In real world contexts of reasoning about evidence, that evi-dence frequently arrives sequentially. Moreover, we often can-not anticipate in advance what kinds of evidence we will even-tually encounter. This raises the question of what we do to ourexisting models when we encounter new variables to consider.The standard normative framework for probabilistic reasoningyields the same ultimate outcome whether multiple pieces ofevidence are acquired in sequence or all at once, and it is in-sensitive to the order in which that evidence is acquired. Thisequivalence, however, holds only if all potential evidence isincorporated in a single model from the outset. Hence little isknown about what happens when evidence sets are expandedincrementally. Here, we examine this contrast formally and re-port the results of the first study, to date, that examines howpeople navigate such expansions.

As AI systems come to permeate human society, there is an increasing need for such systems to explain their actions,conclusions, or decisions. This is presently fuelling a surge in interest in machine-generated explanation. However,there are not only technical challenges to be met here; there is also considerable uncertainty about what suitable targetexplanations should look like. In this paper, we describe a case study which makes a start at bridging between machinereasoning, and the philosophical and psychological literatures on what counts as good reasoning by eliciting explanationsfrom human experts. The work illustrates how concrete cases rapidly move discussion beyond abstract considerationsof explanatory virtues toward specific targets more suitable for emulation by machines. On the one hand, it highlightsthe limitations of present algorithms for generating explanations from Bayesian networks. At the same time, however, itprovides concrete direction for future algorithm construction.

In this paper, we explore the objective-Bayesian principle of minimum information and Maximum Entropy as a solution to the problem of awareness growth: how should rational agents adjust their beliefs upon becoming aware of new possibilities? We introduce the Maximum Entropy principle as a theoretical solution to the problem of awareness growth and present the results of two experiments conducted to compare human reasoners' responses with the theoretical prescriptions of the Maximum Entropy approach. We discover that, although the MaxEnt method may appear computationally demanding, participants' responses are largely consistent with the theoretical prescription.

Recent research suggests that people do not perform wellon some of the most crucial components of causal reason-ing: probabilistic independence, diagnostic reasoning, and ex-plaining away. Despite this, it remains unclear what con-texts would affect people’s reasoning in these domains. Inthe present study we investigated the influence of manipulatingpriors of causes and structural complexity of Causal BayesianNetworks (CBNs) on the above components. Overall we foundthat participants largely accepted the priors and understoodprobabilistic independence, but engaged in inaccurate diagnos-tic reasoning and insufficient explaining away behavior. More-over, the effect of manipulating priors on participants’ perfor-mance in diagnostic reasoning and explaining away was sig-nificantly larger in a structurally less complex CBN than in astructurally more complex CBN.

Recent years have seen a surge of interest in probabilistic measures of argument strength. Such measures have been used to elucidate longstanding questions of both theoretical and practical interest around fallacies of argumentation, scientific argument, legal argument, or argument and evidence aggregation. The measures typically used have drawn, in one form or other, on conditional probability as a central component for measuring argument quality, strength or confirmation. Work in this area has also consistently highlighted the value of Bayesian Belief Networks for computing argument strength in multi-argument (variable) contexts. However, to date, there has been no consideration of whether the key structural property underlying such networks – the notion of conditional independence- has a bearing on argument quality. In this paper, we consider the potential impact of causal structure on the utility of arguments.