have any discriminatory elements.
Hi,
I think it is really an interesting topic as many of us can have a misinterpreted picture of AI being a ‘fair judge’. We assume that because AI does not have any sociocultural influences it cannot make dicriminative and / or unequal decisions. The truth is that AI systems are created and largely influenced by human-driven data. That is why we need to be very careful with what informations we are ‘feeding’ the alogirthms. I am wondering if a simple soultion to thisproblem would be dropping information regarding race, gender, ethnicity, nationality, ability etc. from the datasets used in AI. If technology wouldnt have access to the above-mentioned inputs, there is no way to explain why would it provide dicriminative results.
Let me know what you think about it!
Bartosz