The talk started with a discussion on the meaning of ethics. According to the speaker, “Ethics is a way of life and a moral code”. The speaker traced the roots of ethics to the fourth century beginning from the works of Socrates, who discussed concepts of good, right, just, and virtue. Plato, a student of Socrates, explored the problems of Socrates from an idealistic point of view. Aristotle, a student of Plato, then explored ethics from an applied angle having applications in fields like biology and laws of nature. Modern ethics traces its roots to the works of Hume, who attributed the concepts of good, right, just, and virtuous to individuals and actions using various types of evaluation like the utilitarian.

Then, the speaker re-centers the focus towards ethical ramifications of AI considering that AI is having an increasing impact on every aspect of human life. In the speaker’s own words, “AI will bring about significant change in 50 to 100 years”. Even though everybody agrees with the ethical principles of AI at a broad level, the subjective nature of these principles makes ethics integration of AI a challenging task. For instance, what is fair in India might not be fair in the US. One way to deal with such conflicting situations is to define principles and align with them. Another way to deal with the problem is to find a common ground across all conflicting scenarios and regulate it. While the former is an individualistic approach, the latter approach is a utilitarian one.

Whichever approach we take, the task of integrating AI Ethics principles into businesses is difficult. The speaker cites [1] to discuss the reasons for difficult integration due to various issues ranging from privacy to power imbalance. Additionally, the rate of technological progress in AI has been greater than the rate of regulation. The report contains interviews of chief ethics officers, privacy officers, technical officers from leading think tanks and Fortune 500 companies who stated that they are solving the ethical integration problem through the following ways: common-sense reasoning or in the speaker’s own words “Grandma approval” reasoning; building teams with a hub-spoke model where each team has an AI ethics officer and the company has a chief AI ethics officer; checklists for AI models; privacy preservation; AI algorithmic fairness and explainability.

There has been a lot of push from the research community to push from the black box to grey box solutions that are both accurate and explainable. The speaker however does not advocate for a legal framework for explainability because explainability and accuracy have tradeoffs at present. Moreover, explainability is a derived need that differs based on the target audience which makes it a difficult problem as multiple target audiences need to be considered while devising explainable systems. For instance, a layman who has been rejected a loan should be told what he or she should do to get a loan, while a data scientist needs to be directed towards the internal workings of the model.

Even after designing explainable systems, disputes are unavoidable. The speaker along with his colleagues at the Ohio State University have deliberated over the unavoidable disputes in AI-human settings and call for legal intervention. The speaker discusses a conflicting situation of a Fitbit tracker that has a utilitarian advantage of saving people’s lives, yet causes harm when used to track individual activities. This calls for a legal solution that analyzes the underlying situation like a COVID-19 crisis and then decides to allow Fitbit monitoring for utilitarian benefits. The talk ended with a discussion on the challenges in AI ethics in India, with a reference to [2] to foster more AI ethics discussions in the Indian context, and how we need to focus on talent development in this area by offering courses on the ethical use of data.

References:

1. Dennis Hirsch, Timothy Bartley, Aravind Chandrasekaran, Davon Norris, Srinivasan Parthasarathy, Piers Norris Turner, Business Data Ethics: Emerging Trends in the Governance of Advanced Analytics and AI, 2020.

2. Committee of Experts under the Chairmanship of Justice B.N. Srikrishna, A Free and Fair Digital Economy Protecting Privacy, Empowering Indians, 2018