AI in Criminal Justice

Introduction

Conceptually, Artificial Intelligence is the ability of a machine to perceive and respond to its environment independently and perform tasks that would typically require human intelligence and decision-making processes, but without direct human intervention.

One facet of human intelligence is the ability to learn from experience. Machine learning is an application of AI that allows a programme to analyse a set of data and then learn how to make predictions, or take decisions based on what was learned from previous trials or experiences. The possibilities of artificial intelligence presently is at the level of weak artificial intelligence, wherein the algorithm is able to perform only specific tasks and does not have a general learning capacity. Although they are not at the same level of a broad intelligence, as is the case of human beings, such programs can create opportunities for diverse applications.

            The use of intelligent agents has been developing in the field of criminal justice and has a broad scope for further development. The current use of the and scope of the same is differentiated according to the four general pillars of the criminal justice system, i.e, law enforcement, prosecution, the courts and criminal correctional systems.

AI in Law Enforcement

“The man who pulls the lever that breaks your neck, will be a dispassionate man. And that dispassion, is the very essence of justice. For justice delivered without dispassion, is always in danger of not being justice.”

‌-Quentin Tarantino, The Hateful Eight

‌Though in common discourse AI is interpreted as a futuristic phenomenon that can mostly manifest through sci-fi movies, it has been actively explored since John McCarthy introduced the area of Artificial Intelligence in 1956. That very year, Philip K Dick published Minority Report, a book about future technology that makes it possible to predict crimes and catch criminals before the occurrence, which later was made into a movie starring Tom Cruise. Though we do not possess psychic 'precogs' akin to the ones featured in the movie, data mining and tools such as predictive analytics can indeed do wonders that fall under the ambit of artificial intelligence.

‌Artificial Intelligence is a set of methods and systems used to solve super-complex problems that cannot be solved by direct application of mathematical procedures and hence need a certain level of abstraction or thinking, similar to cognition in humans. Though the human neural network is much more complex and capable of diverse thought, the AI jumps over the logistical barrier of labour and time, accessing databases and identifying patterns exponentially faster. In the context of Law Enforcement, this means collating information about the nature and circumstances of human behaviour can be used for investigating, and maybe, more controversially, predicting crime.

The versatility of AI in this field remains such that it can include recording of behaviour include facial and vocal recognition,  generating new information through data mining that reveals patterns of organized crime, influence decision making through search of relevant information to possibly recording the whereabouts, timings, and profiles of crimes to reveal target areas and vulnerabilities, and maybe even infer from collected data in order to 'forecast' where and when crime may most probably take place.

Though we still aren't quite sure of a virtual Robocop or sentient unmanned sky patrol, cameras can easily recognise faces and detect suspicious behaviour, whether in a shopping aisle or for planting a bomb. Though more abstract than the visual presentation of AI that the masses usually prefer, the best use that AI has found is through software algorithms that mine data and/or influence decisions. This could be the mathematical reduction and description of an entity (modeling), the determination of the methods by which limited resources will be organized (queuing) or gaining insight through reproduction of the dynamics of the system (simulation).

Information has no real utility in itself. But once information is made actionable, it becomes knowledge,  and holds innate value and acts as a resource. Knowledge can be discovered through data mining, or Knowledge Discovery in Databases, which essentially generates knowledge through search of patterns occurring across large batches of data, often collected for different purposes.This often brings forth evidence in criminal cases, especially financial scams as anomalies and patterns stand out distinctively. But researchers have further argued that data surrounding the offender and nature of crimes committed would yield high benefit by throwing light on geographic, temporal and individual probabilities of crime occuring. Predictive Analytics provide the risks and opportunities in data, making law enforcement proactive rather than reactive. This means monitoring of high risk situations, environments and people and prevention of crimes that would probably occur. We encounter several frustrating ethical and logical dilemmas here.The building and training of such predictive tools may very well be imbibed with the bias of the source of training and show that bias in the results and decisions. The 2016 ProPublics investigation of one such tool named COMPAS revealed bias against minorities in the process, hence failing to actually keep up the objective and neutral facade of AI. Furthermore, there are no actual policies governing the use and implementation of the information by police on ground, making the probability of exploitation and subsequent encroachment upon civil liberties seem undeniably high. But if we could, hypothetically, do away with bias and concealment of information,then we would have, the much desired, foreknowledge. As Sun Tzu said in the Art of War, 'foreknowledge’ is “the reason the enlightened prince and the wise general conquer the enemy whenever they move and their achievements surpass those of ordinary men”.

