How can you overcome algorithmic governance challenges?

The context which algorithms operate is essential because there is often a social context in which algorithms operate that the developers of algorithms are either unaware of do not pay sufficient attention to in the development process. This can lead to disastrous consequences because algorithms can produce outputs that are laden with socially unacceptable statements which can also expose developers to legal liability. There have already been a number of examples of poorly trained algorithms leading to discrimination on the basis of protected categories which have also exposed the developers of the algorithm to legal liability. Ultimately algorithms need to be developed in accordance with the values of international human rights law.

Diversity is a powerful tool for eliminating algorithmic bias. Almost all AI works on the garbage in, garbage out principle which means that AI is simply a reflection of the data that it was trained on. If the training data lacks diversity and is only narrowly representative of the purpose it is designed this will be reflected in the outputs of the model. This is why it is essential that training data is garnered from a wide variety of sources and that means being inclusive and diverse when consulting stakeholders in the development phase of the project.

Accountability in the development of AI algorithms is an essential consideration. Implementation of auditing processes is a risk mitigation technique which prevents the exposure of the algorithm and its associated AI project to legal liability. Ensuring that the algorithm has transparency, explainability and verifiability also supports the continuous improvement of the AI system because it allows errors to be identified and rectified. If an AI algorithm is at risk of engaging in discrimination against a category of persons protected under international human rights law or its national equivalents such as civil rights law in the United States, then it is essential to conduct robust auditing processes.

Ultimately technological development needs to be guided by the advancement of the wellbeing of people and the progress of human affairs. The incorporation of ethical principles into the development of algorithmic governance is therefore essential. Particularly in the world of AI and data science, the need to respect the privacy of individuals and their data is paramount but concepts of fairness, justice and human dignity are also essential to the design process of algorithms.