Machine learning (ML) could provide a valuable tool for intelligence analysts in national security but will need care in its application, according to a new report from The Alan Turing Institute.
Its Centre for Emerging Technology and Security has looked at the issue through consultations with stakeholders and emphasises that much depends on the expertise of the analyst.
Titled Human-Machine Teaming in Intelligence Analysis, the report says ML could help analysts shift through vast amounts of data to triage and prioritise information, and that this provides a real potential to reduce risks in the use of the data.
But it adds that those implementing the technology need a good understanding of analysts’ real world work environments, as the way they will handle the output is highly specific to the context. There are also issues in the technical explainability of ML systems, which currently tends to be focused on mathematics, and the need for analysts to trust the technology and the output.
Tailoring explanations
The report’s recommendations include that explanations on the use of ML should be tailored to the person’s expertise, citing the example of data scientists requiring different explanations to analysts and oversight bodies.
It also says ML should be designed from the outset to be integrated into intelligence analysts’ toolsets and workflows, that they should be included in the prototyping and testing of models and associated graphical user interfaces, the language for discussing ML should be standardised, and data science should be offered as a support service to analysts.
In addition, effective adoption of ML requires a system-level approach that takes account of existing policies and practices including legal authorisations.
Anna Knack, lead author and senior research associate at the institute, said: “It’s time consuming for intelligence analysts to work through the vast quantities of data that come across their desks every day. Machine learning can speed up this process, filtering out the irrelevant information helping analysts to act on information much more quickly.
“The use of machine learning to support intelligence analysis presents new challenges such as how to present the right amount of technical information regarding the model's performance to the user, to ensure they maintain the appropriate level of trust in the system.
“The recommendations in this report are aimed at ensuring the responsible, effective and proportionate use of machine learning for intelligence analysis so national security agencies can effectively deploy this technology to help keep people safe.”