AI in Prosecution

A prosecutor is an attorney who represents the federal or state government in court proceedings. They are the principal representative of the state in every matter related to the adjudication of criminal offences. The role of the prosecutor can broadly be divided into two: the investigation process and commencing the proceedings of a trial which occurs if there is substantial evidence on hand. They investigate crimes with the police and have contact with the accused, the victim, and witnesses. Once the preliminary investigations have been completed, they judge whether there is sufficient evidence to bring the case to court.They question the suspect, witnesses and experts in order to establish the suspect’s guilt.

The prosecution is carried out if there are reasonable prospects of securing a conviction and if it serves public interest. Once the charges are laid, the defendant is notified. The hearing and trial take place soon after, followed by the sentencing. Prosecutors are authorised to offer plea bargains and also conduct the trial on behalf of the state and recommend the accused’s sentence.

In the case of investigations, they provide advice and make sure that the evidence required for conviction is present. Artificial intelligence plays a huge role in collecting evidence. They do this in multiple ways, the first being the AI’s propensity for detecting patterns. To do this, AI systems are fed with multiple images found at  crime scene over the years, and are made to recognise patterns and even possible connections between criminal cases. This in turn will alert the police that there are crime patterns and evidence to be collected. The above was an example of an AI software being developed in the University of Leon in Spain, the prototype of which will soon be trialled by the Spanish police force. This is increasingly important due to the vast amount of time required to carry out a proper investigation coupled with the number of cases to be taken care of as well the cost of carrying it all out. Because budgets are insufficient and the police are understaffed, the AI can greatly aid the police force in this capacity. They are able to filter through the immense amount of visual stimuli a lot more efficiently than humans possibly could. Moreover, due to their efficiency and lack of fatigue, the ground covered by an AI system is more extensive. The 8.7 million images of child nudity that had been unearthed on Facebook in 2018 was only possible due to the creation of a software used which was able to identify and flag all possible images of children depicted in any sexual capacity. AI in this foray has an incredible amount of potential but it must be noted that the decision made by the software cannot be changed once the decision is taken to the courts.

DNA collected at the scene of crime is crucial evidence. However the DNA collected usually comes from multiple sources (such as the victim, a pet, a witness, the suspect etc.) It is time consuming for DNA analysts to separate and distinguish the sources of the DNA and most of the time, inaccurate as well. In fact a study conducted on 108 forensic labs in the US wrongly detected DNA material from three people instead of two and in real life this could have resulted in an innocent person being falsely accused and implicated in the crime.

A system called PACE (Probabilistic Assessment for Contributor Estimation) which was developed in Syracuse University, is a machine learning algorithm which has been trained on thousands of dummy samples which contained DNA from multiple sources. The software gradually learned to differentiate between the DNA. Although not completely, accurate it is still more so than the alternate method.

When police are on the lookout for missing persons or murder victims, knowing what the person is helpful. At present, forensic anthropologists work by piecing together fragments of a person’s face and build up the facial tissue using a physical medium such as clay. This task is laborious and very time consuming and the accuracy usually depends on the anthropologist. A system is being developed at the Louisiana State University where the programmer trains the algorithm by feeding it images of people’s faces in order to find a face that would most closely fit the reconstructed skull beneath. In order to do this, the system creates several thousand facial structures and discards thousands more before finding the one that provides the best match

Similar to its role in procuring evidence, AI is widely used in various stages of the trial. It begins by playing a role in legal analytics which is used to predict future events and identify trends and patterns. This is possible because the AI system can do a thorough and comprehensive data search, finding relevant points from past cases.

