Note: this is Part 3 of the series summarizing the recent paper by Lee, Floridi, and Singh (2020)

Parts 1 and 2 of this series have provided a critique of existing mathematical fairness metrics that do not consider the context discussed in ethical philosophy and in welfare economics. Lee, Floridi, and Singh (2020) propose a new approach called “Key Ethics Indicators” (KEIs).

Instead of out-of-the-box fairness definitions that fail to capture the nuanced ethical trade-offs, the authors argue that it is important to devise customised success metrics specific to the context of each model, considering welfare, autonomy, equality, and long-term impact, with close collaboration between the relevant areas of the business and the technical team. The following process helps to achieve this:

  • Define “success” from an ethical perspective. What is the benefit of a more accurate algorithm to the consumer, to society, and to the system itself? What are the potential harms of false positives and false negatives? Are there any fundamental rights at stake?
  • Identify the layers of inequality and bias that are affecting the differences in outcome in order to inform an appropriate mitigation strategy. This may require changes to data collection mechanism or to existing organisational processes, rather than a technical solution.
  • Operationalise these objectives into quantifiable metrics, build multiple models and calculate the trade-offs between the objectives covering all ethical and practical dimensions
  • Select the model that best reflects the decision-maker’s values and relative prioritisation of objectives. This enables a defensible and concrete justification of why a model is more successful in satisfying its pre-defined ethical purpose than the alternative options. The outcome of the evaluation may also be that a machine learning model should not be used at all after weighing the risks and ethical implications against the potential benefits; in such cases, a rules-based model, human judgment, or a combination may be the decision-making process.

Similarly to how a company may define a set of quantifiable values to gauge its achievements using Key Performance Indicators (KPIs), there should be outcomebased, quantifiable statements from an ethical standpoint: Key Ethics Indicators (KEI), enabling developers to manage and track to what extent each model is meeting the stated objectives.

For example, Lee and Floridi estimate the impact of each default risk prediction algorithm on financial inclusion and on loan access for black borrowers. They operationalise financial inclusion as the total expected value of loans under each model and minority loan access as the loan denial rate of black applicants under each model. As Figure 1 from their recent work illustrates, calculating the trade-offs between the two objectives for five algorithms provides actionable insights for all stakeholders on the relative success of each model. Context-specific KEIs should be developed for each use case. For example, in algorithmic hiring, employee satisfaction with a role may be estimated by attrition rates and employee tenure, employee performance may be measured through their annual review process, and diversity may be calculated across gender, university, region, age group, and race, depending on each organisation’s objectives and values.

The roles and responsibilities of a developer are necessarily intertwined with the role of the expert or business stakeholder, as the ethical and practical valuations of what “success” looks like in the model directly influences the algorithm design, build, and testing. It is therefore important to have active and on-going engagement from the beginning between the developer and the subject matter expert to try to understand which inequalities should influence the outcome and how to address the inequalities that should not play a role in the prediction. This process requires engagement from all relevant parties, including the business owner and the technical owner, with potential input from regulators, customers, and legal experts.

One of the key benefits of the outcome-driven KEI trade-off analysis is that it provides interpretable and actionable insights into the decision-maker’s values, which is especially important for complex machine learning algorithms in which the exact mechanism may not be transparent or interpretable. This could also facilitate the discussions with the regulator, the board, or other stakeholders on why a certain model was seen as preferable to all other reasonable alternatives. This may also help reduce the hesitation among decision-makers around the use of machine learning models due to their non-transparent risks, if the analysis shows they are superior to traditional rules-based models in meeting each of the KEIs.