Blogpost by Temitope Lawal (Northumbria University) on behalf of ‘PROBabLE Futures: Probabilistic AI Systems in Law Enforcement Futures’ RAI Keystone Project
Artificial intelligence (AI) is no longer a distant frontier for law enforcement – it is here, reshaping how crimes are investigated, calls are triaged, and public safety is maintained. Over the years, significant progress has been made in developing pipelines that aim to increase the explainability and trustworthiness of autonomous systems. Although the term AI is widely used – particularly following the rise of large language models – it is important to recognise that AI encompasses a diverse range of techniques. These vary significantly from simple deterministic systems to complex multi-modal transformers, with vastly different levels of transparency.
As highlighted during the first external event of the PROBabLE Futures Project on 20 January 2025, the journey toward responsible AI deployment is far from straightforward. Beyond the technical innovations, it is a complex balancing act between efficiency, technical validity, equity, and accountability; one that requires a systemic, interdisciplinary, and deeply responsible approach.
Here, we unpack the key themes from the discussion and assess the challenges and opportunities of embedding AI in law enforcement in a way that truly serves society.
The promise and perils of probabilistic AI
One of the focal points of the event was the increasing use of probabilistic AI systems – tools that work with likelihoods rather than certainties. These systems are already being deployed to triage non-emergency calls, analyse digital evidence, and even identify patterns of grooming behaviour. The possibilities are exciting, but the risks are equally significant.
Take, for instance, the “chaining” effect of AI systems – a point that emerged as a critical concern in the discussion. When outputs from one AI system feed into another or into a series of AI-informed decisions, errors or biases can compound across the justice process. A flawed algorithmic decision at the investigation stage could, for example, inadvertently skew case prioritisation, leading to disproportionate resource allocation or sentencing disparities. These ripple effects underline the importance of stepping back to view AI not as a set of isolated tools, but as part of a larger system with interconnected impacts. This systemic perspective is not just an academic exercise; it is a necessary step toward designing AI frameworks that are both effective and just.
Learning from practice
The event provided concrete examples of AI in action, offering a window into both its potential and its limitations.
In one case study, an AI-powered voice assistant was deployed to handle non-emergency calls for a major police force. The results were impressive – abandoned call rates dropped from 63% to less than 1%, and average wait times fell to just eight seconds. Yet, the team behind the project was quick to caution against overselling the technology; factors and challenges included recognising key words, accommodating varying accents and evaluating the public response. AI alone did not solve the problem of improving the force’s response to 101 calls; it was part of a broader effort involving new leadership, training, and processes. This is a critical reminder that AI is a tool, not a silver bullet.
Another case study focused on the use of AI in digital forensics, where it has significantly accelerated the detection of digital child exploitation materials. AI tools have proven effective in reducing the time needed to process vast amounts of data, identifying content even when hash values are altered. Yet, even here, challenges remain. Current tools struggle to process altered images or adapt to evolving threats, highlighting the need for continuous refinement and robust data governance. It was also noted that these tools require further advancements, particularly in natural language processing, to address the growing demands of data analysis in investigations.
These examples serve as a dual reminder: AI can deliver real value, but its deployment must be accompanied by humility, realism, and a commitment to ongoing improvement.
Taking the industry perspective into account, deploying AI in critical sectors requires a holistic approach and careful consideration, backed up by qualitative measures of the system's trustworthiness. The nature of these qualitative measures is the most problematic part of defining a pre-deployment assessment.
The very nature of autonomous systems (and, more broadly, decision-making itself) is founded on the idea of choosing the most optimal solution given the data available to the AI tool at the time. However, when evaluated in hindsight with additional context and information that should have been considered, the decision may not always be the correct one. Self-driving cars are the best illustration of that situation. In situations requiring rapid action, an autonomous vehicle must make a decision that carries unavoidable consequences. While the action may be optimal given the information available at the time, it may be later judged as inappropriate by an external observer who has a complete understanding of the circumstances in which the decision was made, such as an inspector assessing a crash site.
How do we make AI accountable?
A recurring question throughout the event was: how do we ensure accountability in AI? It is a question without easy answers, especially when dealing with systems that operate as “black boxes,” making it difficult to understand or challenge their outputs.
Several suggestions emerged during the discussion. Transparency mechanisms, such as audit trails and explainability protocols, are essential. But as one contributor insightfully pointed out, accountability is not just about building better algorithms; it is also about the humans and systems around them. Policymakers, developers, and law enforcement agencies must work together to ensure that AI complements (not replaces) human judgement.
Another important consideration came from the call for participatory oversight. By involving communities directly affected by law enforcement, we can move beyond a top-down approach to AI governance. An oversight committee that blends expert knowledge with lived experience could offer a model for more inclusive and responsible decision-making.
A framework for responsible use of AI
What does a responsible AI framework look like in practice? The event reinforced the idea that it cannot be static or one-size-fits-all. Instead, it must be flexible, interdisciplinary, and responsive to both technological advancements and societal changes. The PROBabLE Futures Project is taking steps in this direction, by working to develop a holistic, right-respecting framework to steer the deployment of probabilistic AI within law enforcement, creating a coherent system, with justice and responsibility at its heart. In this regard, AI systems must be assessed not in isolation, but as part of a larger ecosystem. This means examining their cumulative impact across the justice process and ensuring they align with broader societal goals. Furthermore, AI frameworks must evolve alongside technological and legal developments. This requires embedding mechanisms for regular review and iteration, which in turn ensures that the framework remains relevant and effective.
Looking ahead
As the event drew to a close, one message stood out: responsible AI is not about chasing perfection, but about doing better. It is about learning from missteps, listening to diverse voices, and committing to a process of continuous improvement. For the PROBabLE Futures Project, this means not just designing a responsible AI framework, but testing it, refining it, and sharing those lessons widely. Whether through storytelling methodologies to capture lived experiences of those disproportionately impacted by the criminal justice system (such as marginalised communities) or collaborations with partners across different segments, the project is laying the groundwork for a future where AI serves justice, not expediency.
The road ahead is challenging, but it is also filled with opportunity. We can harness the power of AI to build safer, fairer communities by anchoring our efforts in transparency, inclusivity, and accountability. For those interested in joining this journey, the PROBabLE Futures Project invites collaboration and dialogue as we continue to shape the future of AI in law enforcement.
About PROBabLE Futures
PROBabLE Futures, one of Responsible AI UK’s keystone projects, is a four-year interdisciplinary research initiative focused on evaluating probabilistic AI systems across the criminal justice sector. The project aims to develop a responsible, practical, and ‘operational-ready’ framework in collaboration with multiple criminal justice partners.
Back to top
