=Paper=
{{Paper
|id=Vol-3153/paper2
|storemode=property
|title=What if gamified software is fully proactive? Towards autonomy-related design principles
|pdfUrl=https://ceur-ws.org/Vol-3153/paper2.pdf
|volume=Vol-3153
|authors=Esteban Guerrero,Tero Vartiainen,Panu Kalmi
|dblpUrl=https://dblp.org/rec/conf/persuasive/GuerreroVK22
}}
==What if gamified software is fully proactive? Towards autonomy-related design principles==
https://ceur-ws.org/Vol-3153/paper2.pdf
What if gamified software is fully proactive?
Towards autonomy-related design principles ⋆
Esteban Guerrero1[0000−0002−6035−800X] , Tero Vartiainen1[0000−0003−3843−8561] ,
and Panu Kalmi2[0000−0002−1359−5109]
1
School of Technology and Innovations, University of Vaasa, Finland
2
School of Accounting and Finance, Economics, University of Vaasa, Finland
{esteban.guerrero, tero.vartiainen, panu.kalmi}@uwasa.fi
Abstract. Computational agents are a type of software architectures
designed to be autonomous and social, meaning that they can make de-
cisions proactively, reacting also to stimuli from the environment. The
use of such architectures is not common in the gamification field area,
instead, gamified software has traditionally reactive characteristics, re-
sponding to user actions disregarding the possibility of proactive be-
havior. In this paper, we propose four formal principles for designing
autonomous gamified systems, to ensure traceability of gamified outputs,
internal consistency of gamification attempts, coherent agent-user in-
teraction, and formal conditions to assess user actions from a rational
perspective. We present our initial work on these general principles, high-
lighting our empirical future work.
· · ·
· ·
Keywords: Persuasive technology Gamification Software agents
Principles Formal approaches Argumentation dialogues.
1 Motivation
In the artificial intelligence (AI) field, the sense of “autonomy” is not precise,
but the term is taken to mean that software agents’ activities do not require
constant human guidance or intervention [19]. An object is an agent (e.g. a
software agent) if it serves a useful purpose either to a different agent, or to itself,
in which case the agent is called autonomous [12]. Being those purposes the state
of affairs to be achieved, in other words, the goals of an agent. To exemplify this
notion, let us suppose that a person wants to increase her monthly savings by
counting the accumulated expenses using a financial app. That software is her
agent that keeps her savings count, “adopting” her goal to motivate herself for
saving money. Note that the app has a transient agency, if the user does no
longer needs to save money (e.g. she wins the lottery), such an agent becomes a
simple object with no ascribed agency.
©
⋆
Persuasive 2022, Adjunct Proceedings of the 17th International Conference on Per-
suasive Technology. Copyright 2022 for this paper by its authors. Use permitted
under Creative Commons License Attribution 4.0 International (CC BY 4.0).
2 Esteban Guerrero, Tero Vartiainen, and Panu Kalmi
An autonomous agent is not dependent on the goals of others, it possesses
goals that are generated from within rather than adopted [20]. For example, an
autonomous version of that financial app, could change its goal (proactively)
and help her to visualize different philanthropic goals, without any guidance or
intervention.
A high-level question directing this research connects the aforementioned
autonomous agents and gamification field, what if a gamified software became
fully autonomous?. Empirical answers to this question from a gamification per-
spective are scarce, and theoretical frameworks of gamification dealing with this
issue are practically non-existent (see reviews [6,11]). From the AI perspective,
“human oversight” is one of the requirements being put forward as a means to
support human autonomy and agency [13], where theoretical (we will use the
term formal ) guidelines have been proposed, aiming to delineate autonomous
behavior of agents considering responsible and transparent mechanisms [2]. In
fact, high-level principles and guidelines [8,14,16] are commonly used in gamifica-
tion, but most of them are not aligned with autonomous software characteristics,
nor serve as grounded specifications for developing actual software. In this con-
text, our ongoing research proposes four principles for designing autonomous
gamification technology considering: traceability as a mechanism for meaningful
human control, coherence and consistency during the interaction autonomous
agent-human, and rationality of the decisions that a proactive agent makes. In
summary, the proposed principles are:
Principle 1 : Traceability of gamified outputs. Establishes that gamified affor-
dances (outputs) need to provide a transparent and identifiable explanation
of the persuasive attempt.
