Simona Pappalardo, Deputy Director, AI Enablement, Justice AI Unit; and Sumitra Varma, Deputy Director, Probations & Electronic Monitoring Data, Justice Digital Data and Science
Court transcripts writing themselves under human supervision, CCTV automatically
detecting violent incidents in prisons, the location data of people on probation being matched against police crime data - while these measures may have once sounded like science fiction, they are all projects being explored or piloted by the Ministry of Justice as part of its new AI drive.
The MoJ claims its growing use of AI has already saved staff tens of thousands of hours. As Justice Minister, David Lammy stated that the trials of one tool alone - Justice Transcribe - had freed up more than 25,000 hours for those working in the probation system.
The potential benefits of AI extend beyond recovered time. In a context where prison wardens quit at elevated rates due to high levels of violence, the process for victims to view their court records is long and costly, and reoffending rates remain significant, the ministry believes the technology can help safeguard the wellbeing of its own staff, offenders, and of the public.
However, as AI becomes more ingrained, the MoJ faces a problem familiar to departments across government: how to move quickly enough to realise the technology’s benefits while maintaining the security, oversight and public trust required when handling sensitive data.
We spoke to Simona Pappalardo from the ministry’s Justice AI Unit, and Sumitra Varma from its Data Directorate, about how they were navigating these concerns, and about how they were deploying AI more broadly.
____
The Ministry of Justice has long worked with traditional forms of data science and machine learning. Carried out through divisions such as the Data and Analytical Services Directorate, as well as through the wider Justice Digital, Data and Science teams, this work is centred around using analytics to predict, classify, and identify patterns. This in turn is used for functions ranging from reducing the time needed for investigations, to violence prevention, to changing offender behaviour, and beyond.
Following the more recent, rapid proliferation of AI tools, this preexisting work was joined by a new emphasis on frontier AI tooling, embodied by the establishment of a Justice AI Unit in 2024. Justice AI specialises in enterprise and generative AI, developing operational AI use cases - centered around overcoming real-world challenges - as well as integrating platforms such as Microsoft Co-Pilot and ChatGPT Enterprise to support drafting, summarisation, transcription and knowledge search.
Both strands operate through a “scan, pilot, scale” approach, which revolves around
continuously searching for new ideas and technologies and assessing their merit, before testing their efficacy in a low-risk, controlled environment, then scaling solutions that have been successful in pilots.
Perhaps the most high profile AI project to come out of the MoJ recently is Justice AI’s flagship initiative, Justice Transcribe. This tool uses speech recognition to transcribe, summarise, and structure records. While its use in courts is still being piloted, it has been scaled for use in probation, where it has led to a 50% reduction in note-taking time, according to the MoJ.
This tool is “the one [the team is] most proud of”, according to Simona Pappalardo, Deputy Director of AI Enablement in the MoJ, who added that it “helps staff spend less time on note-taking and more time engaging with their users”.
Another tool being assessed by the Justice AI Unit seeks to use AI to enhance CCTV analysis to create safer environments in prisons, courts, and other justice settings, while a project in the pilot stage is Ruby, a digital “red box” helping private offices across government manage complex workflows within a unified system.
In addition, the Data Directorate is piloting an acquisitive crime mapping tool, which compares the location of those on probation for acquisitive crimes with the location of recent offences of a similar kind, as well as a predictive analytics tool which estimates the number of violent incidents a prisoner is at risk of being involved in over the next year. The latter takes into account a prisoner's age and previous behaviour in custody, allowing staff to make better decisions to deal with their behaviour, according to the ministry.
However, despite its apparent success, the in-house development of advanced AI tools never comes without its challenges. This is all the more true for a body which handles as much sensitive information and holds as much ethical responsibility as the Ministry of Justice.
“What we do with [our] data is really important, not just for the people within the criminal justice system, but for the people who've been impacted by it, such as the victims, the public - everyone has an interest to know what happens to these people using that data,” said Sumitra Varma, Deputy Director in the Ministry of Justice.
She added: “People have to trust what we are doing, it has to be assured with the right processes, but at the same time, we need to make sure governance doesn’t just add a bureaucratic layer. It must empower delivery and outcomes.”
Balancing speed with an appropriate level of governance to ensure trust is thus a challenge, beyond simply guaranteeing that data foundations are sufficiently robust to enable AI to work effectively.
Another challenge is ensuring that understandable excitement about new technology does not transform into blind enthusiasm, leading to AI being deployed inappropriately.
Varma said: “I think that there's a perception, now especially, that AI is going to solve a lot of problems, and that it's a matter of ‘AI is the answer. Now, what is the question?’
“But that's not how it should be, we should actually find out what the problem and the user need is.”
It is to prevent this overenthusiasm that the MoJ makes a concerted effort to ascertain which approaches are likely to work prior to committing to a particular one, through its ‘scan, pilot, scale’ approach. Ensuring that people from a range of different disciplines are involved in the same project is also key to preventing blind spots, whilst at the same time aiding in achieving the goal of ensuring, as Pappalardo put it, that “security, fairness, transparency and accountability is built into every stage of development”.
“Bringing together operational colleagues, data protection, policy experts, legal, security ethics and frontline staff has been essential to design solutions that are both practical and trustworthy,” she said.
Varma added: “We work very closely with not just digital counterparts, but also policy, analytical and operational colleagues, because we want to make sure we get this right.”
In addition to a multidisciplinary team, Varma emphasises the importance of a responsible framework in ensuring that citizen needs are met. For the MoJ data science team, this framework provides guidelines for building AI tools, alongside ethical guardrails. According to Varma, the ministry also has robust peer review methods, with the work of one unit continually scrutinised by other teams.
For her, there are three principles other departments should learn from the MoJ’s
example. The first of these is the importance of strengthening data foundations and implementing shared ownership, while the second is the need to invest in AI. To invest in AI is to put the right level of effort and resources into ensuring the appropriate tools are being used, something which she said would allow technical debt to be avoided.
She said: “With most systems, you are creating some level of technical debt. With AI that is multiplied, you can end up creating a lot more technical debt without even realizing it.
“I'll take a small example of things like agents. You have so many departments being encouraged to use agents and try things, but do we have the right amount of governance to tidy up? To do housekeeping? To remove the agents that aren't working? To actually quality assure some of these things?”
Her third and final point concerns the importance of upskilling.
“Just having a tool is not enough,” she said, “you have to know how to use it.”
For both Varma and Pappalardo, AI is here to stay in the Ministry of Justice. It has the potential to assist staff in making decisions on a wider scale, reduce delays, and become so embedded in the digital services that people use that they
won’t know they are even interacting with AI.
That means that government must take the task of ensuring that it is deployed in an ethical and transparent way seriously.
“What would be critical is that we move forward with appropriate governance, transparency and stability,” said Pappalardo. “That's the central point.”