=Paper=
{{Paper
|id=Vol-1419/section0013
|storemode=property
|title=None
|pdfUrl=https://ceur-ws.org/Vol-1419/section0013.pdf
|volume=Vol-1419
}}
==None==
https://ceur-ws.org/Vol-1419/section0013.pdf
High-level inference through mental simulation
Chairperson
Robert Mackiewicz
Department of Psychology, University of Social Science and Humanities,
PL - 03815 Warsaw, Poland.
rmackiew@swps.edu.pl
Discussant
Sangeet Khemlani
Navy Center for Applied Research in Artificial Intelligence, Naval Research Laboratory
Washington, DC 20375 USA
skhemlani@gmail.com
Speakers
Philipp Koralus
Faculty of Philosophy, University of Oxford
Oxford, OX2 6GG, UK
philipp.koralus@philosophy.ox.ac.uk
Robert Mackiewicz
Department of Psychology, University of Social Science and Humanities
PL - 03815 Warsaw, Poland
rmackiew@swps.edu.pl
Walter Schaeken
Laboratory for Experimental Psychology, KU Leuven
B - 3000 Leuven, Belgium
Walter.Schaeken@ppw.kuleuven.be
Marco Ragni
Center for Cognitive Science, University of Freiburg
D-79098 Freiburg, Germany
ragni@informatik.uni-freiburg.de
Reasoners without any background in logic can make Despite considerable theoretical development in the last
valid deductions. They can reason about sentences and 30 years, open questions remain: how does simulation
relations (Mackiewicz & Johnson-Laird, 2012), ascribe synthesize deductive, inductive, and abductive reasoning?
culpability and causality (Bucciarelli et al., 2008), creatively How does it develop? How do reasoners incorporate
generate algorithms to solve tasks (Khemlani et al., 2013), uncertainty into their simulations? Do simulations arise in
make inferences about mechanisms and physical scenes non-linguistic contexts? Researchers have begun to
(Hegarty, 2004; Battaglia et al., 2013), and construct investigate each of these outstanding issues. This
explanations to cope with inconsistencies (Johnson-Laird et symposium highlights recent insights from the last five
al., 2004). Recent evidence implicates mental simulation as years into the pivotal role that mental simulation plays
the conceptual foundation of all these behaviors (Johnson- across a broad swathe of high-level reasoning behavior.
Laird & Khemlani, 2014). People appear to build small- Discussants will highlight developmental trends,
scale discrete mental simulations that mimic the relations of computational models, and new data that provide
what they represent, and Craik (1943) was the first to converging progress toward a unified theory of human
explore their importance in thinking. The idea can be used reasoning based on mental simulation.
to predict reasoning difficulty: the more simulations
reasoners have to build for a given problem, the harder that
problem will be.
22
Illusory inferences and the erotetic theory of difficulty by the initial mental model and the possible
reasoning number of models. In this talk I will first introduce
prominent theories for relational reasoning. In a second step
I will analyze their predictions for cognitive complexity and
Philipp Koralus
discuss if measures from artificial intelligence can provide
additional insights.
Human reasoners are subject to fallacious inferences from
very simple premises that have been described as
tantamount to cognitive illusions (Walsh & Johnson-Laird, Training of Spatial Reasoning
2004; Khemlani & Johnson-Laird, 2009). We present new
experiments that show that these phenomena are much more Walter Schaeken
general and systematic than has previously been thought,
including inferences from disjunctive premises and premises The mental models theory of relational reasoning postulates
involving quantifiers. The novel illusory inferences we that individuals reason by constructing the possible models
present are predicted by the erotetic theory of reasoning of the situation described by the premises. The present
(Koralus and Mascarenhas, 2013). The key idea is that, by article reports two experiments about spatial relational
default, we reason by interpreting successive premises as reasoning and focuses on the possibility of training In
questions and maximally strong answers to those questions, Experiment 1, we compared two different training methods,
which generates the observed fallacies. one in line with the mental models theory and one in line
with the rule-based account Both accuracy and training data
supported the mental models theory. In Experiment 2, we
Kinematic mental simulations in childrens’ compared different training methods for children. Again,
results were in line with the mental models theory. Hence,
abduction of algorithms
training both children and adults in small-scale discrete
mental simulations that mimic the relations expressed by the
Robert Mackiewicz premises enhances the reasoning performance.
The theory of mental models postulates that the creation of
algorithms depends on kinematic mental simulations. We
References
present three experiments with children whose task was to
devise informal algorithms to rearrange the order of cars in Battaglia, P. W., Hamrick, J. B., & Tenenbaum, J. B.
trains (using a siding). Children were able to solve (2013). Simulation as an engine of physical scene
rearrangements of trains containing six cars and the minimal understanding. Proceedings of the National Academy of
theoretical number of moves predicted the difficulty of Sciences, 110, 18327-18332.
rearrangement (Experiment 1). When children were asked to Bucciarelli, M., Khemlani, S., & Johnson-Laird, P.N.
create and verbally describe algorithms for rearrangements, (2008). The Psychology of Moral Reasoning. Judgment
the difficulty of the task depended not on the number of and Decision Making, 3, 121-139.
moves but on the theoretical complexity of the algorithm Craik, K. (1943). The Nature of Explanation. Cambridge:
(Experiment 2). Children used many gestures mimicking Cambridge University Press.
actual moves in formulating their algorithms. Gestures Hegarty, M. (2004). Mechanical reasoning by mental
obviate verbal identifications of cars and descriptions of simulation. Trends in Cognitive Sciences, 8, 280-285.
their moves. A final study supported this hypothesis: Johnson-Laird, P.N. & Khemlani, S.S. (2014). Toward a
children formulated accurate algorithms on 13% more trials unified theory of reasoning. Psychology of Learning and
when they were able to gesture than when they were unable Motivation, 59, 1-42.
to gesture (Experiment 3). Johnson-Laird, P.N., Girotto, V., Legrenzi, P. (2004).
Reasoning from inconsistency to consistency.
Tracing Cognitive Complexity in Relational Psychological Review, 111, 640-661.
Khemlani, S.S. & Johnson-Laird, P.N. (2009). Disjunctive
Reasoning
illusory inferences and how to eliminate them. Memory &
Cognition, 37, 615 – 623.
Marco Ragni
Khemlani, S.S., Mackiewicz, R., Bucciarelli, M., &
Johnson-Laird, P.N. (2013). Kinematic mental
The core interest from a cognitive modeling perspective is
simulations in abduction and deduction. Proceedings of
to find theory inherent predictions for human reasoning
the National Academy of Sciences of the United States of
difficulty typically measured by error rates or response
America, 110, 16766-16771.
times. The theory of mental logic, for instance, claims that
Koralus, P. & Mascarenhas, S. (2013). The erotetic theory
reasoning difficulty depends on the number and kind of
of reasoning: Bridges between formal semantics and the
rules that need to be applied to derive a conclusion. In
psychology of deductive inference. Philosophical
contrast the mental model theory explains reasoning
Perspectives, 27, 312-365.
23
Mackiewicz, R. & Johnson-Laird, P.N. (2012). Reasoning
from connectives and relations between entities. Memory
& Cognition, 40, 266-279.
Walsh, C.R. & Johnson-Laird, P.N. (2004). Co-reference
and reasoning. Memory & Cognition, 32, 96-106.
24