Algorithm charts moral culture in 100k dilemmas

Should we sacrifice the life of one person to save the lives of five others? 

For social scientists, hypothetical dilemmas with stark, life-and-death consequences play an important role in theorising about philosophical first principles, ethical norms, and moral psychology. 

But a recent study analysing 100,000 threads on Reddit found that non-philosophers grapple with ethics and morality in a variety of messy, comparatively minor quandaries — such as how to deal with misbehaving in-laws at a wedding, or how to resolve disagreements about proper use of the workplace fridge.

“In the past decade, artificial intelligence and especially machine learning algorithms have become a lot more powerful and are being more widely used,” said Prof Lexing Xie who heads up the Humanising Machine Intelligence project. “People are interested in endowing algorithms with moral reasoning abilities in order for them to interact better with humans.”

ANU School of Computing researcher Josh Nguyen recently presented Mapping Topics in 100,000 Real-life Moral Dilemmas at the International Conference on Web and Social Media (ICWSM) hosted by the Georgia Institute of Technology in Atlanta.

“Atlanta is a very busy city with lots of highways, but it has beautiful parks, one of which I visited on a morning run,” Nguyen said.

He received ovations for his summation of  the paper’s findings, and for making the 15,000 km journey.

Reddit offerred target-rich environment

Prof Colin Klein of the ANU School of Philosophy identified Reddit’s “Am I the Asshole” (AITA) forum as a rich source of data to investigate machine learning in the moral sphere. The popular online community crowdsources moral deliberation one sticky situation at a time. 

First, users create a thread by summarising events that force them to make a difficult moral or ethical decision. Then, dozens or even hundreds of commenters weigh in. Collective judgments are reached via upvoting and downvoting comments marked by one of five tags: YTA (you are the asshole), NTA (not the asshole), ESH (everyone sucks here), NAH (no asshole here), and INFO (more information needed). 

Last year, Nguyen was brought on to conduct the research, collaborating with Professors Klein and Xie, as well as fellow ANU students Georgiana Lyall, Alasdair Tran, Minjeong Shin, and Nicholas Carroll.

“I think moral content appears everywhere on social media,” Nguyen said. “But this particular subreddit provided fertile ground by supporting discussion from a wide community with a set of strict rules that makes it fairly healthy."

The team analysed 100,000 threads — the largest such study to date —using algorithms to identify topic clusters. Human participants helped to refine the algorithm by deciding when to combine similar clusters and how to assign thematic designations to threads.

The resulting data identifies 47 meaningful topics that fall into five meta-categories: Identities, Aspects, Processes, Events, and Things.  “Aspects,” for example, includes such topics as hygiene, manners, and safety while “Events” includes weddings, breakups, and parties. 

The paper explains that clustering algorithms were designed to allow clusters to overlap, “since both the intersections and the gaps between two intuitive clusters (such as family and money) may be meaningful and interesting”. 

They found that most moral dilemmas included combinations of topics called “topic pairs”.  

Frequently occurring topic pairs included friends/manners, communication/mental health, and celebrations/manners.

“The model we used gives us a ‘soft’ categorization, meaning all topics are probable but some are more than others,” Nguyen said. “There are several reasons we decided to use two topics to define each post, the most important of which is that the human participants in our study tended to choose two topics to describe a post, even if we left the number of topics unrestricted.”

The data also showed that certain topics attracted comments and debate about other topics — for example, children and religion — while some topics repelled others, e.g., breakups and housework.

Previous research in this space, Nguyen said, has focused on automating moral decisions by designing machine learning models to classify and try to predict outcomes.  

“If the majority of the verdicts, say 70 per cent, go in one direction, the models selected that as correct,” Nguyen said. “We think that was skipping a few steps because, first of all, moral problems just don't have one answer. There’s a huge debate, and people with different outlooks come out with different choices.”

“So, if we want to build artificial intelligence that dovetails with human users at the moral and ethical level, we have to understand how people come up with these decisions: the process, not just the end product.”

Findings vindicate Confucians, ancient Greeks

The study suggests that the social sciences should look less at life-and-death decisions on train tracks or in the operating room, which are rare in everyday experience, and more at minor conflicts by which most people gauge morality. 

During his presentation in Atlanta, Nguyen was asked whether every post on AITA is about a moral issue, or whether there were also posts that addressed etiquette or conventional norms. “It was an interesting question, as my co-authors also sometimes disagreed on the definition of moral loading. The conversation was very thought-provoking,” he said.

Nicholas Carroll, a PhD candidate at the ANU School of Philosophy who contributed to the research, said conversations prompted by the “100,000 dilemmas” findings will intrigue social scientists and philosophers as well.

“Most philosophers distinguish between moral norms, such as the norm that we ought to keep our promises, and conventional norms, such as the norm that we ought to allow passengers to exit a busy train before we try to board it,” Carroll said. “This has resulted in philosophers excising from the realm of morality various dilemmas that fall on the conventional side of this distinction. But by contrast, the dilemmas that non-philosophers on AITA post about are mostly about what philosophers would consider conventional norms — dilemmas relating to, for instance, communication, family, friends, manners, and relationships.”

Carroll and Nguyen agreed that one of the most interesting revelations from the new research was how differently philosophers and non-philosophers think about morality.

“There is a gap between what philosophers think of morality and what everyday people think of morality,” said Nguyen. “These nuanced scenarios are encountered much more often than the idealised moral dilemmas that philosophers have been pondering for years.”

The study points out that a core tenet of early Confucian philosophy is that the everyday challenges and exchanges that people experience are of profound importance to morality. 

“Philosophers might do well to pay more attention to these kinds of interactions—doing so would bring contemporary moral philosophy more in line with the moral philosophy of the Ancient Greeks and Confucians, for instance,” Carroll said.

The debates expressed on AITA challenge the traditional distinction between moral norms and conventional norms like rules of etiquette. 

“The realm of morality is far larger than philosophers ordinarily think it is,” Carroll said.