Data Ethics Service

Maximise the ethical use of data across your organisation

Bias and discrimination in AI can occur in several dimensions of data and cloud services, such as the setting of business objectives, training data, bias in algorithms, and bias in the interpretation of the results. To mitigate bias at each phase, and to promote fairness and equality, it is important that the relevant data collection, data use, data interpretation and data ethics issues are considered at the outset and throughout the data analytics process.

Our 7-step Ethics Assessment reviews your systems and processes to identify relevant ethical risks and potential mitigation measures to ensure your systems maximise the ethical use of data from end-to-end. 

The 7-steps are:

Step 1. Ethical assessment of business objectives

To integrate ethics into the business understanding phase, we include the requirements for ethical and trustworthy AI in the list of requirements and test the business objectives against the ethics requirements. This includes an assessment of whether any special issues (e.g. vulnerable populations, sensitive data such as medical data or biometrics, etc.) are likely to be involved, and if so, guidelines for these special issues will be included.

This step is needed to
identify any possible tensions between the business objectives and ethics requirements.

Step 2. Ethical assessment of data objectives

In step 2, we test the data objectives against the ethics requirements

We do this because even if the business objectives are compatible with the ethics requirements, the data objectives may be formulated in a way that is not compatible (e.g., it may propose a segmentation of people into social categories that was not referred to in the business objectives and that does not fit well with principles of fairness and equality).

Step 3. Stakeholder analysis (a) or involvement (b) in the business understanding phase

Inclusion of ethical criteria in the development process benefits from a stakeholder analysis, in which direct and indirect stakeholders to the project are identified and their values and interests are assessed. 

This makes it easier to identify more specific ethical requirements, make ethical assessments, and assess possible tensions between objectives and requirements and ethical criteria.

Step 4. Ethical data collection and assessment

To integrate ethical requirements into this phase of the process, we start by evaluating the data collection choice. 

At this stage, bias, discrimination, fairness and diversity, privacy, and data quality are particularly important. Data selection and fair algorithmic design are coupled with an ongoing ethical need to understand the historical and social contexts into which these systems are being deployed. 

While definitions and statistical measures of fairness are certainly helpful, they cannot consider the nuances of the social contexts into which an AI system is deployed, nor the potential issues surrounding how the data were collected. As a result, the assessment process examines the role of human judgment in the decision-making loop to continuously assess whether the AI system has sufficiently minimised unfair bias so that it can be safely released for use. 

The assessment draws on many disciplines including data science, law, and ethics.

Step 5. Ethical data description, exploration, and verification

To integrate ethical requirements into the rest of the process, we evaluate the ethical consequences of describing, exploring, and verifying the data

At this stage, issues relating to privacy, data quality, precision, accuracy, transparency, explainability, bias, discrimination, and fairness and diversity will be particularly important. 

We examine the “gross” or “surface” properties of the acquired data (such as format and quantity), and evaluate whether the data satisfies the relevant requirements

We pay special attention to data mining questions that concern patterns in the data (e.g., distribution of key attributes, relationships between pairs of attributes, properties of significant sub-populations, simple statistical analyses), through queries, visualization, and reporting techniques. 

Finally, we examine the quality of the data, including completeness, correctness, and missing variables (e.g. were more affluent neghborhoods more easily accessible for data gathering?)

Step 6. Ethical assessment of modelling

Next, we assess whether ethical criteria are considered in the modelling stage, and that the selection of the model(s) are evaluated relative to these ethical criteria. 

Issues that may be particularly relevant are those relating to transparency, and safety and robustness. This step examines the chosen modelling technique (e.g., neural network generation with backpropagation, or decision-tree building with Python) and assesses it for accuracy and lack of bias.

Step 7. Ethical assessment of outcomes

Finally, an ethical assessment will be performed on the results. 

Possible outcomes are that ethical issues have been dealt with in a satisfactory way, that further development is needed, or that specific guidance for or restrictions on deployment and use need to be in place to mitigate ethical issues.

Furthermore, we can support your organisation in planning, implementing, monitoring, and overseeing ethical improvements using a risk-based and rights-based approach to prioritise activities.

Please contact us to find out more about how we can help you. 

G cloud

Our solutions are available to purchase through G-Cloud.

Talk to us

If you’d like to find out more about our Data Ethics Service, fill out the contact form below  and we’ll be in touch.