Empowering Justice: Charting the Course for AI in Indian Undertrail Management
- Post by: Arjun Kumar
- July 11, 2024
- No Comment
Kinshuk Kandpal[1]
[1] Pursuing post-graduation in Development Studies at Indian Institute of Technology (IIT) Madras
hs22h030@smail.iitm.ac.in
| Title: | Empowering Justice: Charting the Course for AI in Indian Undertrail Management |
| Author(s): | Kinshuk Kandpal |
| Keywords: | Artificial Intelligence, Legal Sphere, COMPAS (Correctional Offender Management Profiling for Alternative Sanctions), Public Policy, DPI (Digital Public Infrastructure) |
| Issue Date: | 11 July 2024 |
| Publisher: | IMPRI Impact and Policy Research Institute |
| Abstract: | The legal sphere resolves transgressions and has a sanctified space for human judgment. It is increasingly encountering AI-related aids, and it must be adequately adjusted to function how it has been designed. This paper sees the legal sphere’s interaction with undertrial cases as a deep transformational leverage point for a dedicated AI-based intervention. So, the Indian framework i.e. the policies, datasets, and data collection are examined and discussed in the context of COMPAS (Correctional Offender Management Profiling for Alternative Sanctions), which assesses offenders’ recidivism, to craft a policy for an Indian AI-powered undertrial management system. |
| Page(s): | 83-87 |
| URL: | https://iprr.impriindia.com/empowering-justice-charting-the-course-for-ai-in-indian-undertrail-management/ |
| ISSN: | 2583-3464 (Online) |
| Appears in Collections: | IPRR Vol. 3 (1) [January-June 2024] |
| PDF Link: | https://iprr.impriindia.com/wp-content/uploads/2024/07/YV1_Empowering-Justice_Kinshuk-Kandpal-IPRR-V3I1-Jan-June-2024.pdf |
(January-June 2024) Volume 3, Issue 1 | 11th July 2024
ISSN: 2583-3464 (Online)
1. Introduction
Communities form a society or are recognized as being part of a society in terms of shared attitudes and interests. So, what takes care of the interests must be the bedrock of ‘the community,’ which is justice. This concept in modern times has come to embody fairness and reciprocity, which leads to the endowment of legitimacy to the societal structure, where the value given to human judges and their foresight is vast. A system change is desired here to effectively engage with the rigid structure and narratives surrounding the legal system. Leverage points are places to intervene in a system. Identifying and utilizing a leverage point is necessary because these offer a way to generate system change (Abson, et al, 2016). I argue that AI and related tools can help in administering cases, thus enabling the legal system to do what it is meant to do better. The data reveals around 278 thousand undertrials out of a total of 381 thousand prisoners. This means more than two-thirds of the criminals are undertrials, underscoring the worthiness of the issue (Sahoo & Jain, 2015). The paper will discuss mainstreaming measures and accelerating actions. Finally, we will look at policy support for AI in the legal sphere while considering the ethical considerations one must make to operationalize the aforementioned system.
2. An Indian Undertrail Management System
Correctional Offender Management Profiling for Alternative Sanctions is an artificial intelligence program that utilizes deep learning to train the algorithms used by several US courts constantly. It uses a 137-question assessment, requiring direct input from the accused and data from the authorities related to incarceration. This provides the risk score of recidivism on a 10-point scale, where a higher score means a denial of bail and a higher sentence ceteris paribus (Yong, 2018). There are around three levels to interpret the need and risk scales with the outcomes wholly, which use the crime-producing issues and sentencing guidelines to start with, environmental issues, skill issues, and criminogenic findings to deepen the analysis—multitudes of criminological theories to underpin the final working.
The problems of importing such an algorithm become apparent immediately. The questionnaire and data collection process poses a challenge for many accused from educationally deprived communities who might need help to capture the gravitas of the questions at hand. The police force is known for requiring more checks and balances and possibly false information. Moreover, the normative data is too few for an algorithm to train in an Indian Context. Furthermore, the norms, groups, and weightage added to categories would have to be reworked entirely in Indian society with overlapping caste, religion, and even tribal systems and practices. It needs to account for the expressivist theory of justice, which views punishment as a conventional device for expressing attitudes of resentment (Feinberg, 1965), and its relevance and validity in an Indian context transfixed upon several media trials. All this substantiates the call for an indigenous India-specific algorithm.
