What does the future hold for Al within the field of criminology?

Table of Contents:

Al can be applied in many different ways to benefit the field of crime research and prevention. As discussed earlier, Al makes it possible to discover patterns that were not possible to find by using more traditional research methods. The downside to the use of the large data sets is that to establish such an extensive set of data, one inevitably has to give up some, or possibly much, of his/her privacy.

The potential of Al techniques is currently a prominent discussion outside the field of criminology as well, more specifically in the field of Al itself, hi Europe prominent researchers have joined forces in the CLAIRE¹⁴ initiative, in an attempt to establish a European centre of excellence with an emphasis on /»/man-centred AL Similarly, top scientists in the Netherlands have recently launched an Al manifesto¹⁵ regarding the future of Al and the challenges Al faces. In this statement the authors point out the limitations of techniques like ML as well. According to the authors of that manifesto ‘Al models simply do not understand the world yet at a level that humans do, and this lack of context makes it very hard for them to generalise to new situations at the same level as humans do’. Humans can learn quickly; they sometimes only need one example, while an ML algorithm needs a large data set of examples to draw on. Humans can interpret the data based on situational circumstances while ML cannot; ML algorithms lack background knowledge and do not have the level of understanding of cause and effect as humans do. So, for example, when an algorithm has been developed to predict crime in Amsterdam, that algorithm cannot directly be used for a different city. The algorithm is tailored to a certain area based on the available data set (e.g., see Chapter 5, this volume). If you want to predict crime in London you will need a new data set to train the algorithm to do so.

Another focus area within the future of Al is called explainable Al, making people understand how algorithms come to their decisions. As Al systems such as ML and decision-making algorithms will significantly affect their users, it is

Big data and criminology from an Al perspective 3 7 important to be able to explain how and why an Al system produced the effect that it did. The example of COMPAS is a nice illustration of the complications of complex algorithms. Decisions are made but it is unclear what they are precisely based on. The aim in Al is to design tools to help make the systems more transparent. People should be able to understand how advanced Al systems work before they can be used to perform sophisticated tasks.

A third challenge can be described as responsible Al: Al that complies with our nonns and standards. Today’s complex Al systems have become less predictable while at the same time their potential impact on our society has increased. Hence to guarantee trust in these systems we need to develop algorithms that respect human values. For instance, ML algorithms might adopt the same biases as humans (e.g., with respect to race and gender). When these patterns exist in the data set, the consequences might be critical. Many examples of biased Al systems exist.¹⁶ One of these was discussed earlier in this chapter; when a minority is overrepresented in a police data set because they were racially profiled, the output of the algorithm will be that that minority has a higher probability of committing crimes in the future, when there is in fact the possibility that they actually may be the victim of a biased police coips. Examples of systems based on a biased data set outside the field of criminology exist as well. For example, a recruiting tool for technical jobs used by Amazon that favoured men,¹⁷ or even Google Translate, which translated the gender neutral word ‘doctor’ from Turkish to a male form in English, while the gender neutral word ‘nurse’ was supposed to be female.¹⁸

With the field of Al identifying these key issues as extremely important for future research, we should also take into account that current applications often lack these properties. To develop systems that add to criminological research, it is important that these limitations are acknowledged and solved. You do not want to put someone in prison based on a biased, non-transparent algorithm. One option is to only use data sets that are not biased, to exclude the possibility of unfair predictions. However, since it is not always possible to know whether or not a data set is fair or might be biased, it is very important to realise and accept that there might be some problems and then be very carefill about drawing conclusions based on these data sets. As mentioned earlier, it might not be that bad to predict crime at a certain location which turns out to be incorrect. But making predictions about individuals committing future crimes and attaching consequences should not be allowed when you are not sur e about the fairness of the data these predictions are based on. Developing a mechanism that is be able to understand what would be fair, in a way that humans do, is considered to be the holy grail in AL Extending algorithms with built-in ‘fairness’ would solve the problem but remains a monumental challenge for now.

To conclude, the possibilities of using Al and especially big data within criminological research are endless but it is up to us to guide this in the right directions, to obtain context-aware, transparent, and unbiased systems that make the world a safer place. It is therefore essential that the research conununity in computer science puts these societal aspects of Al high on its agenda.

