USE CASE: PRISON REFORM

Humans and AI working together to improve lives

Novus logo

 

Novus is the largest secure-training provider in the UK prison system. Novus coordinates education and rehabilitation services for over 60,000 prisoners in England & Wales.

Delivering consistent, high-quality provision across 34 prisons generates a lot of admin for front line-staff. Tutors lose up to 60% of their working week on administrative tasks associated with diagnosing learner needs, tracking progression and recognising achievement. By freeing tutors from the burden of administration, could Novus buy tutors more time with students? 

Our team set-out to help Novus embed document-understanding AI into these vital processes. The project forms part of a wider initiative to introduce large-scale digital-transformation in the prison system. 

 

Collaboration objectives:

  • Coordinate assessment standards across the secure estate

  • Deliver better, more targeted support to prisoners

  • Automate formative assessment and progress tracking activities

Enhancing these processes can make a big difference to people’s lives. Every improvement in diagnosing learning needs equates to better planning decisions for those students. Simplifying the process of recognising achievement increases a prisoner’s life chances upon release. Improving the quality and consistency of provision leads to better, more reputable qualifications. Every minute saved in admin equates to a minute more quality time that can be made available to students.

 

 

42 minute time-saving per worksheet

 

A detailed manual review of a 16 question worksheet took an average of 47 minutes to complete. Typical time for the AI was 32 seconds. Tutors would still need to spend 3-6 minutes reviewing the evidence and drawing educational conclusions.

23.1% improvement in grading consistency

 

Without AI, two tutors evaluating the same piece of work were 48.1% likely to assign the same score. The ‘human + AI’ model achieved a grading consistency of 71.2%.  This equates to a 23.1% improvement in the detection of learning needs.

Digitisation of paper-based assessment

 

Working in close-collaboration with Novus, the Fluence team developed a process to fully digitise the assessment and review of paper-based work samples.

 

Accurate detection of ‘curriculum coverage’

 

Trained on curriculum textbooks, the AI succeeded in detecting which aspects of the curriculum any student had covered and in what level of detail. The software is in a position to track when sufficient evidence has been produced by a learner to justify a qualification.

Delivered a framework for 'human + AI' collaboration

 

The framework allows tutors and AI to develop consensus on how work should be assessed. It ensures that human experts and AI are approaching work-review tasks in a similar way and allows humans and AI to operate on an interchangeable basis.

(‘AI can moderate human decision’ AND ‘human can moderate AI-derived decision’)

Download the Novus impact evaluation

Fluence recently completed a large-scale AI project with the largest training providers in the UK prison system.

 

Access the report using the link below:

 

Simplified AI-training

Making it easy for subject-experts to train AI

The task of getting data-ready can seem daunting. Prison work is completed using pen and paper, which complicated the task of developing a Novus data lake. Our team developed an accessible process for converting paper-based work into machine readable text before allowing tutors to train the AI to reflect their logic and values.

 

  • Turned AI-training into an accessible, non-technical process
  • Systematic approach for generating reliable predictive models 
  • Aggregating and moderating feedback from multiple tutors

    AI-assisted document feedback

    Novus-specific feedback, in 32 seconds

    Time is a precious resource for tutors. Following the AI-training work, predictive models were connected to Fluence’s document-feedback service. When a new piece of work is submitted, the system automatically generates dozens of predictions (as trained by the tutors themselves). Tutors can add colour-coded actions to the system, which triggers when a specific outcome has been predicted on a document (e.g. if ‘score’ is greater than 4, trigger action ‘Well done’ in green). 

    • Autonomously track multiple learning needs at once
    • From ‘work-submitted’ to ‘detailed-feedback’ in 32 seconds
    • Put students in charge of their own education
    • API framework to embed Fluence conclusions into any business processes

    Automated analytics

    Evidencing that just happens

    As part of Novus’ ongoing work to deliver digital transformation in prisons, we have succeeded in turning the tedious task of assessing work into a passive process that runs automatically in the background. Tutors can gather actionable intelligence about any individual or group they are teaching simply by logging onto the system.

     

    • Pinpoint learning needs and education priorities
    • Obtain trended analysis of needs, abilities and performance
    • Understand patterns of behaviour on an individual, group and prison level 

    Data control

    Understand precisely how decisions are made

    AI is only useful if operators can trust the decisions it makes. Together with Novus, the team audited the precise capabilities of the AI, including defining how AI-derived decisions were made. Service allows leadership teams and front-line staff to build-up trust in the algorithms.

    • Ongoing maintenance of predictive models
    • Track outlier cases
    • Understand how different inputs influence the model
    • Identify bias and inconsistency in the data
    • Train-out unwanted influences from predictive models 

    Curriculum oversight

    Big-picture curriculum understanding

    After connecting student work to Fluence’s Consul suite, the team was able to audit curriculum performance across all participating prisons. The system was used to understand the relationship between the curriculum and student performance. For example, the weakest ‘terminology’ responses were all linked to ‘Financial Planning.’ Developing concrete answers to qualitative questions allows education leads to rethink processes and to make strategic adjustments to their education plans. 

     

    • Understand strengths and weaknesses in curriculum offer
    • Obtain trended analysis of needs, abilities and performance
    • Understand patterns of behaviour on an individual, group and prison level 

    Anonymous tutor survey

     

    “It was thoroughly enjoyable. For the first time I felt part of the future and positively contributing to how that may look. I felt involved and a worthwhile employee. Fluence were professional and wholly supportive whilst engaging and offering feedback to me on my efforts. I would work with them again without any apprehension.”

    Steve Grix, Digital Learning Lead, Novus

     

    “Our collaboration with Fluence is a key sustainable development investment; looking to the future of integrating technology into our teaching, learning and assessment. Their technology is opening up new possibilities for the future; to functionally improve the teachers quality of life and the experience of the learner. Many of us that work with EdTech now know that it has great potential to improve the assessment process. Fluence’s technology has the potential to transform the assessment process for all educators; it is that significant.

    It has been wonderful working with Fluence. They are an exciting and innovative bunch and have found several solutions to deal with the challenges of our context. I have been fortunate enough to speak with them regularly and have taken so much away from our collaboration. I look forward to our continued work together and the day where we can leverage what we have worked upon to transform our delivery.”

    Download the Novus impact evaluation

    Fluence recently completed a large-scale AI project with the largest training providers in the UK prison system.

     

    Access the report using the link below:

     

    With our thanks to the Ufi VocTech Trust for making this project possible.