Léa, your research employs AI techniques to predict the performance of company directors. What do you see as the current shortcomings in board appointments, and what alternative is AI offering?
“Boards have nominating committees that are responsible for selecting new board members and, theoretically, the CEO should not be involved. In reality, however, it is not uncommon for the CEO to have significant influence, which is a problem, of course, because it is the board’s role to oversee the CEO’s performance.
“Some boards hire external search firms and there are a lot of benefits to that because a third-party can help identify potential candidates that are not part of the board’s existing network. Having said that, it’s clear from my talks with recruitment companies that proposed candidates are often discarded and the final appointment goes to someone that is already known to the board in any case.”
“We designed an algorithm that mimics what boards should be doing to select candidates: predicting their performance and selecting from the top. We used demographic information, experience and education, while also taking into account the biographies of incumbent directors. The algorithm worked remarkably well in ranking directors and predicting the distribution of outcomes.
“We then looked at incoming directors selected by boards and compared them with the directors that the algorithm had ranked highly. What was striking was the confirmation of something that has been understood since the days of Adam Smith: there is a lot of homogeneity on boards that effectively self-select. The machine learning algorithm, by contrast, was able to identify directors who not only performed better but were also more diverse.”
Léa Stern
Are there risks in delegating appointments to an algorithm?
“It is highly unlikely that companies will hire directors at the push of a button any time soon, so I am not worried on that front. However, companies should learn from previous failed attempts, should they rely on algorithmic decision aids.
“A few years ago, Amazon tried to automate some of its hiring decisions, but there were some fundamental problems with the algorithm. The company tried to predict the likelihood of promotion among potential candidates. The results found that men were far more likely to be hired than women. However, the algorithm used to determine this was based on past hiring and promotion decisions, thereby amplifying existing biases. Algorithm design is critical. Third-party algorithm audits will likely be crucial elements in the future of AI to avoid bad practices and unintended consequences on discrimination, for example.”
Léa Stern