In witness testimonies, it is important to ascertain the accuracy of their accounts. AI can help here because it can detect whether the witness is lying. This is referred to as demeanour evidence which is used to assess behaviour, conduct and mannerisms in the hopes of establishing more credibility or lack thereof, to a witnesses testimony. Facial lie recognition uses micro expression, movement of individual facial muscles and body language. Such AI algorithms are already being used at checkpoints between the border crossing points in Europe. A software used in the court called DARE (Deception Analysis and Reasoning Engine) which was developed and designed in the University of Maryland, was programmed with videos from the courtroom. It managed to spot 92% of the micro expressions displayed. In the case of bails, AI systems have been used in risk assessment to determine the extent of recidivism (whether the person is likely to repeat the crime). Judges often grant bail (or do not) based on this.

It is clear to see that AI plays a substantive and vast role, one that is ever expanding. Although AI does make for a more efficient system, it is far from perfect. In fact, its usage calls many other ethical matters into consideration, one such issue being privacy, another being that AI systems lack empathy and discretion. Is it possible to allow a machine the sovereignty  to impinge upon the fate of a human being? Moreover, because most software programmes are proprietary, these companies are not liable to sharing their code. As a result of this, there is a judicial system that is not required to explain itself. Due process of the law allows for cross examination on the part of the defendant which is no longer possible once AI is thrown in the mix. In the case of an unfair or faulty ruling even judges are apprehensive of changing the same and take the AI’s input into consideration because it has reviewed thousands of similar cases. It will take away from the transparency and accountability of the justice system, which is perhaps the biggest ethical violation. Although AI systems may be less biased than a human may be. the role of the programmer still plays an important role and any biases he or she may have, creeps into the programme as well; as was seen when an AI system wrongly identified dark skinned members of Congress in the US as criminals.    

Although artificial intelligence is not yet being used to its full potential it still plays a larger role than most people are aware of. Opinion continues to be divided on whether AI systems can somebody be competent enough completely take over the roles of attorneys. Others however believe that AI no matter how advanced can never fully take over and will merely remain aides in the criminal justice system.

AI in the Courts

            Once a crime has been committed and a violator has been identified by the police, the case goes to court. A court is a system that has the authority to make decisions based on law. Criminal cases are heard by trial courts with general jurisdictions. Usually, a judge and jury are both present. It is the jury’s responsibility to determine guilt and the judge’s responsibility to determine the penalty, though in some states the jury may also decide the penalty. Unless a defendant is found “not guilty,” any member of the prosecution or defense (whichever is the losing side) can appeal the case to a higher court.

There are numerous researches which attempt to apply, justify, or as a matter of fact, unjustify the use of the Artificial Intelligence in court rulings. AI software used for finding patterns in the process of decision-making are suggestable options in predicting the outcome of court trials. As reported by an article in The Guardian, a group of computer scientists at University College London devised an AI Judge to predict the results of real life cases. The artificial judge arrived at approximately the same verdicts as the judges at the European Court of Human Rights in almost four in five cases involving torture, derogatory treatment and privacy. The software was designed to accommodate legal evidence along with moral questions of right and wrong. The algorithm examined data sets for related cases. In each case, the software analysed information and made a judicial decision, 79% of which were the same as the verdicts delivered by the court.

However, the concept of artificial judicial judgements would require replicating a human conscience altogether, which falls under the purview of strong artificial intelligence. Technology has not made such high advancements that it could replicate a human brain.

Another important research was conducted by the National Bureau of Economic Research in the USA. A software to measure the likelihood of defendants fleeing or committing new crimes while they are awaiting trial in liberty was developed (Júnior, 2017). The algorithm assigned a risk score based on the offense they are charged with, when and where the person was detained, their age and their criminal record. The software has been tested on numerous criminal cases in New York and has been proven to be more efficient at assessing risk than judges.

However, the question of their accountability and transparency of these algorithms still stand since they may reproduce human prejudice and prevalent racial disparities. Such algorithms should be verifiable and auditable in order to prevent non-transparent decision making criteria.

The character of justice. Using hyper-complex modelling decision making techniques guarantees that cases with meaningfully identical features always have the same outcome. (Brennan-Marquez, & Henderson, 2018)

However, even if the AI can make the “right verdict” on a case, should it be allowed to?

