Integrating Artificial Intelligence (AI) into the legal sector signifies a significant shift from traditional, labour-intensive methods to a more efficient, technology-driven approach. This transformation goes beyond mere convenience; it represents a fundamental change with the potential to revolutionise access to legal services and improve procedural efficiency. However, with these benefits come various ethical challenges that necessitate careful and thorough consideration. As AI reshapes the legal landscape, it is crucial to thoughtfully address these concerns to ensure that technology adoption upholds, rather than undermines, the principles of justice and fairness.

Innovative Horizons in the Legal Domain

Historically perceived as slow to adopt technological advancements, the legal sector has now emerged as a leader in integrating AI technologies.[1] These innovations are pivotal in bridging the accessibility gap in legal services, offering sophisticated tools for drafting wills, amending contracts, and facilitating participation in remote depositions. AI’s applications extend further, assisting litigants with digital divorce templates and enabling law firms to use data analytics to predict outcomes in complex legal disputes.[2]

The transformative potential of AI in family law is particularly noteworthy. By analysing extensive historical and legal datasets, AI can generate equitable asset division plans, parenting agreements, and divorce terms, thereby revolutionising the field and addressing longstanding challenges in family law.[3] This capability expedites the legal process and introduces a systematic approach to dispute resolution—areas traditionally relying heavily on human judgment.[4] These advancements promise significant benefits, instilling optimism about the future of legal services. However, they also necessitate careful consideration of AI implementation’s ethical and practical challenges in sensitive legal contexts.[5] It’s important to note that while AI can assist in these processes, human oversight is crucial to ensure that the technology is used responsibly and ethically and to maintain the importance of human judgment in the legal field.[6]

AI’s Role in Legal Decision-Making

AI is being extensively used in the legal field and can be classified into two main categories: knowledge-based systems and statistical learning technologies. Knowledge-based or expert systems use predefined rules to deduce behaviours within specific domains.[7]

Conversely, statistical learning technologies, such as machine learning and deep learning, improve decision-making capabilities by continuously analysing new data.[8]

AI implementation in legal contexts is widespread. From e-discovery platforms that handle large amounts of data to predictive software like COMPAS (Correctional Offender Management Profiling for Alternative Sanctions) used by adjudicators, AI is transforming legal professionals’ access to, analysis, and application of information.[9]

These tools promise to enhance legal research efficiency and improve legal outcomes’ accuracy by extracting insights from a broader range of data than were previously available.

Ethical and Practical Implications

While AI’s capabilities inspire considerable enthusiasm, they are accompanied by significant ethical concerns, particularly regarding algorithmic bias. The data that informs AI algorithms can reinforce existing societal biases, leading to skewed outcomes in sensitive legal areas such as child custody decisions and bail settings. A notable real-world example is the COMPAS algorithm, as mentioned above, used in the U.S. criminal justice system to assess the likelihood of a defendant reoffending. Studies have shown that COMPAS sometimes predicts higher recidivism risks for Black defendants compared to white defendants, even when controlling for similar prior criminal histories.[10] This case underscores the dangers of algorithmic bias when AI operates without sufficient oversight.

The discourse advocates for a “human-in-the-loop” (HITL) approach to address these concerns, ensuring that AI operates closely with human expertise.[11] This methodology mitigates biases and accommodates the nuances of individual cases, emphasising the necessity of sustaining the human element in legal decision-making. In this way, technology serves as a facilitator of human judgment rather than a surrogate.

Despite AI’s advantages, its application in the legal field is fraught with challenges, particularly regarding ethics and bias. The reliance on historical data and algorithmic decision-making risks perpetuating existing biases, compromising legal outcomes’ equity. This issue is further exacerbated by the opaque nature of some AI systems, raising concerns about accountability and transparency. The discussion underscores the need for transparency, accountability, and the incorporation of diverse datasets to address these challenges. It calls for developing AI systems that are both technologically sophisticated and ethically sound, capable of rendering equitable and unbiased decisions. By adopting these measures, the legal profession can ensure that AI supports justice and fairness rather than undermining them.

The Future Trajectory of AI in Legal Decision-making: The Role of the Human-in-the-Loop Approach

Looking ahead, integrating AI into legal practices offers significant potential to transform the delivery of legal services. AI’s ability to enhance efficiency, increase accessibility, and improve accuracy holds the promise of revolutionising the legal landscape. However, this technological evolution also brings a profound responsibility to ensure that AI is developed and deployed in a manner that upholds the highest ethical standards. The HITL approach emerges as a critical framework for achieving this balance, ensuring that AI operates as a complement to, rather than a substitute for, human expertise in the legal realm.

