The 3rd iteration of the Obfuscation Workshop is coming to Europe! It will take place at TU Delft, in the Netherlands, 11 and 12 May 2020.
Obfuscation can be seen as the art and science of protecting your privacy in contexts where your actions are being monitored and analyzed by other humans, organizations, or e.g. AI-powered technology. Obfuscation models are an important topic in the BEHAVE program, as obfuscation is considered a useful strategy to masque one’s true moral motivations, in contexts where giving them away might lead to contempt or feelings of shame.
This interdisciplinary workshop convenes researchers, scientists, policy makers, developers, and artists to discuss a broad range of technical, theoretical, and policy approaches to obfuscation, including tools, simulations and experimental methods that people and artificial agents use to obfuscate themselves and their environments in asymmetries of power and information.
You can read more about the last iteration of the workshop at http://www.obfuscationworkshop.org/report/ .
The organizing committee, consisting of Helen Nissenbaum @ Cornell University, Caspar Chorus & Seda Gurses @ TU Delft and Ero Balsa @ KU Leuven, will soon be sending out invitations and an open call for submissions. For now, please save the dates and feel free to forward this announcement to anyone who could find it of interest.
Our research on moral uncertainty for ethical AI is going to be featured in the AAAI/ACM conference on Artificial Intelligence, Ethics, and Society which will take place on February 7-8 in NY. PhD candidate Andreia Martinho will present our re-conceptualization of a metanormative framework for decision-making under moral uncertainty using Discrete Choice Analysis techniques and its operationalization using a latent class choice model. The relevance of moral uncertainty is illustrated in a proof of concept in which we conceptualize a society where AI Systems are in charge of making policy choices and investigate whether the choices of morally uncertain AI contrast with the choices of morally certain AI.
In November, three presentations based on BEHAVE-work were given to enthusiastic audiences. On November 6, the TU Delft Tracks in Transport conference hosted a special session organized by Andreia and Caspar, in which Andreia, Tom and Bing Huang presented work on the topic “On moral men and machines: Real ethical issues on the road”. In this special session highlights of recent work done in the BEHAVE project were presented, including new theoretical insights underpinned by empirical evidence on three thought-provocative questions: Can we predict aggressive driving behavior based on drivers’ moral values? Why do people find accidents caused by autonomous vehicles more unacceptable than those caused by human drivers? and What can bioethics and automotive industry reports teach us about dealing with moral issues surrounding autonomous vehicles?
On 20 November, Caspar gave a lunch lecture for the aiTech community, which brings together scholars with diverse AI-related disciplinary perspectives ranging from computer science to ethics and the behavioral sciences. Caspar’s talk, titled “Morality and taboos for men and machines”, started with a quick overview of how the BEHAVE-team integrates the newest insights from moral psychology into tractable mathematical formulations of moral choice behavior. Furthermore, it was shown how, building on such a mathematical representation of human morality, a human-inspired moral compass can be designed for a ‘morally uncertain’ Artificial Intelligence (AI).
On 28 November, as part of the so-called Dag van het Gedrag in The Hague, Tom, Maarten en Caspar co-hosted a special session on the difficulties associated with stearing human behavior through influencing moral values and attitudes. The session was oversubscribed, and participants (mostly civil servants) engaged with the speakers in a lively discussion about the pros and cons of tapping into moral values when trying to steer behavior towards more societally beneficial outcomes.
During the past couple of weeks, two new BEHAVE-research manuscripts have been published. The first is a Chapter about the moral dimension of regret, and its implications for choice set design. The design, or architecture, of choice sets is an increasingly established marketing tool to steer choices towards products with, e.g., a particularly high profit margin, in a subtle wat that is often not detectable by consumers. This manuscript shows, using the random regret minimization model which has been developed at TU Delft, how some of these choice architectures generate disproportional levels of regret and hence should be considered morally problematic. Food for thought for marketing professionals! The Chapter has appeared in the book The Moral Psychology of Regret, published by Rowman International.
Another manuscript, of which Ahmad Alwosheel is first author and Sander van Cranenburgh second author, was published in the Journal of Choice Modelling. It reconceptualizes techniques from the computer vision field to develop a procedure to build trust in the use of so-called Artificial Neural Networks (ANNs) for choice behavior analysis. ANNs are increasingly used to analyze and predict choice behavior, but their ‘black box’ nature causes problems in terms of explainability and interpretability. Especially in morally sensitive choice situations, this hampers the use of these models, as it precludes learning what moral values and trade-offs were at stake. The idea of having the trained neural network generate so-called prototypical examples helps gain trust, among analysts, that a particular neural network has learned intuitive relations and behavioral processes underlying the observed choice data. This in turn will help pave the way towards using these machine learning approaches for the analysis of moral decision making.
The Journal of Choice Modelling has published the call for papers for the special issue on Models of moral decision-making, which is connected to the special session of the International Choice Modelling Conference which we discussed further below. We welcome submissions on a range of fascinating topics, such as: Norm formation and its effect on choices; Altruistic and pro-social behaviour; Anti-social behaviour, deceit, obfuscation, taboos; Guilt, shame, remorse as determinants of choice behaviour; Decision-making in moral dilemmas; (social) Context effects on moral choice behaviour. Submission deadline: 30 September.
