How do you include human values in your algorithms?

1 Why values matter

Human values are the principles, standards, and preferences that guide our actions and judgments. They can be influenced by our culture, religion, education, experience, and personality. Values can also vary across different contexts, situations, and stakeholders. For example, some people may value privacy more than convenience, while others may prioritize efficiency over fairness. When you create or use an algorithm, you need to consider how it aligns with or conflicts with the values of yourself and others who are affected by it. Ignoring or neglecting values can lead to ethical, legal, and social problems, such as discrimination, bias, harm, or mistrust.

2 How to identify values

One of the first steps to include human values in your algorithms is to identify them. This means understanding what values are relevant, important, and shared by the different stakeholders involved in the algorithm's development and deployment. Stakeholders can include yourself, your team, your organization, your clients, your users, your regulators, and your society. To identify values, you can use various methods, such as surveys, interviews, focus groups, workshops, or scenarios. You can also use frameworks or models that categorize or rank different types of values, such as the Schwartz Value Survey, the Rokeach Value Survey, or the Value Sensitive Design approach.

3 How to incorporate values

Once you have identified the values that matter for your algorithm, you need to incorporate them into your design and implementation. This means making decisions that reflect and respect the values of your stakeholders, and avoiding or minimizing trade-offs or conflicts between them. For example, you may need to balance accuracy and fairness, transparency and security, or autonomy and accountability. To incorporate values, you can use various tools and techniques, such as value elicitation, value specification, value operationalization, value verification, or value communication. You can also use standards or guidelines that provide recommendations or best practices for incorporating values, such as the ACM Code of Ethics, the IEEE Ethically Aligned Design, or the EU Ethics Guidelines for Trustworthy AI.

4 How to evaluate values

Another step to include human values in your algorithms is to evaluate them. This means assessing how well your algorithm meets or violates the values of your stakeholders, and how it impacts them positively or negatively. For example, you may need to measure the fairness, accountability, or explainability of your algorithm, or the satisfaction, trust, or well-being of your users. To evaluate values, you can use various methods, such as metrics, tests, audits, reviews, or feedback. You can also use frameworks or models that provide criteria or indicators for evaluating values, such as the FATML Principles, the AI Ethics Canvas, or the Value Sensitive Design Evaluation.

5 How to adapt values

The final step to include human values in your algorithms is to adapt them. This means updating or modifying your algorithm to address any issues or gaps that arise from your evaluation, or to respond to any changes or challenges that occur in your environment. For example, you may need to adjust your algorithm to accommodate new laws, regulations, or norms, or to cope with new data, users, or contexts. To adapt values, you can use various strategies, such as value learning, value negotiation, value alignment, or value governance. You can also use frameworks or models that provide methods or processes for adapting values, such as the AI Ethics Lifecycle, the Responsible Innovation Framework, or the Value Sensitive Design Iteration.

6 How to learn more

Including human values in your algorithms is not a one-time or easy task. It requires continuous and collaborative effort, as well as critical and creative thinking. If you want to learn more about how to include human values in your algorithms, you can explore some of the resources and references that we have mentioned in this article, or look for other sources that suit your needs and interests. You can also join or create communities of practice that share and discuss ideas, experiences, and challenges related to algorithm ethics and social impact.

7 Here’s what else to consider

This is a space to share examples, stories, or insights that don’t fit into any of the previous sections. What else would you like to add?