<img height="1" width="1" style="display:none" src="https://www.facebook.com/tr?id=648880075207035&amp;ev=PageView&amp;noscript=1">

BayesiaLab Webinar Series

shutterstock_708011203_1600x1067

Artificial Intelligence & Ethical Reasoning

Wednesday, November 13, 2019, 11 a.m. – 12 p.m. (CST, UTC-06)

Abstract

With more and more autonomous interactions of computers and robots with the world around us, the questions about their ethical behavior emerge. In the context of self-driving cars, hypothetical scenarios are often proposed as examples, such as the choice of driving off a cliff or plowing into a child. How could a computer decide ethically when facing such a dilemma? Examples like these are cited to imply that Artificial Intelligence would not be able to make ethical or moral judgments the way a human could.

In this webinar, we propose that Artificial Intelligence is able to make more reliable, ethical decisions compared to humans. Why? Humans are not particularly good in performing probabilistic inference in general, and their performance presumably deteriorates further when forced to decide within a split-second.

That is not to say humans should not specify moral values and ethical standards based on their beliefs, that is undoubtedly a human prerogative. What we are saying is the Artificial Intelligence may be better equipped for systematically reasoning based on given moral standards.

Example: damned if you do and damned if you don't

In this webinar, we showcase a tragic dilemma, which does not seem to provide an option for a universally acceptable "good" decision. As a result, the objective is to look for the "lesser evil" among several terrible choices. A further burden to this quandary is the presence of uncertainty in terms of situational awareness and outcomes.

By framing this situation as a Bayesian network and assigning human-defined values as utilities, we show that the "least bad" decision can be clearly identified and justified. Importantly, this approach allows for separating value judgments and structural considerations. Hence, different decision-makers could be in perfect agreement as to how the world works, yet based on different values, they could each arrive at a decision that is right for them.

Webinar Registration for Nov. 13, 11 a.m. (CST, UTC-06)