06 October 2022
I thought I would never understand nor care about algorithms. I was wrong. Algorithms are now being used in the public sector and government — and they are making decisions that affect us directly. This is why we should know, and care, about algorithmic transparency.
Initially, the word “algorithm” sounded to me like something extremely abstract and complex — the sort of term only geeky people would understand. I didn’t know how to put into plain words what an algorithm did nor how it could be used by government. When I first carried out the classic Wikipedia search, I read: “In mathematics and computer science, an algorithm is a finite sequence of rigorous instructions, typically used to solve a class of specific problems or to perform a computation. Algorithms are used as specifications for performing calculations and data processing.” This was my face afterwards:
Luckily, when I joined Oxford Insights, I had the opportunity to learn about how Artificial Intelligence is used by governments across the world to improve their public services (check our AI Readiness Index 2021). I was also given the chance to learn about Trustworthy AI. This is when I finally started to understand in practical terms how algorithms are used in the public sector, and my interest in the topic of ‘algorithmic transparency’ grew.
How are algorithms used in the public sector?
In the last few years, algorithms have been used in the public sector. Governments are using them to make a wide range of decisions, from dealing with waste collection to allocating heart transplants. Algorithms have been used to support investigations of reports made concerning possible illegal holiday rentals, but also to support child protection services, resource allocation planning (for example, homeless shelter planning), or predictive policing. Algorithms have also been adopted to accelerate efficiency—for instance, by helping to prioritise investments in road work through the analysis of traffic-bottlenecks—or expedite large-scale routine services (such as the processing of VISA applications.)
Algorithms are also supplementing or replacing decision-making previously undertaken by humans. Algorithmic models will analyse data and make predictions based on them. These predictions will then inform decisions. And these decisions could have a very concrete impact on our lives. For instance, the UK government is trialling an algorithmic model to predict whether universal credit claimants should receive benefits based on their perceived likelihood of committing fraud in the future. While in the past this would have been an activity done by civil servants, now an automated model is analysing historic fraud data and deciding who should receive an economic aid based on the results.
When I learned about this, I couldn’t help but ask myself: how do these algorithms actually work? Who builds them? Are there any ethical concerns about the way they are built? Can I challenge a decision made by an algorithm?
Algorithmic bias and discrimination
Not knowing the answers to those questions can be dangerous, as automated systems can in fact be discriminatory or biased. Safiya Noble talks about this in her book “Algorithms of Oppression” (check out her TEDx talk about algorithm bias too), where she discusses how algorithms behind search engines embed negative biases against women of colour and privilege whiteness. For instance, she gives the example of how when she first searched for the term “black girls” in Google ten years ago, the top results all led to porn sites. This didn’t happen when searching for “white girls”. According to the author, because search engines such as Google are motivated by commercial interests, big companies with influence and money can skew search results to their own benefit. This makes algorithms not neutral.
Another element that contributes to algorithmic bias is the source and quality of the data that is fed into them. For instance, if you want to train an algorithm to recognise faces, then you should train it with images of faces of people with different age, sex, ethnic background, etc. If you only feed an algorithm with white faces, it will then not learn to recognise faces from other ethnicities. If you only feed an algorithm with old faces, it will then not learn to recognise young ones. Not having quality data representative of every segment of society can hence make algorithms biased.
Learning how algorithms can foster discrimination and bias alarmed me. It made me realise how important it is to open up algorithms for public scrutiny, so that their purpose, structure, actions and outcomes can be challenged or disputed. Also, that it is imperative to make algorithms’ use and characteristics understandable to citizens, so that the latter can exert oversight on them.
Why algorithmic transparency?
All these thoughts brought me to the importance of what we call ‘algorithmic transparency’. How do we promote it and what does it look like? According to the Ada Lovelace Institute, in order to obtain meaningful transparency there are some key questions that need to be answered about an algorithmic system. These include: what are the data sources of a system? What is the logic of the system? Are there groups impacted by it? There are different ways for governments to find out and publicise the answers. Mechanisms such as the assessment and evaluation of automated systems, the standardised disclosure of the data which was used to create those systems, or the publishing of the code behind their functioning are some of them.
And why is having algorithm transparency important? Algorithmic transparency is essential to challenge automated decisions and ensure that our rights are protected. And this is because even algorithms which were created with a good purpose could undermine our rights. For instance, in the Netherlands, a court ruled that an algorithm which had been set up to detect welfare fraud was breaking the right to privacy.
Transparency is also critical to ensure that algorithms established to protect vulnerable populations are used fairly. According to the Fundación Éticas, the algorithm VioGen — launched in Spain to protect victims of sexual harrassment — raises concerns around lack of independent oversight, accountability, and end-user engagement. Being able to carry out external audits of these automated systems, especially when they take decisions which are far-reaching and related to sensitive topics, is key to ensure their correct functioning and maintain their fairness. By having algorithmic transparency, citizens would be able to dispute algorithmic decisions or outcomes. This, in turn, would reduce the existence of algorithmic bias.
How do we promote algorithmic transparency?
Establishing legal frameworks and international standards that include algorithmic transparency would make governments accountable to promoting the latter. It would ensure that there are rights in place for citizens to oversee and dispute algorithmic decisions. The OECD AI Principles are a good example, as they include a principle on “Transparency and explainability” that mentions how AI Actors should “enable those affected by an AI system to understand the outcome; and, to enable those adversely affected by an AI system to challenge its outcome based on plain and easy-to-understand information on the factors and the logic that served as the basis for the prediction, recommendation or decision.”
Some national governments have also started to adopt their own ethical guidelines around the use of algorithms — as mapped by the AI Ethics Guidelines Global Inventory by Algorithm Watch. This is a good way to ensure an ethical use of AI systems and avoid bias. Having public registers of algorithms is also a good step towards more transparency, as citizens can get familiarised with the cities’ algorithmic systems and give feedback on their use. Localities such as Amsterdam and Helsinki have already established their own.
For citizens to exercise public scrutiny over these systems, they need to receive digestible, simple information about them. This is especially so in the case of people who may not be comfortable using, or may lack the access to, technology. Making it easier to understand algorithms, through educational videos or public engagement campaigns, is a potential way to foster public participation in this area. Strengthening civil servants’ knowledge around digital and data rights as well as ethical concerns around the use of these systems is also important. This could contribute to establishing accountability checks within the government itself.
As technology and automated systems are more widely adopted by our own governments, it is critical for civil society to maintain scrutiny on how they are designed and used. The adoption of algorithms for decision-making in the public sector can be very beneficial — to speed up processes, analyse very high amounts of data rapidly, or provide additional evidence to make certain decisions — but it can also be harmful. To avoid them being so, it is key that these systems remain transparent.
In conclusion, no matter how confused you may initially be around what algorithms are (remember the meme?), I hope I made my point that it is very important for you to know and care about algorithmic transparency.
At Oxford Insights, we care about the trustworthy use of AI. Please get in touch if you would like to seek advice at firstname.lastname@example.org