Preventing Data Catastrophes
In the realm of data science and machine learning, Information Resilience has emerged as a crucial component, ensuring the robustness, adaptability, and sustainability of data operations and model performance in the face of disruptions. This article will delve into the significance of Information Resilience, drawing on a recent case study involving a covid symptom tracking app.
The covid symptom tracking app, designed for short-term respiratory infections, was used in a research paper to draw inaccurate conclusions about the prevalence of Long Covid. However, the app's data collection practices conflicted with the workflows of on-the-ground partners, who were not compensated for their extraneous effort. The project did employ automation to support manual work, but errors in state reporting mechanisms were often caught by eagle-eyed data scientists.
The faulty research paper's results were widely shared, offering false reassurance about Long Covid prevalence. This incident underscores the importance of understanding the context for data, as gathering data for one context (short-term mild respiratory infections) was used for another (long-term neurological, vascular, and immune diseases), resulting in missing and incomplete data. Moreover, the app didn't include common Long Covid symptoms, had a frustrating user-interface, and made erroneous assumptions about recovery.
Ignoring the expertise of long covid patients led to lower quality data and erroneous research conclusions. It is crucial that the involvement of the people impacted in AI systems is meaningful, ongoing, and compensated. Dr. Timnit Gebru proposed Datasheets for Datasets as a means to make the context of data more explicit.
Inaccurate research and incomplete data from the covid tracking app could have been avoided by drawing on the expertise of patients. The Information Resilience centre, focused on detecting and responding to failures and risks in data, has recently launched, funded by the Australian government's top scientific funding body (ARC).
Information Resilience encompasses sourcing, sharing, transforming, analyzing, and consuming data. It involves the capacity of data systems and governance frameworks to continue operating effectively despite internal changes or external shocks. In machine learning, it includes continuous updating and fine-tuning of models to improve prediction accuracy and adapt to new data or threats.
Maintaining data availability and integrity is crucial since data is the 'oxygen' powering ML and data science workflows. Information Resilience supports quick recovery from attacks or failures, minimizing downtime and operational risk. It enables organizations to adapt data governance and machine learning processes to evolving internal and external environments, ensuring long-term sustainability and compliance.
Enhancing security and resource optimization is also vital for scalable and robust ML systems. Improving resilience can progress through maturity levels from reactive and fragmented approaches to autonomous, AI-optimized systems that proactively manage risks and recover efficiently.
In summary, Information Resilience is fundamental in data science and machine learning for ensuring robust, adaptive, and sustainable data operations and model performance in the face of disruptions, thus protecting organizational value and enabling reliable insights and decisions over time. Collecting data directly from patients, understanding the context, and involving the people impacted in AI systems are key to achieving high-quality, accurate, and meaningful data.
- The significance of Information Resilience in data science and machine learning is highlighted in a recent case study involving a covid symptom tracking app.
- The faulty research paper's results, based on the covid tracking app's data, offered false reassurance about Long Covid prevalence.
- Ignoring the expertise of long covid patients led to lower quality data and erroneous research conclusions.
- Dr. Timnit Gebru proposed Datasheets for Datasets as a means to make the context of data more explicit.
- The Information Resilience center, funded by the Australian government's top scientific funding body (ARC), focuses on detecting and responding to failures and risks in data.
- Information Resilience involves continuous updating and fine-tuning of models in machine learning to improve prediction accuracy and adapt to new data or threats.
- Maintaining data availability and integrity is crucial since data is the 'oxygen' powering ML and data science workflows.
- Enhancing security and resource optimization is crucial for scalable and robust ML systems, progressing through maturity levels towards AI-optimized systems that proactively manage risks and recover efficiently.