The argument boils down to the fact that in any liberal democracy, there should be an aspect of role-reversibility to judgements. In some contexts, those who exercise judgment should be vulnerable, in reverse, to its processes and effects. And those subject to its effects should be capable, reciprocally, of exercising judgment (Brennan-Marquez, & Henderson, 2018). What matters is whether decision-makers are situated to imagine themselves into the role of an affected party, and vice versa—such that both participants, and in some sense the entire moral community, can understand judgment as a democratic act.

Even in the case of jury trials, role reversibility is exercised to some extent. Even when a jury trial does not lead to a different outcome than a trial before an institutional judge, it facilitates the systematic recognition of judgment’s human toll. Thus transforming the trial into a fairly democratic act.

Should the execution of laws of the community be entrusted to AI? Even though the delegation of such power can lead to consistent and accurate decisions, relieving us from the agony of decision making. Each decision is primarily based on a value system and an implementation outcome (which has its roots in logic). To ensure the same, it is important to keep humans ‘in the loop,’ exercising ultimate say over the decision-making process.

Scope for future development. Although having AI judges is a debatable concept for now and in the near future, it is feasible for AI to play a supporting role in the decision making process. A decision making version of the Eisenhower Matrix can be employed  which helps distinguish between what is important and what is urgent. Urgent tasks are time sensitive whereas important tasks are more strategic. When it comes to decision making, decisions can vary in their reversibility and the level of consequences that they may have. (Farnam Street, 2018)

            Weighing these factors can help one delineate possible deadlines for a task and prioritise the same. It can also help decide whether delgatation of the task would be a feasible option. Making use of the Eisenhower matrix can help increase the productivity of a system as well as direct flow of labour efficiently in the right direction.

            If a software could be developed wherein algorithms could process case information and organise them on the basis of their urgency, importance, reversibility or consequences, decision making processes could be more addressing criminal cases considering said factors.

AI in Criminal Correction Systems

              When it comes to the field of criminal corrections, AI has been shown to do two things remarkably. Get the job done and/or fail spectacularly at it. The applications for AI in the field of Criminal Correction systems is actually seemingly endless. Given the number of individuals trapped within the confines of the legal system and ultimately within jails, it is estimated that 1 in every 38 Americans is or has served some prison time. Thus AI equipped with Re-Offender algorithms plays an important role here.

            It estimates the capacity of each individual to not only commit a crime but also whether they will commit the same crime again. It has been proved however, time and time again that the AI fails rather miserably. It has intense difficulty identifying faces of colour and at times has even mistaken members of Congress for criminals (Hao, 2019). It is stated that modern algorithms are driven by training based on historical crime data.

            The error here might lie in the fact that AI uses a strategy of machine learning such algorithms instead of deep learning modules that could help advance the scoring mechanisms.

            AI has also successfully been used in systems of modern prison management using Bayesian algorithms. It has been used in three broad areas in the prison systems, namely, overcoming one-sided cell allocation strategies (where instances like individuals known to the criminal are avoided); lack of scientific guidance (Where one needs to overcome the human errors in cell allocation) and lastly the influences of uncertainties of allocation results (where previous offenders might try to escape based on available conditions in relation to their housing cell). These systems have been implemented in management in areas in China (Jang, Wang and Wu. 2018)

References

Al Fahdi, M., Clarke, N. L., & Furnell, S. M. (2013). Towards An Automated Forensic Examiner (AFE) Based Upon Criminal Profiling & Artificial Intelligence. Retrieved August 13, 2019, from  https://pdfs.semanticscholar.org4fb1/0dbfc73cf8c1b1f4e387344bf8f4af9a3060.pdf

Angwin, J., Mattu, S., Larson, J., & Kirchner, L. (2016, May 23). Machine Bias. Retrieved from https://www.propublica.org/article/machine-bias-risk-assessments-in-criminal-sentencing

Brennan-Marquez, K., & Henderson, S. E. (2018). Artificial Intelligence and Role-Reversibility. The Journal of Criminal Law and Criminology, 109(02), 137-164. Retrieved August 14, 2019