One of the HITL approach’s most significant strengths is its ability to mitigate the ethical risks associated with AI, particularly the risk of algorithmic bias. AI systems often rely on large datasets containing historical biases—such as racial or gender disparities—that could be perpetuated if AI were to make autonomous decisions.[12] Legal professionals can assess and correct any biases AI algorithms might introduce by involving a human in decision-making. This human oversight ensures that legal outcomes remain fair and equitable, especially in sensitive areas such as child custody, bail determinations, or sentencing decisions, where ethical considerations are paramount.[13]

Moreover, the HITL approach ensures that human expertise is applied to the nuanced interpretation of legal complexities. The legal profession often requires judgment beyond data analysis, particularly in areas where laws are subject to interpretation, and cases present unique circumstances.[14] AI excels at quickly processing vast amounts of information but cannot fully grasp legal disputes’ intricacies and human elements. The involvement of human legal experts in the HITL approach ensures that AI-generated insights are interpreted within the broader context of the law, allowing for more refined and appropriate decisions.[15]

Another key advantage of the HITL approach is its role in maintaining transparency and accountability. AI systems are sometimes criticised for their “black box” nature, where the decision-making process is opaque and difficult to explain. This lack of transparency is particularly problematic in legal contexts, where the ability to justify decisions is essential.[16] The HITL approach mitigates this issue by ensuring that a human is always involved in the final decision, providing a necessary layer of accountability that AI systems alone cannot offer. By having humans review AI recommendations, the legal profession can maintain clear documentation and justification for decisions, thereby improving transparency and building trust in AI-assisted processes.[17]

However, the HITL approach has its challenges. One significant drawback is the potential for reduced efficiency. AI is often lauded for its ability to process information quickly and autonomously, which can significantly improve efficiency in legal practices.[18] Yet, the HITL approach necessitates continuous human intervention, which can slow down processes that AI could otherwise expedite. For routine tasks—such as document review or contract analysis—where AI might operate effectively without human oversight, introducing a human in the loop can diminish some of AI’s speed benefits. Balancing the need for human oversight with the efficiency gains promised by AI is a critical challenge in implementing the HITL approach.[19]

Another potential weakness of the HITL approach is the risk of over-reliance on AI outputs. While the goal of HITL is to maintain human control over the decision-making process, there is a concern that legal professionals might defer too heavily to AI recommendations over time, assuming that the technology is infallible.[20] This could reduce critical thinking and analytical skills, with legal professionals becoming passive facilitators rather than active decision-makers. Ensuring that humans remain vigilant in their oversight of AI and do not simply rubber-stamp AI outputs is crucial for maintaining legal practice quality.[21]

The HITL approach can also be resource-intensive, requiring significant investment in technology and human resources. Law firms and legal departments must invest in AI systems and train their staff to oversee them effectively. Smaller firms or those with limited resources may struggle to implement a comprehensive HITL approach, particularly when it comes to ensuring that human oversight is meaningful and continuous. This hybrid model, in which humans and AI collaborate, can increase operational costs and limit the accessibility of AI technologies for some legal professionals.[22]

Finally, scalability is another challenge associated with the HITL approach. AI is often praised for its ability to scale legal processes, enabling firms to handle larger volumes of cases with incredible speed.[23] However, the requirement for human review in every AI-assisted decision can limit the potential for scaling. In high-volume legal environments, such as contract law or compliance work, where AI could automate large-scale tasks, human intervention may slow down the number of cases that can be processed simultaneously.[24] This presents a challenge in balancing the benefits of AI with the need for human oversight in high-volume scenarios.[25]

In summary, the human-in-the-loop approach is an essential framework for the responsible integration of AI into legal practice. While it offers significant advantages such as ethical safeguards, nuanced decision-making, accountability, and flexibility, it also presents challenges such as potential reductions in efficiency, resource intensiveness, and scalability issues. The future of AI in law will be determined by how effectively the legal profession can address these challenges, ensuring that AI complements human expertise without compromising the integrity of legal practice. By striking the right balance between AI’s capabilities and human oversight, the legal sector can fully leverage the transformative potential of AI while upholding justice, fairness, and ethical responsibility in legal services.

[1] Iain Rodgers, John Armour, and Mari Sako, ‘How Technology Is (or Is Not) Transforming Law Firms’ (2023) 19(1) Annual Review of Law and Social Science 299 https://doi.org/10.1146/annurev-lawsocsci-111522-074716

[2] Gulnara Said and others, ‘Adapting Legal Systems to the Development of Artificial Intelligence: Solving the Global Problem of AI in Judicial Processes’ (2023) 1(4) International Journal of Cyber Law https://doi.org/10.59022/ijcl.49

[3] A. Zafar, ‘Balancing the Scale: Navigating Ethical and Practical Challenges of Artificial Intelligence (AI) Integration in Legal Practices’ (2024) 4 Discover Artificial Intelligence 27 https://doi.org/10.1007/s44163-024-00121-8

[4] E. Alsamhan, ‘AI and Online Dispute Resolution: Mediation’ (2023) 4(13) Journal of Scientific Development for Studies and Research (JSD) 283 https://doi.org/10.61212/jsd/113