Notes

1 www.scoop.it/u/dahl-winters
2 https://nl.wikipedia.org/wiki/Criminaliteits_Aiiticipatie_Systeem
3 https://nl.wikipedia.org/wiki/Criminaliteits_Anticipatie_Systeem
4 www.predpol.com/
5 https://nl.wikipedia.org/wiki/Criminaliteits_Anticipatie_Systeem
6 www.predpol.com/
7 Issie Lapowsky. “How the LAPD Uses Data to Predict Crime”, www.wdred.com/story/ los-angeles-police-department-predictive-policing''
8 wwv.cdcr.ca.gov/rehabilitation/docs/FS_COMPAS_Final_4-15-09.pdf
9 Ed Yong. “A Popular Algorithm Is No Better at Predicting Climes Than Random People”. www.theatlantic.coni/technology/archive/2018/01/equivant-compas-algorithm 550646/
10 Julia Angwin, Jeff Larson, Surya Mattu and Lauren Kirchner. “Machine Bias”, www. propublica.org/article/machhie-bias-risk-assessments-in-crimhial-sentencing
11 www.trapwdre.com/
12 Charles Arthur. “Trapwire Surveillance System Exposed in Document Leak", www. theguardian.com/world/2012/aug/13/trapwire-surveillance-system-exposed-leak
13 Scott Shane. “WikiLeaks Stirs Global Fears on Antiterrorist Software”, www.nytimes. coni/2012/08/14/us/trapwire-antiteiTorist-software-leaks-set-off-web-furor.html
14 https://claire-ai.org/
15 Special Interest Group on Artificial Intelligence, http://ii.tudelft.nl/bnvki/wp-content/ uploads/2018/09 Dutch-AI-Manifesto.pdf
16 Daniel Cossins. “Discrimhiating Algorithms: 5 Times Al Show'ed Prejudice”, www. new'scientist.com^f'aiticle/2166207-discihnhiathig-algoritluiis-5-thiies-ai-showed-prejudice/
17 “Amazon Ditched Al Recruiting Tool that Favored Men for Technical Jobs”, www. theguardian.comtechnology/2018/oct/10/amazon-hiring-ai-gender-bias-recniiting-engine
18 Adam Hadhazy. “Biased Bots: Artificial-Intelligence Systems Echo Human Prejudices”, ww'w'.princeton. edu/news/2017/04/18/biased-bots-artificial-intelligence-systems-echo-human-prejudices

References

Bhattacharyya, S., Jha, S., Tharakunnel. K„ & Westland, J. C. (2011). Data mining for credit card fraud: A comparative study. Decision Support Systems, 50(3), 602-613, Elsevier.

Bishop, C. M. (2006). Pattern recognition and machine learning. New'York, NY: Springer.

Chen, H„ Chung, W„ Qin. Y, Chau, M„ Xu. J. J., Wang, G„ ... Atabaklish, H. (2003). Crime data mining: An overview and case studies. National Conference on Digital Government Research, 1-5. Boston. MA.

Chen, H., Chung, W., Xu, J. J., Wang. G., Qin. Y, & Chau, M. (2004). Crime data mining: A general fiamew'ork and some examples. Computer, 37(4), 50-56.

Chiranjeevi, C. B., & Revathy, R. (2015). Ding trafficking suspect prediction using data mining. International Journal of Advanced Research in Computer Science and Software Engineering, 5(8).

Dick, P. K. (1956). The minority report. Short Story- in Fantastic Universe, 4(6).

Friedman, J. H. (1998). Data mining and statistics: What’s the connection? Computing Science and Statistics, 29(1), 3-9.

Gerber, M. (2014). Predicting crime using Twitter and kernel density estimation. Decision Support Systems, 61, 115-125.

Geiritsen, C., & Yan Breda, W. (2015). Simulation-based prediction of collective emotional states. Proceedings of the International Conference on Social Computing and Social Media, LNCS, 118-126, Springer.

Kohavi, R„ & Provost, F. (1998). Glossary of terms. Machine Learning, 30, 271-274.

Lin, Y, Chen, T., & Yu, L. (2017). Using machine learning to assist crime prevention. Proceedings of the 6th IL4I International Congress on Advanced Applied Informatics (IIAI-AAI), 1029-1030. Hamamatsu, Japan.

McClendon, L., & Meghanathan, N. (2015). Using machine learning algorithms to analyze crime data. Machine Learning and Applications: An International Journal (MLAIJ), 2(1). doi:10.5121/mlaij.2015.2101.1

Pang. B., & Lee, L. (2008). Opinion mining and sentiment analysis. Foundations and Trends in Information Retrieval, 2(1-2). 1-135.

Perry, W. L., & McInnis, B. (2013). Predictive policing: The role of crime forecasting in law enforcement operations. Santa Monica, CA: RAND.

Rienks, R. (2015). Predictive policing: Taking a chance for a safer future. Korpsmedia. PDC.

Samuel, A. (1959). Some studies in machine learning using the game of checkers. Journal of Research and Development, 3(3), 210-229, IBM.

Siegel, E. (2013). Predictive analytics: The power to predict who will click, buy, lie or die. Hoboken. NJ: John Wiley and Sons Inc.

Skeem, J. L., & Louden. J. E. (2007). Assessment of evidence on the quality- of the Correctional Offender Management Profling for Alternative Sanctions (COMPAS). file:///C:/Users/ EIgenaar/Downloads/Assessment_of_Evidence_on_the_Quality_of_the_Corre%20(l).pdf

Uclrida, C. D.. & Swatt, M. L. (2013). Operation LASER and the effectiveness of hot spot patrol: A panel analysis. Police Quarterly, 16(3), 287-304, Sage.