Brigham, K. (2019, March 17). Courts and police departments are turning to AI to reduce bias, but some argue it’ll make the problem worse. Retrieved from https://www.cnbc.com/2019/03/16/artificial-intelligence-algorithms-in-the-criminal-justice-system.html

C. B. (2019, March 4). The New Weapon in the Fight Against Crime. Retrieved from http://www.bbc.com/future/story/20190228-how-ai-is-helping-to-fight-crime

Farnam Street. (2013, April). Eisenhower Matrix: Master Productivity and Eliminate Noise. Retrieved August 15, 2019, from Farnam Street: https://fs.blog/2013/04/eisenhower-matrix/

Farnam Street. (2018, September). The Decision Matrix: How to Prioritize What Matters. Retrieved from Farnam Street: https://fs.blog/2018/09/decision-matrix/

Hao, K. (2019, January 21). AI is sending people to jail - and getting it wrong. Retrieved August 14, 2019, from MIT Technology Review: https://www.technologyreview.com/s/612775/algorithms-criminal-justice-ai/

Ha-Redeye, O. (2019, March 24). Using Artificial Intelligence for Demeanour Evidence. Retrieved from http://www.slaw.ca/2019/03/24/using-artificial-intelligence-for-demeanour-evidence/

Johnston, C. (2016, October 24). Artificial intelligence 'judge' developed by UCL computer scientists. Retrieved August 14, 2019, from The Guardian: https://www.theguardian.com/technology/2016/oct/24/artificial-intelligence-judge-university-college-london-computer-scientists

Júnior, O. P. (2017, March 12). How can artificial intelligence affect courts? Retrieved August 15, 2019, from Institute for Research on Internet and Society: http://irisbh.com.br/en/how-can-artificial-intelligence-affect-courts/

Martin, M. (2019, June 15). San Francisco DA Looks To AI To Remove Potential Prosecution Bias. Retrieved from https://www.npr.org/2019/06/15/733081706/san-francisco-da-looks-to-ai-to-remove-potential-prosecution-bias

   National Research Council. 2001. What's Changing in Prosecution?: Report of a                Workshop. Washington, DC: The National Academies Press. https://doi.org/10.17226/10114.

Philipsen , S., & Themeli, E. (2019, May 15). Artificial intelligence in courts: A (legal) introduction to the Robot Judge. Retrieved from http://blog.montaignecentre.com/index.php/1942/artificial-intelligence-in-courts-a-legal-introduction-to-the-robot-judge/

Rigano, C. (2018, October 8). Using Artificial Intelligence to Address Criminal Justice Needs. Retrieved August 14, 2019, from National Institute of Justice: https://www.nij.gov/journals/280/Pages/using-artificial-intelligence-to-address-criminal-justice-needs.aspx

Thompson, D. (2019, June 20). Should We Be Afraid of AI in the Criminal-Justice System? Retrieved from https://www.theatlantic.com/ideas/archive/2019/06/should-we-be-afraid-of-ai-in-the-criminal-justice-system/592084/

Weber, S. (2018, January 10). How artificial intelligence is transforming the criminal justice system. Retrieved from https://www.thoughtworks.com/insights/blog/how-artificial-intelligence-transforming-criminal-justice-system

Wu, S., Wang, J., & Jiang, Q. (2012). The Application of Artificial Intelligence in Prison. Advances in Intelligent and Soft Computing, 159, 331-332. Retrieved August 13, 2019, from https://link.springer.com/chapter/10.1007/978-3-642-29387-0_49

Credits

Group 8, AI in Criminal Justice:

1. Debargha Roy,1833208- Documentation

2. Sai Siddharth, 1833210- Presentation

3. Parth Malhan, 1833216- Presentation

4. Rishabh Bapat, 1833218- Scriptwriting

5. Rohit Jaiswal, 1833219- Video Editing and Direction

6. Sriram Nair 1833225- Acting

7. Y Arulvel, 1833226- Videography

8. Nathan Zachary Fernandez, 1833237- Documentation

9. Radhika Rastogi, 1833280- Documentation

10. Simone Diya, 1833294- Documentation

11. Therese Liam Tom, 1833297- Acting