Exploring AI Safety: Membership Use Cases Across Industries
This insight provides a glimpse into various membership use cases in the realm of AI Safety, showcasing innovative applications across different sectors. Released on 09/05/2024, it highlights how organisations leverage artificial intelligence to address critical challenges and advance their respective fields.
Each use case demonstrates the transformative power of AI in tackling complex issues while prioritising privacy, efficiency, and improved outcomes. If you're interested in contributing a use case, please reach out to [email protected].
(Akrivia) SME AI Adoption in life sciences
Akrivia Health curates one of the largest psychiatric electronic health record (EHR) databases in the world, providing curation and AI-structuring services to 19 NHS Mental Healthcare Organisations (HCOs).
Problem: To find precision psychiatric treatment requires rich phenotypic data representing the full heterogeneity of patient states and care. Electronic health records can potentially provide this, but a particular challenge of NHS psychiatric EHRs is that ~85% of actionable information is recorded as unstructured text - difficult to analyse at scale, and too sensitive for third party access.
Solution: Akrivia’s NLP solution is a multi-stage pipeline, which uses AI in different ways, at different stages. Using large language models (like BERT) directly for named entity recognition (alongside rule-based/regex models. A key premise of our data structuring activity is to reduce the need for exposure to sensitive free-text documents in research. The NLP outputs meant researchers can leverage the rich transdiagnostic data captured in clinical free text without the need for manual review.
AI is helping Akvrira (and our NHS, academic and industry partners) make better use of routinely gathered health data, whilst preserving patient privacy.
Interested in reading more? You can do so here: Akrivia Health - life sciences.docx
_______________________________________________________________________
(Faculty AI) Using Machine Learning Techniques to Understand Parkinson's Disease
The Francis Crick Institute, a leading biomedical research centre in London, formed through a collaboration between prominent academic and research institutions.
Problem:
-
To build data science capabilities - giving Crick members the understanding and skills needed for independent use of data science techniques;
-
Apply data science skills - jointly working towards solving a specific technical problem to further build capabilities;
-
To gain traction - building momentum and enthusiasm for wider adoption of data science approaches. This initiative addressed the "Skills, Training, and Recruitment" theme, ensuring that members could independently employ data science techniques for their research.
Solution: AI revolutionised research by swiftly analysing vast volumes of cellular imaging data. Alongside this, Faculty identified skills shortages, and helped build capabilities through developing a tailored training program empowering researchers with data science skills relevant to their work.
The implementation of AI led to facilitating quicker insights into cellular behaviour and disease mechanisms. Specifically, the AI-enabled models helped stratify patients based on drug responses, identify biomarkers for drug efficacy, and predict disease states in PD.
Interested in reading more? You can do so here: Faculty AI - life sciences.docx
__________________________________________________________________
(Lenus Health Ltd.) Business Strategy – Chronic Condition Management in the NHS
Reducing admissions using Machine Learning based Risk Stratification in respiratory disease.
Problem: COPD is a long-term respiratory condition affecting over 1million people in the UK. In 2022 there were 113,304 COPD related admissions across the NHS in England with a readmission rate of 43% the study aims to identify these patients earlier to reduce the cost and impact on the NHS.
Solution: Lenus Health used machine learning models to generate risk prediction scores for COPD patients in NHS Greater Glasgow and Clyde. The models were used to stratify patients and identify who was at high risk.
The project proved that AI can be used to shift care from a reactive to preventative approach, enhancing patient outcomes and reducing healthcare burdens and costs.
Interested in reading more? You can do so here: Lenus Health - life sciences.docx
___________________________________________________________________________
(Lux Aeterna (NVIDIA)) Organisations have used NVIDIA AI tools to enhance their own capabilities.
Problem: Ability to create more creative parallax occlusion maps (a method for creating the effect of depth for 3D textured surfaces).
Solution: NVIDIA’s AI helped by augmenting Lex Aeterna’s AI architecture. Lex Aeterna’s attendance at a tech festival hosted by NVIDIA and Digital Catapult allowed the SME to receive feedback, iteration and guidance on how to better deploy AI.
Lex Aeterna, by using and learning how to better deploy NVIDIA’s AI tools, improved its services and the quality of its operation.
Interested in reading more? You can do so here: NVIDIA - creative industries.docx
___________________________________________________________________________
(RainBird (AiDan)) Fusing decision intelligence with end-to-end automation.
Problem: DAC Beachcroft's Property Recoveries team found that the initial process of assessing claims for recovery prospects was very manual and often lacked all the information required for an accurate assessment. This led to excessive time and resources invested in ultimately unsuccessful cases.
Solution: Identifying more recoveries and driving better results. Standardise and improve triage efficiency, enhancing recovery performance. Ingest and validate high volumes of unstructured and inconsistent client data.
By automating labour intensive and repetitive aspects of the triage process, DACB enables their triage and legal teams to work smarter, increase efficiency, and provide an even better service to their clients.
Interested in reading more? You can do so here: Rainbird (AIDan) - legal.docx
__________________________________________________________________________
(SkenarioLabs) Predictive data analytics for real estate
SkenarioLabs offer predictive data analytics, specialising in real estate risk assessment and valuation. They operate globally in 13 market areas, facilitating the monitoring of over £42bn of real estate each month.
Problem: Collateral portfolios often comprise hundreds of thousands of assets, all of which fluctuate in value. Repeated expert valuations for all units are not cost-effective, and index-based tracking of value change developments is not particularly reliable.
Solution: Algorithms can be used to cost-effectively maintain an accurate valuation of a bank's collateral assets and provide an estimate of their future value, with particular reference to green credentials.
SkenarioLabs' adoption of AI within their models enables the provision of an enhanced, scaleable and efficient service. This allows banks to maintain a more accurate picture of collateral risks at a lower cost than before and respond to evolving sustainability-led reporting obligations, such as those required by TCFD.
Interested in reading more? You can do so here: SkenarioLabs.docx
_________________________________
If you found this case study document on AI Safety interesting and want to learn more about techUK’s work in AI Safety, please email [email protected].
Tess Buckley
Tess is the Programme Manager for Digital Ethics and AI Safety at techUK.
Usman Ikhlaq
Usman joined techUK in January 2024 as Programme Manager for Artificial Intelligence.
