AstraZeneca data and AI ethics

HR Workforce Analytics - Case Study

“My team piloted an analytics tool aimed at understanding how people collaborate when using various business technologies. This tool provided valuable insight on how working patterns have evolved through the COVID-19 pandemic. We worked with our Data Privacy Office and the Data and AI Ethics principles to drive responsible use of the tool. This helped us maintain focus on making AstraZeneca a Great Place to Work.

With the framework, our values, and our pre-existing commitments in mind, we identified the following essential criteria:

• Respecting our fellow employees’ privacy and act within reasonable expectations of how we would use their data. Privacy is a fundamental right, and respecting has always been a top priority. Fortunately the tool only uses aggregated data rather than individual-level data, so privacy is maintained.
• Using the tool in a way that benefits our employees. The insights generated have fed into positive discussions, ideas and potential actions to help maintain our position as a top employer while technology continues to transform the workplace.
• Assigning accountability for the analytics system and each use case across its lifecycle.
• Maintaining the ongoing security of the data used by the analytics system.

The principles also helped us develop a firm ethical foundation to specifically enable future HR analytics projects.

We will need to draw insights from an increasingly varied set of technologies and data to stay competitive and to provide the best support to our colleagues. This is especially true as predictive analytics become more commonplace. In this dynamic technology landscape, we are optimistic about our AI solutions because of our combined approach: robust privacy and security framework, peoples’ good judgement, and robust Data and AI Ethics Principles.”

Pedro Nunes, Head of Global Business Services Analytics, AstraZeneca

Scientific Collaboration with BenevolentAI - Case Study

“Our ongoing collaboration with BenevolentAI is AstraZeneca’s largest AI collaboration to date. It presents an exciting opportunity for us to develop novel targeted drugs for two complex diseases with large unmet medical needs- Idiopathic Pulmonary Fibrosis (IPF) and Chronic Kidney Disease (CKD).

For a successful AI partnership, we need to implement our Data and AI Ethics Principles together. Here’s how we’re doing it:

• Private and Secure: BenevolentAI underwent a rigorous IT security review to appropriately safeguard the data involved. This resulted in a secure data sharing solution for the research team that shielded access from each company’s internal data & analytics environments. Some of the datasets included genomic data, which poses a unique challenge for de-identification. The team created a novel model to abstract the data, ensuring the privacy of patients’ data while maintaining its scientific utility for researchers.
• Explainable & Transparent: The machine learning models are normally not directly interpretable. However, we jointly interrogate the models through methods such as latent class clustering. This helps explain which variables were used by the AI system to generate their biological predictions. Also, specific techniques are used to infer causal or likely causal relationships within the data. This enables a more transparent discussion on disease mechanisms than can be achieved through association only. Finally, expert review is an integral part of prioritising predicted targets for further experimental work, which helps us stay aligned with the disease area strategy. Overall, we ensure assumptions are clear, algorithms are documented, and results have clear provenance throughout. 
• Fair: The data used for the collaboration stems from a variety of sources ranging from literature to omics datasets, both internal and external. Each dataset was meticulously annotated with relevant metadata to maintain consistent use of data originating from many different sources. Bespoke pre-processing pipelines were also established to normalise quantitative observations and adjust for confounding factors. 
• Accountable: A very clear data access process was designed, with training required for access to the data and the knowledge graph. The process is monitored and governed to keep usage in line with the collaboration aims and compliant with consent and privacy policies. 
• Human-Centric and Socially Beneficial: The collaboration provides an outstanding opportunity to identify novel targets which will open the doors for treatments for diseases that cause suffering to hundreds of millions of people each year. By holistically leveraging available data and information, better decisions will be made for the benefits of patients. 

If we satisfy these principles, we can continue to push the boundaries of science while living up to our core values.”

Claus Bendtsen (Executive Director, Data Sciences & Quantitative Biology, AstraZeneca)