# On the meaning of uncertainty for ethical AI: philosophy and practice

**Authors:** Cassandra Bird, Daniel Williamson, Sabina Leonelli  
**arXiv:** [2309.05529](https://arxiv.org/abs/2309.05529)  
**Date:** 2023-09-11  
**Summary page:** https://www.summarizepaper.com/en/paper/2309.05529v1/

## Key points

- Accountability for data scientists, statisticians, and modellers in relation to AI systems
- Proposal to enhance transparency and accountability in decision-making using AI systems
- Benefits of increasing transparency and accountability in AI systems:
- Enhances responsiveness to feedback for continuous improvement
- Improves the quality and meaning of uncertainty associated with outputs
- Increases transparency for evaluation and identification of biases or shortcomings
- Use of Posterior Belief Assessment (PBA) as a method to enhance belief ownership from complex AI structures
- Incorporating ethical considerations into mathematical reasoning for implementing ethical AI in statistical practice
- Case study involving statisticians developing models during the Covid-19 pandemic
- Focus on combining different models using PBA to report uncertainties ethically and principled manner
- Importance of reporting uncertainty through an Ethical AI lens

## Summary

This paper addresses the issue of accountability for data scientists, statisticians, and modellers in relation to the AI systems they develop. The complexity of AI systems and the challenges in comparing and synthesizing competing claims arising from their deployment for data analysis have made this topic controversial and highly debated. To address this issue, the authors propose enhancing the transparency and accountability of decision-making using AI systems by explicitly acknowledging the statistical foundations that underpin their development. This includes considering how these foundations dictate how the results of AI systems should be interpreted and acted upon by users. The authors argue that increasing transparency and accountability in AI systems can lead to several benefits. Firstly, it enhances the responsiveness of models to feedback, allowing for continuous improvement based on user input. Secondly, it improves the quality and meaning of uncertainty associated with the outputs of AI systems, providing users with a better understanding of the reliability and limitations of these outputs. Lastly, it increases the transparency of AI systems to evaluation, making it easier to assess their performance and identify potential biases or shortcomings. To exemplify their approach, the authors extend Posterior Belief Assessment (PBA) as a method to enhance belief ownership from complex and competing AI structures. They argue that incorporating ethical considerations into mathematical reasoning is crucial for implementing ethical AI in statistical practice. The paper presents a case study involving a group of statisticians from the University of Exeter who were tasked with developing models to track variants of concern during the Covid-19 pandemic in England. These models were used to provide insights to SPI-M-O (Scientific Pandemic Influenza group on Modelling, Operational sub-group), which then reported key findings to SAGE (Scientific Advisory Group for Emergencies) for informing government decisions. In this case study, the authors focus on one particular week near the beginning of the Omicron wave in December 2021. They explore how different models developed by the statisticians could have been combined using PBA to report uncertainties to SPI-M-O rather than simply passing on a collection of model results for synthesis. The aim is to demonstrate importance reporting uncertainty through an Ethical AI lens while discussing its process at length. It is important to note that this study does not present specific results relevant to pandemic or critique work done by statisticians or other groups involved; instead its purpose is showcase techniques developed synthesizing models & reporting uncertainty ethically & principled manner .Overall ,this paper highlights significance transparency ,accountability & ethical considerations in development & statistical practice .By explicitly acknowledging statistical foundations & implementing methods like PBA ,decision making using AIs can be made more transparent ,responsive & reliable .