Constructing, Weighing and Evaluating
        Arguments to Solve Wicked Problems

                                Thomas F. Gordon

                              https://www.tfgordon.de



      Abstract. Most work in AI on knowledge representation and belief re-
      vision is founded in philosophical realism, and its simplifying assumption
      that objects and relations in the real-world can be represented, one-to-
      one, in a knowledge base, at least approximately, and improved over time
      via learning. According to this perspective, knowledge can first be ac-
      quired and then applied to solve problems. Here we present components
      of a decision-support system based on an alternative argumentation-
      based approach to knowledge representation and problem solving. In
      this approach, knowledge is socially constructed and subject to debate
      during the problem-solving process, in dialogues, in a more iterative way.

      Keywords: Argumentation · Issue-Based Information Systems · Prac-
      tical Reasoning · Knowledge Representation · Belief Revision


    In the classical symbolic AI approach to problem solving, knowledge is first
acquired and represented, using some knowledge-representation language, and
then applied to solve particular problems, using a knowledge-based or expert
system. This approach is founded in philosophical realism, which assumes that
objects and relations in the real-world can be represented, at least approximately,
one-to-one in a theory or conceptual model (correspondence theory of truth).
Over time, the theory can be revised and improved to take into account new
information.
    However, as Kunz and Rittel noticed in the early 1970s [8], in many problems
domains, such as the law, politics and city planning, the task of formulating or
framing the problem, acquiring knowledge about the problem domain and apply-
ing this knowledge to solve the problem are deeply intertwined and interdepen-
dent. In such domains, knowledge is socially constructed and subject to debate
during the problem-solving process, in dialogues. Kunz and Rittel proposed ar-
gumentation as a more dynamic and iterative method suitable for solving such
“wicked” problems. In this talk, components of a computational model of ar-
gument are presented which are designed for use in interactive decision-support
systems for helping people to collaboratively solve such problems. The compo-
nents provides support for automatically constructing (inventing, generating)
arguments, using presumptive inference rules, called “argumentation schemes”,
as well as evaluating these arguments by resolving attack relations among ar-
guments and weighing and balancing pros and cons to determine which options
proposed as solutions to issues have the best support.




                                        1
     The mainstream line of research in the field of computational models of
argument is based on Dung Abstract Argumentation Frameworks (AFs) [2, 1],
which are focused on modeling and resolving attack relations among arguments.
Dung intended AFs to be used in a pipeline model for solving problems, where
arguments are first generated from a knowledge-base, evaluated, and finally used
to determine which statements (propositions) can be accepted as true. Thus,
these mainstream computational models of argumentation are also well within
the foundations of philosophical realism, since they continue to assume that
knowledge can be acquired and modeled in some objective way before problem-
solving begins, without being subject to debate during the solving of particular
problems.
     Moreover, a common and arguably more typical form of human argumenta-
tion, where pros and cons are weighed and balanced to choose among alternative
options, cannot be simply and intuitively reduced to attacks, as in Dung AFs. In
[7, 5] we defined a new formal model of structured argument which generalizes
Dung AFs to provide better support for argument weighing and balancing, en-
abling cumulative arguments and argument accrual to be handled without caus-
ing an exponential blowup in the number of arguments. Dung’s pipeline model,
which evaluates statements after arguments, sequentially, makes it impossible to
make the acceptability or weight of an argument depend on the labels of their
premises. To overcome this problem, in our model the weight of arguments and
labels of statements can depend on each other, in a mutually recursive manner.
     The role of a knowledge-base in our approach is played by a rule-based rep-
resentation of a set of presumptive inference rules called argumentation schemes
[9, 10, 3]. In [6] we introduced a high-level declarative programming language for
representing argumentation schemes, where schemes represented in this language
can be easily validated by domain experts, including developers of argumentation
schemes in informal logic and philosophy, and serve as executable specifications
for automatically constructing arguments, when applied to a set of assumptions.
Since argumentation schemes are presumptive inference rules, both premises
and conclusions of schemes can be second-order schema variables, i.e. without
a fixed predicate symbol. Our language for representing argumentation schemes
is based on Constraint Handling Rules (CHR), a declarative, Turing complete,
forwards-chaining, rule-based programming language [4]. The expressiveness of
our scheme language has been validated by using it to represent twenty of the
most common argumentation schemes.
     We conclude with a discussion about the extent to which this argumentation-
based approach to knowledge-representation and problem-solving overcomes the
simplifying assumptions of philosophical realism and provides better support for
decision-making in domains with “wicked” problems.


References

 1. Baroni, P., Gabbay, D., Giacomin, M., van der Torre, L. (eds.): Handbook of Formal
    Argumentation. College Publications (2018)




                                        2
 2. Dung, P.M.: On the acceptability of arguments and its fundamental role in non-
    monotonic reasoning, logic programming and n-person games. Artificial Intelligence
    77(2), 321–357 (1995)
 3. van Eemeren, F.H., Garssen, B., Krabbe, E.C., Henkemans, A.F.S., Verheij, B.,
    Wagemans, J.H.M.: Handbook of Argumentation Theory. Springer (2014)
 4. Frühwirth, T.: Constraint Handling Rules. Cambridge University Press (2009)
 5. Gordon, T.F.: Defining argument weighing functions. IfCoLog Journal of Logics
    and their Applications 5(3), 747–773 (2018)
 6. Gordon, T.F., Friederich, H., Walton, D.: Representing argumentation schemes
    with constraint handling rules (chr). Argument and Computation 9(2), 91–119
    (2017)
 7. Gordon, T.F., Walton, D.: Formalizing balancing arguments. In: Proceeding of
    the 2016 conference on Computational Models of Argument (COMMA 2016). pp.
    327–338. IOS Press (2016). https://doi.org/10.3233/978-1-61499-686-6-327
 8. Kunz, W., Rittel, H.W.: Issues as elements of information systems. Tech. rep.,
    Institut für Grundlagen der Planung, Universität Stuttgart (1970)
 9. Walton, D.: Argumentation Schemes for Presumptive Reasoning. Erlbaum (1996)
10. Walton, D., Reed, C., Macagno, F.: Argumentation Schemes. Cambridge University
    Press (2008)




                                        3