Big Data has demonstrated great potential for its use as artificial intelligence in a number of different roles, such as efficient data management in legal activity. Utilizing technological advancements in AI, we can maximize the abilities of law enforcement professionals while leaving logical tasks– such as data calculation and work using pre-existing concepts– to automated systems. This in turn reduces the repetitive and menial work left to professionals, who can then use their abilities in other, more productive ways that are not yet possible through AI.
However, in applying new technology to a given professional field, we must also determine new limits of that technology, particularly in regards to ethical use of AI. For example, there has been recent conflict over the use of data analysis in crime prevention and reducing recidivism. Big Data is used to collect and analyze probabilistic information from past crimes to predict which individuals are likely to commit crimes in the future. While seemingly efficient and unbiased in theory, this application has its shortcomings. First, we cannot assume that these “pre-crime” algorithms can completely eliminate crime; there will be some individuals and acts that the algorithm cannot account for. Secondly, no matter how perfect the AI is, it is impossible to make a certain prediction about who will commit a crime, as in the science fiction tale The Minority Report by Philip K. Dick[1]. The algorithm can only predict likelycrimes, or possible future re-offenders.
Despite these limitations, pre-crime technology is already widely used; in the United States, policemen utilize pre-crime AI to identify specific areas that are more prone to criminal activity by analyzing patterns in data about previous crimes. This way, law enforcement agents can be more methodically distributed to high-risk areas. Chicago in particular uses this technology to analyze publicly available information on social media networks to predict possible victims of violent crimes.[2]
While seemingly efficient and benign, these systems are an ethical minefield; they, in essence, punish crimes that have not occurred. Even when they are fed with information from real cases, they are then subject to the biases and inequalities inherent to the failures of the criminal justice system. Here comes a quandary to consider while this technology might be apt at predicting crime based on pre-existing data, it is entirely based on data provided by police officers, who are prone to their own biases and prejudices as to what characteristics label a person a criminal.
For example, say that in a given location the police use zip code to predict crime, in an attempt to eliminate prejudice associated with other criteria. They then isolate neighborhoods with the highest index of criminal residents and move forward with the pre-crime process. Despite this criterion– zip code– not demonstrating explicit prejudices such as race or ethnicity, it still carries the weight of social prejudice; this system does not account for historical segregations of black and white communities, which still exists today despite having been outlawed years ago.
This situation occurred clearly in numerous experiments with pre-crime technology in America– in the cities of Ferguson, Missouri; Newark, New Jersey; and New York, New York– in which black people disproportionately constituted more than 80% of police approaches, despite only making up about half of the general population. Police officers are drawn to them based only on suspicions and biases, and most of these approaches do not result in any significant action.[3] Specifically, in the city of Newark, black people account for 54% of the population, but 85% of pedestrian approaches, as well as 79% of the prison population; in short, black people are 2.5 times more likely to be approached and 2.7 times more likely to be searched when compared to the rest of the population. However, the Newark Police Department asserts that the probability of encountering evidence of a crime in these approaches is equal regardless of the pedestrian’s race.[4]
Here we see a higher rate of search and seizure among certain racial or ethnic groups, based on them fitting a police officer’s idea of “suspicious.” Furthermore, due to historical segregation and class gaps, race and ethnicity can also inform other criteria, such as housing and incomes. This means that these criteria, despite not explicitly being racial, would also be subject to the biases of law enforcement officers. Thus, pre-crime systems using data provided by these officers would continue to maintain these prejudices and, given its efficiency compared to human operation, would further widen the racial disparity in suspected criminals.
This creates an urgent need for algorithmic transparency, since these technologies will inherently carry the ideologies of their programmers and their data providers. The only way to resolve this is through rigorous supervision of AI systems and effective control of the data that feeds them. They must be subject to constant review and revision to ensure maximum efficiency and minimum bias.
In summary, data analysis and crime prevention systems have great potential to not only help predict crimes, but also to understand the sociological circumstances that influence criminal offenses. In spite of this, they cannot become mechanisms that propagate inequality and maintain prejudices; there must be special attention and precautions taken when using such AI methods and when determining which data will be entered into the systems, especially in light of the enormous impact they generate.
We can say that the advancement of technology has many considerable benefits to society, since artificial intelligence can quickly perform menial tasks. However, we should also take into account that what we recognize as artificial intelligence cannot be considered “artificial consciousness,” capable of independent thinking in historical issues and providing true and unbiased qualitative data analysis. They can only produce data analysis as objective as the data provided to them.
It is of the utmost importance to discuss algorithmic transparency in order to address these shortcomings, as well as the other ethical implications of artificial intelligence, since data management is not an exact science, and lacks human understanding and caution. In this matter, we are not only discussing possible abstract results, but concrete impacts in human lives that are affected day after day.
[1] KINDRED DICK, PHILIP. The Minority Report. Published in: 1956. More Information In: https://en.wikipedia.org/wiki/The_Minority_Report.
[2] SOUTHERLAND, Vincent. With AI And Criminal Justice, The Devil Is In The Data. American Civil Liberties Union (ACLU – aclu.org). Published on: 09 Apr. 2018. Access on: 11 10. 2018. Available In: https://www.aclu.org/issues/privacy-technology/surveillance-technologies/ai-and-criminal-justice-devil-Date.
[3] SOUTHERLAND, Vincent. With AI And Criminal Justice, The Devil Is In The Data. American Civil Liberties Union (ACLU – aclu.org). Published on: 09 Apr. 2018. Access on: 11 10. 2018. Available In: https://www.aclu.org/issues/privacy-technology/surveillance-technologies/ai-and-criminal-justice-devil-%20data.
[4] FLATOW, Nicole. At Least 3/4 Of Newark Pedestrian Police Stops Had No Constitutional Basis, Justice Department Finds. Think Progress – thinkprogress.org. Published: 22 Jul. 2014. Access on: 12 10. 2018. Available In: https://thinkprogress.org/at-least-3-4-of-newark-pedestrian-police-stops-had-no-Constitutional-basis-justice-department-finds-1844d9baba8a/
This piece was contributed as part of the 2019 Harvard Legal Technology Symposium organized by the Harvard Law & Technology Society. The Symposium was the largest student organized legal technology event in the world. It brought together an interdisciplinary and international community to think deeply about how technology can improve and shape the law.