Implementing an ethical AI framework

People

Principal investigator

Research areas

Description

The purpose of this project is to implement an ethical framework and some algorithms for answering questions in this framework.  

With the increasing importance of automated decision making, there is a will to enforce that only ethical, morally permissible, acts are taken by machines.  This is one of the topics we tackle in the HMI (Humanising Machine Intelligence) ANU Grand Challenge project (<a href="/%3Ca%20href%3D"https://hmi.anu.edu.au/">https://hmi.anu.edu.au/</a>">https://hmi.anu.edu.au/">https://hmi.anu.edu.au/</a>).  This is particularly hard because even philosophers struggle to define what makes a decision ethical or not.

The implementation will be based on a recent research paper from Dietrich and List [D&L, 2017] that involves several layers: an ``option'' (the possible result of a decision) in a given ``context'' (the current situation) features a number of ``properties''; then, a relation exists that compares sets of properties, and indicates which ones are preferable.  

Using this framework, it is possible to answer some questions, such as: is a specific option morally permissible?  what other acceptable options are there?  is there actually a morally acceptable option (moral dilemmas)? etc.  One can use this framework directly in order to answer the questions.  It is also possible to find better procedures.  

Given some additional assumptions on the moral theory, for instance the fact that the relation is transitive (if A is better than B, and B better than C, then A is better than C --- this relation is not always satisfied), there exist algorithms more efficient than the standard, naive, ones.  The framework you will have to implement should be able to automatically build on the existing results, so as to automatically identify the best algorithm.

This project will expose the student to ethical AI, a topic that is very popular among modern companies.

[D&L, 2017] Dietrich, Franz, and Christian List. "<b>What matters and how it matters: a choice-theoretic representation of moral theories</b>." <i>Philosophical Review</i> 126.4 (2017): 421-479.

 

Requirements

Python3

Background Literature

[D&L, 2017] Dietrich, Franz, and Christian List. "<b>What matters and how it matters: a choice-theoretic representation of moral theories</b>." <i>Philosophical Review</i> 126.4 (2017): 421-479.

 

Keywords

AI; ethics

Updated:  1 June 2019/Responsible Officer:  Dean, CECS/Page Contact:  CECS Marketing