Richard Vidgen presents his approach to addressing bias and unfairness in algorithm development.
Algorithms are playing a greater and greater role in our daily lives, impacting on job applications, medical treatment, policing, criminal justice, and loans and financial services. There are undoubted benefits to algorithmic decision-making in general, and artificial intelligence (AI) in particular, such as in medical diagnoses. However, there is also a dark side to algorithms. The Guardian newspaper has investigated “How algorithms punish the poor” and reports...
One in three councils are using computer algorithms to help make decisions about benefit claims and other welfare issues, despite evidence emerging that some of the systems are unreliable.
The Markkula Institute for Applied Ethics identifies five ethical dimensions:
- Utilitarian: focuses on outcomes and attempts to maximise good done while reducing harm
- Rights: humans should have the ability to choose freely what they do with their lives and have rights such as being told the truth, to not be injured, and a right to privacy
- Fairness (or Justice): all humans should be treated equally and where people are treated unequally then this must be done on the base of defendable criteria (e.g., to compensate for biases in society)
- Common Good: society is more than the sum of individuals and there are common conditions (e.g., policing, health care, education) that are needed to protect the welfare of all members of society
- Virtue: our actions should be consistent with ideal virtues that promote the full development of our humanity, such as truth, beauty, honesty, courage, compassion, generosity, tolerance, love, and integrity.
Figure 1: the business ethics canvas template
The canvas elements, which are addressed going in clockwise order around the canvas, are:
- A proposed analytics solution that address the needs of specific customers (we recommend that these are generated using an opportunity canvas)
- Identification of stakeholders that can affect or are affected by the proposed analytics solution
- An assessment of stakeholder utility of the analytics solution
- An assessment of the rights of stakeholders
- An assessment of the fairness (justice) of the solution
- Implications for the common good
- Reflections on the virtue of the proposed analytics solution