Academics call on New Zealand to regulate AI as Brookings issues guidance
A pair of academics from the University of Otago are calling for the New Zealand government to establish a new independent regulator to monitor artificial intelligence technologies and address risks associated with their use by government agencies. In a post to The Conversation, Postdoctoral Fellow John Zerilli and Associate Professor in law and ethics Colin Gavaghan say the country is a leader in government AI, but concerns about transparency, meaningful human control, data protection and bias must be addressed.
Their 92-page report (PDF) considers the current state of New Zealand law and government AI practice. The researchers recommend the creation of a public register for predictive algorithms used by the government, perhaps along with explanation systems, as well as consideration of the legal status of inferred data. The risk of bias and discrimination based on “dirty data” is also noted, along with the possibility of mitigating this risk by including results for specific sub-populations in public evaluation of AI tools.
Bias is also dealt with in a report on “Algorithmic bias detection and mitigation: Best practices and policies to reduce consumer harms” from the non-profit Brookings Institution. The Brookings report argues that operators and other stakeholders need to proactively address factors which contribute to bias.
To aid that effort, a framework for “algorithmic hygiene” is presented, identifying some specific causes of bias, along with best practices to deal with them. The goal of the report, the authors write, is to juxtapose the concerns of computer programmers and industry with those of policy makers and civil society. More than 40 thought leaders from across industry, academia, and civil society contributed to the report and framework.
Recommendations include updating laws to apply to digital practices, implementation of a set of self-regulatory best practices, and a system to improve algorithmic literacy and promote feedback between industry and civil society.
The report presents examples of algorithmic bias, describes bias detection strategies, and discusses the trade-offs between fairness and accuracy. It emphasizes the importance of ethical frameworks, and offers proposals for bias mitigation. A template for a bias impact statement, consisting of 16 questions, is also provided.
In concluding, the authors warn that the use of algorithms in the U.S. criminal justice system are likely to perpetuate bias and result in longer prison sentences for African-Americans and poor people.
“As outlined in the paper, these types of algorithms should be concerning if there is not a process in place that incorporates technical diligence, fairness, and equity from design to execution,” the authors write. “That is, when algorithms are responsibly designed, they may avoid the unfortunate consequences of amplified systemic discrimination and unethical applications.”
The problem of potential AI bias has received increasing attention over the past couple of years, and a Forbes editorial recently suggested that some companies may need to consider creating a “Chief Bias Officer” position to combat it.