Applications and Implications of Algorithmic Decision-Making for Just Societies: The Case of Crime Prediction through Big Data

Emmanuel
Letouzé
Short bio: 

Emmanuel Letouzé is the Director and co-Founder of Data-Pop Alliance on Big Data and Development, co-created by the Harvard Humanitarian Initiative (HHI), MIT Media Lab, and the Overseas Development Institute (ODI), where he is respectively a Fellow, a Visiting Scholar, and a Research Associate. He is also a PhD Candidate in Demography at UC Berkeley (ABD, finishing this year) and a Non-Resident Adviser at the International Peace Institute (IPI). In 2011-12 Emmanuel worked as a Development Economist at UN Global Pulse, where he wrote UN Global Pulse's white paper “Big Data for Development". Before that he worked for UNDP in New York (2006-09) and in Hanoi, Vietnam, for the French Ministry of Finance (2000-04). Emmanuel is a graduate of Sciences Po Paris (BA, Political Science 1999, MA Economic Demography, 2000) and Columbia University SIPA (MA Economic Development, 2006) where he was a Fulbright Fellow. As a political cartoonist ("in the making since 1975") Emmanuel contributes cartoons to Rue89, a French news website, to the satirical blog Stuffexpataidworkerslike as well as illustrations for development reports and campaigns.

Abstract: 

Crime is arguably one of the most salient symptoms and drivers of social fragmentation, exclusion, and disenfranchisement. Violent crimes such as homicides and rapes constitute infringements on human rights and all crimes are impediments to social progress. Given the high individual and societal benefits of reducing crime levels—including possibly ‘before they happen’—the opportunity cost of not using reliable predictive tools as they may become available and more sophisticated is significant.

The increased availability of fine-grained behavioral data is making crime prediction possible with growing accuracy. ‘Predictive policing’ is currently being used by an increasing number of police departments of large metropolises in the US, UK and a few other countries. Additionally, a nascent body of academic literature involving cell-phone and transportation data analysis has emerged, investigating the possibility and effectiveness of calling and mobility patterns to help forecast crime hotspots. These predictive models may be completely a-theoretical or based on some hypotheses about causal processes that may help devise public policies (e.g. the features of public transportation systems (such as frequency and coverage) and their effect on crime patterns and trends).

At the same time, the development and deployment of these methods and tools raise a number of hard ethical questions, notably around individual and group privacy—another fundamental human right. One ‘intrinsic’ facet of the problem is the fact that the informed consent of the emitters of data is typically not clearly established in such initiatives. An ‘instrumental’ aspect is the risk of profiling, harassment, and reinforcement of existing inequities and prejudices that may result from blind or overreliance on algorithmic predictions.

Against this general background, the paper will (1) take stock of the state of research and practice in the field, (2) discuss whether the use of aggregated and supposedly ‘anonymized’ datasets is a sufficient safeguard against possible harms, included unintended, (3) assess whether profiling risks and reinforcements of inequality that may result from such data-driven approaches can be mitigated and the costs of error balanced with the benefits of prediction in light of model accuracy, and (4) interrogate and suggest ethical principles and requirements and associated legal, regulatory and institutional frameworks that should inform future applications and developments in the field—with a focus on the need to foster opportunities for members and representatives of at-risk communities to have a voice and be actors in related dialogues, decisions and initiatives.