Prof. Caspar Chorus

Caspar Chorus, Professor of Choice Behavior Modeling, TU Delft.

Obfuscation maximization for discrete choice analysis.

Wednesday 6 March 2018, 10:00 to 11:00
Institute for Transport Studies, room 1.11, University of Leeds

Abstract:

Discrete choice theory is founded on the idea that the preferences of individuals echo through in the choices they make. I argue that in some situations, this key assumption does not hold; decision-makers may want to obfuscate – ‘hide’ – their preferences from onlookers, e.g. to protect their privacy, or to avoid punishment. Such obfuscation behaviour could for example be relevant in in the context of flirting; decision-making in the context of moral dilemmas; and political decision-making. In these and many other contexts, the decision-maker has an incentive to be less than fully transparent about his or her preferences. In this presentation, I propose a model of choice behaviour which formalizes this notion of obfuscation based decision-making, by combining the notions of Bayesian updating and Shannon entropy maximization. In short, the model postulates that the decision-maker believes that the onlooker (which could be a choice modeler) attempts to learn his or her preferences by means of a Bayesian learning process using his or her choices as input. The result of this process is a posterior probability distribution (in the mind of the onlooker) over the preference sets employed by the decision-maker. The decision-maker then maximizes the Shannon entropy (i.e., the uncertainty in the mind of the onlooker) which is implied by this posterior. A decision maker who operates in this way, will succeed in keeping the onlooker ‘in the dark’ regarding his preferences. After having presented this formal model of obfuscation based discrete choice, I will show how the model can be used to understand decision making as well as to help onlookers design choice sets that maximize the cost (to the decision-maker) of obfuscation. Finally, some very recently obtained empirical results derived from an Obfuscation-game will be shared.

About Caspar:

Caspar's research aim is to develop and empirically validate models of human decision-making that combine high levels of behavioral realism and mathematical tractability. In recent years, funded by personal Veni (2010) and Vidi (2012) grants from NWO, Caspar and his co-workers have developed choice models based on bounded rationality, including the now widely used random regret minimization model suite.

Most of his current work is focused on designing moral choice models, which capture the heuristics and considerations that humans employ in morally sensitive situations. This work is funded by means of an ERC-Consolidator grant, and should ultimately lead to a specific, 'human' type of artificial morality for AI. See behave.tbm.tudelft.nl for more information about this research program.