Operationalising Data Ethics

Companies that leverage their data are attractive to clients, employees and shareholders – but they are being scrutinised accordingly. As advances are made within the data space, the risk grows that an organisation’s use of data may lead to unexpected and unfamiliar ethical issues.

We began to explore this topic recently on the blog, setting out how decision-makers must navigate operational and reputational risks when leveraging their data. Data ethics make up a critical part of this picture. In the absence of robust regulation, organisations must proactively engage with data ethics. Here’s how they can do it by harnessing a data ethics framework. 

The importance of a flexible framework 

A data ethics framework is required by organisations to monitor their use of data and the ethical issues surrounding its usage, introducing governance to bring tangibility to the topic. Operationalising data ethics should be viewed as a journey that will develop over time. This is because It isn’t static. Data ethics are dynamic. 

Hence, a framework must reflect developments as new thinking, regulations and technologies emerge and respond to new inputs. This means that a framework must be flexibly designed so innovation isn’t stymied. 

By producing a flexible framework, data ethics can be incorporated into data handling and manipulation processes to become ‘business as usual’ can become ‘business as usual’, and that should be the aim for all organisations who set out to leverage their data. 

A principles-led approach

Principles are a key part of developing a framework. This is especially true when developing, deploying and using AI. These should be clear and actionable, enabling you to start defining “what good looks like” when it comes to data ethics.

Defining what data ethics means to your organisation ensures everyone is on the same page. They should be defined in advance as the basis of the framework, however, they must also reflect the framework’s flexibility. 

In order to define its principles, an organisation should: 

• Engage a range of stakeholders
  

In order to reduce bias in models and decision making, stakeholders from a range of departments and backgrounds must be included. 

• Perform a data ethics survey

A data ethics survey can help you to understand what matters to your organisation by generating ideas and encouraging debate. 

• Run data ethics workshops

Workshops should define your data ethics code and principles, bringing stakeholders together to actively engage in discussion. 

Tech as an enabler 

In order to uplift the framework, technology should be deployed. First, it’s crucial that ethical metrics, such as fairness, explainability and accountability, are established as these will make the framework tangible. Technology can assist by measuring and improving these ethical metrics and they should be adopted where possible, as they will bring transparency.

An organisation should also strive to achieve fairness, as fairness is a key pillar that supports ethical AI. Unfairness comes about mainly due to human bias existing in the training data, and it is important to mitigate the degree of bias a model may have. Engaging a wide range of stakeholders from a range of departments and backgrounds can help here, too. 

Observational criteria can help discover discrimination, but there are tools and methods (for example, preprocessing, optimization at training time, and post-processing) to better quantify fairness across the stages of a model’s lifecycle. A trade-off between accuracy and fairness usually exists, hence model objectives should be kept in mind when making any adjustments.

It’s also important that organisations work towards building Explainable AI. This means that when measuring the metric of explainability, explainable AI can be adopted to simplify model development as well as explain a model’s behaviours to key stakeholders. This can be used to verify that a model is behaving as expected, weed out bias and make improvements.

Embedding the framework 

Once it has been developed and a range of tech enablers identified, a framework must be properly embedded within an organisation. If it isn’t, the framework will soon grow redundant. 

Your organisation can embed a framework by following these steps: 

• Policies, procedures and guidelines
  

Documentation and training materials should be provided to ensure the framework generates value. Guidelines should be clear, so it is easy for an organisation to follow.

• Roles and responsibilities
  

Roles and responsibilities should be defined, and a function assigned as the data ethics facilitator. Ideally, a ‘Data Ethics Officer’ should act as a key point of contact.

• Governance forums
  

Forums and Working Groups should be established and the appropriate cadence for review diarised. Their effectiveness should be monitored by the data ethics facilitator. 

• Training and comms
  

The data ethics framework should be clearly communicated to relevant stakeholders, and training requirements for all members of an organisation updated accordingly.

• Culture

Ownership of the data ethics framework must be implemented at all levels. A culture that embraces data ethics will reduce risk and maximise the value it can derive from its data. 

• External guidance and internal audit

Regulatory advice should be sought from external bodies to ensure the data ethics framework is satisfactory. Further, external bodies can support with reviews and audits.  

If these steps are implemented, there’s no reason why a data ethics framework can’t be deeply embedded within an organisation’s day-to-day ‘business as usual’. 

A vital part of the embedding process is to be aware of how important risk is when it comes to data ethics. Needless to say, it must be constantly reviewed as the needs of the organisation, regulations and current business climate evolve. Risk is dynamic and a flexible framework will allow your organisation to adapt to the risk requirements over time, too. 

Bringing it all together: Ethics as a business value driver

We all know that there can be a wide chasm between business theory and practise. 

Hence, it is easier to talk about subjects like data ethics in presentations and in meetings than it is to put them into operation. It is perhaps useful then to not think in terms of ‘data ethics’ but in terms of ‘responsible AI’ posing the question – ‘how can we use AI responsibly to advance our business?’

By following the steps we set out above, your organisation can integrate an effective data framework, operationalising AI responsibly and leveraging the promise that your data has to offer.