3. Mainstreaming, Ecosystem Creation and Acceleration
The purpose behind mainstreaming a process is to make a transformative change to achieve a remodelled system, here, one with AI. Progress is the goal, and intelligence is the way to progress. The Supreme Court AI Committee has been constituted. As early as 2021, the erstwhile CJI Sharad Bobde recognized AI’s aid and assistive capacity and introduced SUPACE (Supreme Court Portal for Assistance in Courts Efficiency). SUPACE is useful in augmenting the effectiveness of the judges in working on cases by making it efficient to extract relevant information, read case files, and draft case documents (Rituraj & Singh, 2021). Such initiatives are a welcome step; however, they are still confined to a minimal space. Nevertheless, the Supreme Court hopping on the AI bandwagon gives it credibility and opens discourses.
A study used Pune Districts and Sessions Court as a straw in the wind to extrapolate the number of bail applications Indian courts get in a year. The Pune court handled 5337 bail applications in 2020. There are at least 748 District & Sessions courts. The number of bail applications handled by the system in a year is 748*5337= 39,92,076, as a ballpark estimate (Srinivasa Rao & Gore, 2023). So, this zone is always a potential bottleneck. State governments have taken the lead in this area where the Uttar Pradesh government has launched an artificial intelligence-powered mobile application ‘Trinetra,’ which uses facial identification based on a 500-thousand-strong criminal database for gang identification (Ministry of Electronics and Information Technology, 2022), although the opaqueness and the ‘black box’ problem ensure ambiguity.
However, this has enthused the central government. This was evident in the recently held G20 conferences where the Minister of State for Electronics and Information Technology relayed the government’s will to assemble and allow usage of significant forms of the anonymized non-personal dataset collected and harmonized under a ‘Data Governance Framework’ policy to stimulate the growth of startups working in the AI sector, thus, developing an open AI-positive ecosystem (Anonymous, 2023). A highly relevant foundational tool by the NIC is the National Judicial Data Grid (NJDG) portal, a repository of data pertaining to all court cases, including onboarding the ones at every district and taluka level court. It provides statistics related to the case types, the pendency of cases, the stage of the cases, etc., updated daily (Ministry of Law and Justice, 2023). Thus, there are building blocks in place to build an ecosystem where innovators can create socially desired products.
4. Deployment and Recommendations
There are no laws in India concerning AI regulation. The space is open for deploying the aforementioned Indigenous India-specific undertrial management system. The government has offered access to anonymized non-personal data, and the judiciary has operationalized a comprehensive database like the NJDG, both of which can help provide datasets. The data collection from the undertrials has to be accurate, too. We need representative Indian parameters and a responsive model, which means following the risk-needs-responsivity approach such that interventions should not only manage the risk of recidivism but also help individuals achieve their primary goals (United Nations Office on Drugs and Crime, 2018b). We need a process that ensures data collection is not corrupted. Huikahi’s selective circle approach from Hawaii reduces the risk of recidivism by creating a circle of the incarcerated individual, their close ones, and prison officials to map out their post-prison life (United Nations Office on Drugs and Crime, 2013). I propose that such an approach be used to aid the incarcerated individuals when filling out the questionnaire. The prisoner files can be filled out upon admission by social workers, instituting a system of checks and balances that attest to the accuracy of the data.
One of the main issues with COMPAS has been that the algorithm is proprietary, and the individuals are faced with a black box problem with the absence of transparency or explainability. I propose a Digital Private Infrastructure (DPI) approach for the same which creates an efficient and independent ecosystem that makes technological goods public by hosting them on the India stack, an open-source framework (Kriplani, 2023). This means that the proposed program would be decoupled from corporations while garnering resources from the government. I suggest that for deployment, we look at Brazil’s strategy, which has fostered innovation by authorizing the courts under its National Council of Justice to develop their own models (Guthrie, 2024). This decentralized approach will ensure more competition and more players while having a centralized judicial oversight committee that makes sure the models pertain to the standards and communicate with each other.
