RUDI
To bridge the gap between the academic evidence base and everyday policing we provide a framework offering practical guidance to support forces considering the use of complex machine learning algorithms, where the outcomes have potential consequences for members of the public.
Conceptualisation
Public-facing applications require extra diligence and documentation. Therefore, an essential step before embarking on producing a public-facing model is to decide if this is the best use of the resources and that there is capacity to integrate and maintain the product after development.
From prototyping to implementation, the process and decisions require comprehensive modelling. The Rationale section will guide you.
Care and due process during the development stages can lead to better models and more ethically responsible applications.
​
​
​
Data integrity is paramount in systems where it is stored in disparate databases and managed by various organisations.
Decisions in deployment of the technologies including ensuring continaul model integrity, user experience design, and the operantional interventions that informed the models are key to ethical use of AI tools in policing.
RUDI is a practical framework police forces can follow to build and implement data modelling. It is specific to models that use machine learning techniques, the outcome of which are likely to drive police action and have possible consequences for members of the public (e.g., models that prioritise people or predict future criminal behaviour or locations). These models require compliance with governmental transparency and data protection guidelines, which are compatible with RUDI. It covers four main stages:
​
Rationale: Documenting the process, making decisions explicit and justifying actions during unification, development and implementation
Unification: Merge data sources together for modelling and ensure validity and reliability of data
Development: Build and test models, evaluating for bias, performance and limitations and choose preferred model
Implementation: How the model feeds into current practice and how it will be maintained over time
Development is not a linear process. As the process continues the rationale, need for unification and development may change, evolve and feed into each other. However, conceptualisation of the problem always comes first, and implementation of the proposed solution always comes last.
​
There is no ‘one right way’ to develop data models, it is a series of decisions that must be made whilst balancing multiple competing concerns. RUDI is a framework that sets out some of the decisions involved and provides a way for forces to document the process to improve:
​
-
Transparency: being honest and open about decisions and trade-offs),
-
Justifiability: making decisions defensible
-
Lawfulness: using proportionate methods and data
-
Accountability: being answerable to critique
It does not prescribe an exact formula for algorithmic modelling, nor does it reduce the need for in-house domain or data expertise (see Data Scientist Expertise for details of data expertise required).