[5] M. Agbese and others, ‘Implementing AI Ethics: Making Sense of the Ethical Requirements’ in Proceedings of the 27th International Conference on Evaluation and Assessment in Software Engineering (June 2023) 62 https://doi.org/10.1145/3593434.3593453

[6] Claudio Novelli, Mariarosaria Taddeo, and Luciano Floridi, ‘Accountability in Artificial Intelligence: What It Is and How It Works’ (2024) 39 AI & Society 1871 https://doi.org/10.1007/s00146-023-01635-

[7] I.H. Sarker, ‘Multi-Aspects AI-Based Modeling and Adversarial Learning for Cybersecurity Intelligence and Robustness: A Comprehensive Overview’ (2023) 6(5) Security and Privacy e295 https://doi.org/10.1016/j.chb.2022.107468

[8] Bahman Zohuri and Farhang M. Rahmani, ‘Artificial Intelligence Driven Resiliency with Machine Learning and Deep Learning Components’ (2023) 1(1) Japan Journal of Research

[9] Francesco Lagioia, Roberta Rovatti, and Giovanni Sartor, ‘Algorithmic Fairness Through Group Parties? The Case of COMPAS-SAPMOC’ (2023) 38 AI & Society 459 https://doi.org/10.1007/s00146-022-01441-y

[10] Jackson E, Mendoza C. Setting the record straight: what the COMPAS core risk and need assessment is and is not. Harvard Data Sci Rev. 2020. https://doi.org/10.1162/99608f92.1b3dadaa

[11] Katerina Kyriakou and Jahna Otterbacher, ‘In Humans, We Trust’ (2023) 3 Discover Artificial Intelligence 44 https://doi.org/10.1007/s44163-023-00092-2

[12] ‘Part Front Matter for PART 2 Human-Centered AI Framework’ in Human-Centered AI (Oxford, 2022; online edn, Oxford Academic, 17 February 2022) https://doi.org/10.1093/oso/9780192845290.001.0001

[13] Sharon Alon-Barkat and Madalina Busuioc, ‘Human–AI Interactions in Public Sector Decision Making: “Automation Bias” and “Selective Adherence” to Algorithmic Advice’ (2023) 33(1) Journal of Public Administration Research and Theory 153

[14] S. Kumar and others, ‘Applications, Challenges, and Future Directions of Human-in-the-Loop Learning’ (2024) 12 IEEE Access 75735 https://doi.org/10.1109/ACCESS.2024.3401547

[15] I. Glenn Cohen and others, ‘How AI Can Learn from the Law: Putting Humans in the Loop Only on Appeal’ (2023) 6 npj Digital Medicine 160 https://doi.org/10.1038/s41746-023-00906-8

[16] F. Khan, ‘Human-in-the-Loop Approaches to Improving Machine Translation’ (2024) 7(1) Academic Journal of Science and Technology 1

[17] A. Akbar and O. Conlan, ‘Towards Integrating Human-in-the-Loop Control in Proactive Intelligent Personalised Agents’ in Adjunct Proceedings of the 32nd ACM Conference on User Modeling, Adaptation and Personalization (June 2024) 394

[18] C.O. Retzlaff and others, ‘Human-in-the-Loop Reinforcement Learning: A Survey and Position on Requirements, Challenges, and Opportunities’ (2024) 79 Journal of Artificial Intelligence Research 359

[19] O. Gómez-Carmona and others, ‘Human-in-the-Loop Machine Learning: Reconceptualizing the Role of the User in Interactive Approaches’ (2024) 25 Internet of Things 101048

[20] M. Taherisadr and S. Elmalaki, ‘HILT: Personalized and Adaptive Privacy-Aware Early-Exit for Reinforcement Learning in Human-in-the-Loop Systems’ (2024) arXiv preprint arXiv:2403.05864 https://doi.org/10.48550/arXiv.2403.05864

[21] Roman V. Yampolskiy, ‘On Monitorability of AI’ (2024) AI Ethics https://doi.org/10.1007/s43681-024-00420-x

[22] S. Järvelä, A. Nguyen, and A. Hadwin, ‘Human and Artificial Intelligence Collaboration for Socially Shared Regulation in Learning’ (2023) 54(5) British Journal of Educational Technology 1057 https://doi.org/10.1111/bjet.13325

[23] D. Agarwal, Learning and Inferencing Challenges in Human-in-the-Loop Decision Systems (2024) https://hdl.handle.net/2097/44182

[24] Y. Bengio and others, ‘Managing Extreme AI Risks Amid Rapid Progress’ (2024) 384(6698) Science 842 https://doi.org/10.1126/science.adn0117

[25] Sourati, J. and Evans, J.A., 2023. Accelerating science with human-aware artificial intelligence. Nature human behaviour, 7(10), pp.1682-1696.