5. Ethical Considerations and Conclusion
The introduction of the proposed India-specific AI is meant to aid the legal system in exercising this acumen more effectively. So, human oversight has to be a mainstay of the new system design to make sure that the interests of innocent defendants find adequate redressal rather than being buried in an unaccountable code. However, the active involvement of legal practitioners also has other broader implications, such as potentially feeding into an automation bias (propensity to privilege recommendations from automated decision-making programs) or bias blind spot (the cognitive bias where one delineates the effect of biases on others’ judgment, yet is unable to check the biases on one’s judgments). So, even with human oversight, one can sense the possibility of a value lock-in. In this situation, precedents consistently turn the legal status quo into a rigid structure to the detriment of jurisprudential evolution, not at pace with the evolving society. This introduces us to technological due process, where system design itself has to incorporate and evolve mechanisms for accountability and oversight to uphold the essence of the judicial process and its function of respecting and validating rights and liberties (Citron, 2008). The proposed changed system of a solid judicial AI oversight committee overseeing a decentralized system with the added imperative of confidential third-party assessment is the foundation for a step in the right direction in light of the ethical considerations.
References
Abson, D. J., Fischer, J., Leventon J., & Newig, J. 2016. Leverage Points for Sustainability Transformation. Ambio, 46(1), 30–39. [online] Available at: https://doi.org/10.1007/s13280-016-0800-y.
Sahoo, N. & Jain, V. (2015). Justice System in Crisis: The Case of India’s Undertrial Prisoners. (online) orfonline.org. Available at: https://www.orfonline.org/research/justice-system-in-crisis-the-case-of-india-s-undertrial-prisoners.
Yong, E. (2018). A Popular Algorithm Is No Better at Predicting Crimes than Random People. (online) The Atlantic. www.theatlantic.com/technology/archive/2018/01/equivant-compas-algorithm/550646/
Feinberg, J. (1965). THE EXPRESSIVE FUNCTION OF PUNISHMENT. The Monist, [online] 49(3), pp.397–423. Available at: https://www.jstor.org/stable/27901603.
Rituraj, V. & Singh, S.K. (2021). Supace: an Artificial Intelligence Solution for Reforming Indian Judiciary. International Journal of Management and Applied Science (IJMAS), Volume 7 (Issue 11), pp.17–20.
Srinivasa Rao, A.S.R. & Gore, A.P. (2023). Indian Courts of Law Can Benefit Immensely by Adopting Artificial Intelligence Methods in Bail Applications for Speedy and Accurate Justice. Handbook of Statistics, Volume 49, pp.227–236.
Ministry of Electronics and Information Technology (2022). 75 @ 75 India’s AI Journey. Pp.32.
Anonymous, (2023). How ready is India for ‘Responsible AI’? • Software Freedom Law Center, India. [online] Available at: https://sflc.in/how-ready-is-india-for-responsible
Ministry of Law and Justice. (2023). The flagship e-Courts project completes full circle with Supreme Court of India onboarding the National Judicial Data Grid portal. https://pib.gov.in/PressReleasePage.aspx?PRID=1957318
United Nations Office on Drugs and Crime (2018b). The Prevention of Recidivism and the Social Reintegration of Offenders. CRIMINAL JUSTICE HANDBOOK SERIES. [online] Available at: https://www.unodc.org/documents/justice-and-prison-reform/18-02303_ebook.pdf.
United Nations Office on Drugs and Crime (2013). Handbook on strategies to reduce overcrowding in prisons. CRIMINAL JUSTICE HANDBOOK SERIES. [online] Available at: https://www.unodc.org/documents/justice-and-prison-reform/Overcrowding_in_prisons_Ebook.pdf.
Kriplani, M. (2023). Private Innovation in Digital Public Infrastructure. [online] Gateway House. Available at: https://www.gatewayhouse.in/private-innovation-in-digital-public-infrastructure/
Guthrie, A. (2024). Brazil’s Overwhelmed Judiciary, Desperate for Help, Turns to Artificial Intelligence. [online] Law.com International. Available at: https://www.law.com/international-edition/2024/01/16/brazils-overwhelmed-judiciary-desperate-for-help-turns-to-artificial-intelligence/?slreturn=20240401075125
Citron, D. (2008). Technological Due Process Technological Due Process. Washington University Law Review, [online] 85(6), p.1249.