Weekly Feature: “If Your Company Uses AI, It Needs an Internal Review Board”

February 7, 2021

Reid Blackman recently published a piece in the Harvard Business Review where he argues that companies using AI should develop Internal Review Boards modeled off those within the medical field.

In Blackman’s view, companies using AI generally know they need to worry about ethics; but when it comes to actually implement strategies, they fall short. Internal Review Boards can help businesses save not only money and brand reputation, but, more significantly, they can promote better, more responsible applications of AI.

Why companies fail with AI ethics

According to Blackman, discussions of AI ethics follow a similar flawed pattern between many organizations. It begins with a narrow definition of AI ethics as an issue rooted in “fairness,” rather than a more holistic approach mindful of the complex, interrelated concerns related to using AI.

Additionally, companies tend to search for technical tools and quantitatively-based bias-mitigation strategies after they identify potential AI ethics issues. While this isn’t inherently problematic, Blackman argues that “the truth is that many ethical issues are not reducible to quantitative metrics of KPIs.” Even further, Blackman suggests that technical tools do not cover all types of bias; there are some cases of bias where no technical tool exists.

A better solution: IRBs

Internal Review Boards (IRBs) were introduced in the medical field to mitigate the ethical risks of conducting research on human subjects. IRBs carry out their function by approving, denying, and suggesting changes to proposed research projects—proactively, not reactively, promoting the idea of “do no harm.”

Blackman argues that there are similar ethical risks in medicine and in AI. In both, there is potential for harming individuals and groups, imposing physical and mental distress, invading privacy, and undermining autonomy.

Moreover, IRBs can promote AI ethics by systematically and exhaustively identifying ethical risks before AI tools are even created. In addition to approving and rejecting proposals, perhaps most importantly, IRBs can help researchers and product developers by making ethical risk-mitigation recommendations.

Read the full piece here.