We partnered with Women in Data® and UCL to uncover bias and transform outcomes for women's health with GenAI.
KubrickGenAI for Women's Health Research
The challenge
- A leading professor in autoimmune disease research at UCL wanted to test her hypothesis that most studies on Systemic Lupus Erythematosus (SLE) which disproportionately affect women don’t adequately address the risk of sex bias in their methodologies when using Machine Learning.
- In partnership with Women in Data®, Kubrick provided the voluntary expertise of two of their Generative AI specialists to help expedite and improve the systematic review processes to uncover sources of bias.
The solution
- Kubrick’s consultants worked in agile to create a POC which they scaled to an MVP in order to test and develop a GenAI tool using OpenAI GPT-3.5 Turbo LLM.
- They worked collaboratively with their stakeholder to align their technology to domain protocols so as to augment highly manual processes, including sourcing and vetting papers for eligibility, scoring papers on their proficiency in addressing and remediating potential sex bias, and scoring papers in their proficiency to follow Machine Learning best practice to remediate potential bias.
- The GenAI model is designed to address multi-class classification challenges, with in-built self-reflection and stringency assessment to act as a assistant to human assessment.
The results
- The MVP sourced and scored 160 papers with a run-time of 2 minutes per paper for assessment, demonstrating the ability reduce the workload of researchers from weeks to days. In doing so, researchers will be able to more easily identity areas of novel, niche, and underfunded research, overcoming some of the barriers in women's health research.
- Kubrick also provided a comprehensive report on the tool and their development process to ensure explainability and transparency of GenAI to enhance trust and encourage human oversight.
- The tool demonstrated a startling gap in acknowledgement and remediation of sex bias in research. Kubrick are now working with Women in Data® to scale access to the tool for other women’s health initiatives and pursue support to improve remediation of sex bias in women's health research.
