turn, such choices might have consequences on the
methods used for translating the input problem in a
suitable form, as well as in the process of identifying
Discussion and Open
The proposed challenge is extremely hard if we con-
sider its ultimate goal, that is, to set a real competi-
tion between computers and humans on mathemati-
cal puzzles. However, it allows for a number of
different intermediate steps with increasing difficul-
ties, thus providing a nice playground for AI
researchers. Having in mind the challenge, and for
the sake of discussion, we cite a few research direc-
tions, knowing that being exhaustive would be
Means for Evaluating the Challenge
Given the breadth and the complexity of the challenge, the reader might question whether it is broad
enough to foster research advances with regard to
existing state-of-the-art solutions. In this respect, an
important feature of research challenges is the possibility of confronting them in a stepwise fashion, thus
providing short-term goals as well as long-term ones.
The stepwise nature of the proposed challenge is
threefold. First, we can approach problems at increasing complexity levels: among mathematical puzzles
of the primary school, we could first approach simpler then more complex problems, providing different scores in the competition. Second, we can proceed gradually, starting from mathematical puzzles
that can be solved using a single technique (for example constraint solving) to an ensemble of solution
techniques that should be selected by the agent.
Table 1. Overview of Recent Related Works and Challenges.
Domain Diagrams Everyday
; Query Answering
Seo et al. (2015) Geometric
Shi et al. (2015) Math Word
Roy, Vieira, and
Hosseini, et al.
Kushman, et al.
Zhou, Dai, and
et al. (2015)
Liang, et al.
Lev, et al. (2004) Logical Puzzles FOL Reasoner
Todai Project24 University Access
Mitra and Baral