Politics by Algorithm: Classifying Migrant Risk at ICE

There has been a manipulation of the RCA algorithm over time, an observation based on the careful reverse-engineering by researchers of freedom of information disclosures from the government body, US Immigration and Customs Enforcement (ICE). Image adapted from, Evans, K. and Koulish, R., 2020. Manipulating risk: immigration detention through automation. Lewis & Clark Law Review., 24, p.789.

PPA has explored the genesis and transformation of risk classification tools used by the US Immigration and Customs Enforcement (ICE), and reviewed the impact that alterations of an algorithmic decision-making tool had on detentions, and the political provenance of such alterations. 

We adopt the position that algorithms are inherently political, but not only for prominently discussed reasons such as bias. Through the ICE case study we see algorithms as an arm of branches of government which can make changes to other branches when influence can’t be exerted through policy or other ‘normal’ channels. This in turn challenges claims that algorithms primarily increase the efficiency. Instead, we argue that in some cases algorithms were designed to be political with politics baked in intentionally in order to create a ‘digital work-around’. We therefore posit that one under-acknowledged drive to digitalise is the inability of administration to achieve policy in other ways, which in turn has implications for policy design, implementation, and outcomes in a range of sectors, including refugee resettlement.

PPA carried out a mapping of the US refugee resettlement and repatriation ecosystem, including the internal structure of the ICE risk-based decision-making RCA algorithm, the decision points of which are shown in outline above. Note that the risk-based decisions are based on an assessment of flight risk and risk to public safety, at high, medium or low levels. Within the system the ability of decision-makers to revise the recommendations of the algorithmic tool was initially designed into the system but this ability to override was controversially upgraded at different points of the development of the algorithm at ICE. This mapping is also partly based on the Evans, K. and Koulish, R., 2020.

Please click on this image to see an enlarged version. This shows the overview of the asylum seeking pathways carrying refugees to (and in some case returning from) the USA. This starts from a generalised country of origin seen on the right (otherwise known as the ‘Asylee Producing Countries’ or APCs according to the authors of Refugee Roulette), passing then to country where a claim is made, seen in the middle here. On the right, the resettlement pathways to and within the USA are shown, including the use of the RCA algorithm at ICE.

Statement on refugee matching by Will Jones and Alexander Teytelboym, 2016.

Next
Next

Voices of Tomorrow: Redesigning Refugee Resettlement Decision-Making Systems