During the past semester, three MSc-students supervised by BEHAVE-members wrote theses on moral choice behavior. Nienke Pieters, hosted by Goudappel Coffeng where she was supervised by Dr. Matthijs Dicke-Ogenia, studied to what extent moral considerations influenced consumers’ decisions for safety-enhancing Advanced Driver Assistance Systems (ADAS); her thesis can be found here. Belle Visee, hosted by Mobycon where she was supervised by Babet Hendriks, studied the views of road users on (im-)polite behaviors by automated vehicles; her thesis can be found here. Anne-Fleur Tjon Joe Gin, hosted by Accenture where she was supervised by Rozemarijn de Koomen, studied to what extent consumers are willing to pay for more environmentally friendly package delivery services, and how this relates to their innate morality; her thesis can be found here. Congratulations to all three (former) MSc-students with obtaining their degree!
A paper written by Tanzhe Tang (PhD-candidate) and Caspar Chorus has been accepted for publication in the Journal of Artificial Societies and Social Simulation (JASSS), which is a leading journal in the field of social simulation. The paper presents a new model of opinion dynamics. In contrast with previously proposed models, our so-called AOI (action opinion inference) model postulates that people learn about each others’ opinions not by observing them directly, but by observing each others’ actions and interpreting those. This added level of behavioral realism has important implications for the predicted population-shares of various types of opinions. As such, the AOI model provides an important stepping stone for more realistic models of moral norm formation, which are currently being developed in the BEHAVE-program. Congratulations with this publication, Tanzhe!
The full abstract reads as follows: Opinion dynamics models are based on the implicit assumption that people can observe the opinions of others directly, and update their own opinions based on the observation. This assumption significantly reduces the complexity of the process of learning opinions, but seems to be rather unrealistic. Instead, we argue that the opinion itself is unobservable, and that people attempt to infer the opinions of others by observing and interpreting their actions. Building on the notion of Bayesian learning, we introduce an action-opinion inference model (AOI model); this model describes and predicts opinion dynamics where actions are governed by underlying opinions, and each agent changes her opinion according to her inference of others’ opinions from their actions. We study different action-opinion relations in the framework of the AOI model, and show how opinion dynamics are determined by the relations between opinions and actions. We also show that the well-known voter model can be formulated as being a special case of the AOI model when adopting a bijective action-opinion relation. Furthermore,we show that a so-called inclusive opinion, which is congruent with more than one action (in contrast with an exclusive opinion which is only congruent with one action), plays a special role in the dynamic process of opinion spreading. Specifically, the system containing an inclusive opinion always ends up with a full consensus of an exclusive opinion that is incompatible with the inclusive opinion, or with a mixed state of other opinions, including the inclusive opinion itself. A mathematical solution is given for some simple action-opinion relations to help better understand and interpret the simulation results. Finally, the AOI model is compared with the constrained voter model and the language competition model; several avenues for further research are discussed at the end of the paper.
During the first half of 2019, various BEHAVE-studies have been presented at symposia, workshops, and conferences. Tanzhe Tang presented his newest work on obfuscation and its effects on moral norm formation at the INAS symposium in St Petersburg, which brought together experts on Agent-Based Modelling for Theory Building in Social Sciences. Nicolas Cointe presented his work on how obfuscation influences coalition formation in multi-agent systems at the EXTRAAMAS workshop in Montreal, which focused on Explainable Transparent Autonomous Agents and Multi-Agent Systems. His paper was nominated for best paper award! Caspar Chorus presented BEHAVE-research at various occasions, including an opening keynote at the National Econometricians Day, a seminar at Leeds University’s Choice Modelling Centre, a plenary talk at a workshop on Collective Decision-Making at the University of Amsterdam, and a guest lecture at the Netherlands Defence Academy. Are you interested in receiving slides? Please send us an email. Note that much our most recent work will be presented at the forthcoming International Choice Modelling Conference in Kobe, Japan, where we are also hosting two special sessions on moral choice models (featuring work from scholars outside the BEHAVE-team). More updates to follow!
The International Choice Modeling Conference 2019, which is the premier venue for choice modelers world-wide, will feature a special session on Models of moral decision making. The session, which is sponsored by the BEHAVE-program and organized by Caspar Chorus, Jürgen Meyerhoff and Ulf Liebe, will present novel work concerning the (theoretical) development and (empirical) testing of mathematical representations of human decision-making in morally sensitive contexts. Authors of accepted abstracts will be invited to submit full papers to a corresponding special issue, to be published in the JCR-listed Journal of Choice Modelling. The call for abstracts can be found here. We look forward to receiving your submission! Note that submission to this special session does not count against the ‘one presentation per registration’-quotum of the conference.
Andreia Martins Martinho and Teodóra Szép presented their research in Utrecht at the TRAIL PhD Congress 2018.
Andreia’s presentation titled ‘Novel perspectives on ethics and Autonomous Vehicles’ can be found here, and Teodóra’s presentation, ‘Decision making with moral dimensions – Comparison of incentives for cooperative driving’, can be found here. Their abstracts -along with all other presenters’- can be found in the program book.