Principle 2 : Internal consistency of a gamification attempt. Defines formal
requirements of the informational elements (e.g. content, visualization, etc.)
of a persuasive attempt.
Principle 3 : Coherent gamified interaction. Characterizes the type of interac-
tion that a persuasive agent should (or should not) make.
Principle 4 : Rational persuasive gamification: Determine the formal condi-
tions that an agent needs to consider if a user action can be considered as
rational.
We formalize these principles in propositional logic to be used by designers
of formal mechanisms of decision-making for software agents (e.g. [4] among
others).
2 Methods
We use a formal method (framework) based on argumentation-based games [17]
for describing the agent-user interaction. In this paper, a user model is a tuple:
U = ⟨B, I, Be, PI , PB , ⪯B , ⪯I ⟩. In which the probability distributions PI and
PB relate the subjective probability of intentions and beliefs, and hierarchies of
beliefs and intentions are given by ⪯B , ⪯I .
Autonomous gamification 3
Option 1: User
Ag take a loan selection
𝜮gf g-move
Gamified story-scenario
Option 1:
𝜮gm take a loan
Option 2:
modest
consumption
Fig. 1: The user-agent gamified interaction used in this paper.
In this paper, we consider gamification mechanisms (gm) e.g. avatars, stories
and leader boards, and gamification feedback (gf ), which are visual affordances
presented to a user e.g. rewards, feedback messages, etc. This classification of
gamification affordances follows a taxonomy presented in [5]. We assume that
the agent has two databases Σgm and Σgf , containing gm and gf affordances.
We also consider preferences among affordances gm and gf , which is a pre-order
function ⪯gm and ⪯gf , we assume a preexisting order. In this context, a user
and an agent exchange information regarding a particular topic T , e.g. about
financial literacy.
We use propositional logic with ¬ to express logical negation, x denoting un-
certainty (w.r.t. a true, false valuation), ⊢ deductive inference, and ⊢s semantic
interpretation, and ≡, ≡s for syntactic and semantic equivalence. We also use a
handy function for updating information UPD(old, new).
Agent Ag, as a gamified persuasive technology
is oriented to generate as output a gamified assert(p,Persuader )
move (gmove), which is a tuple ⟨sa, cont, vis⟩
formed by: 1) a speech act (sa) that is the assert(q,User)
assert(r,User)
intended action of the agent within the per-
assert(q1,Persuader )
suasive exchange as a dialogue, 2) a persua- …
sive content (cont) that is the underlying mes- assert(q2,User)
assert(q3,User)
sage to be transmitted to a user, and 3) a vi-
…
sual cue (vis). sa are predefined actions such …
as accept, assert, question, reject, ignore (see Fig. 2: Protocol of gamified per-
technical details in [5]). We use the nota- suasive interaction
tion gmoveAg→U
ti to express that a gmove was
made from Ag to U at ti , which is omitted if time and move direction is evident
or unnecessary. We also use three handy functions CONT(gmove) = {content},
VIS(gmove) = {visual} and SA(gmove) = { speech act } to return the content
and the visualization of a given gmove. An agent Ag uses as input the aforemen-
tioned three databases ΣT , Σgf and Σgm , and a model of the user U ∗ (where
U ∗ ⊆ U, denoting that it may have not perfect information).
4 Esteban Guerrero, Tero Vartiainen, and Panu Kalmi
3 Results - Principles for autonomous gamified systems
We define these principles considering three assumptions: 1) a user establishes
a communication with the agent using gamified affordances, 2) the agent has
information about beliefs, intentions, and preferences of the user, and 3) the
user follows some protocols of communication with the agent.
Principle 1 (Traceable gamified output) A persuader agent should be able
to provide traceable explanations for every gamified output. Formally, if S ⊢s
gmoveAg→U is the gamified move output, and S is the knowledge source of the
move, the following criteria should be fulfilled:
Formalism Explanation
A gamified output of an agent should be the
S ⊆ (Σgf ∪ Σgm ∪ U ∗ ) consequence of an inference process based on a
set of gamified mechanisms and the user model.
Determinants of a gamified move should be
S ̸= ∅
identifiable.
This first principle relies on the traceability of the semantic inference (⊢s ). In
the AI literature, violations of these formalisms have been investigated to avoid
black-box -style algorithms [18], and handcrafted processes with no back-track
inference mechanism.
Focusing on the internal definition of the gamified output, the following set
of principles establishes basic conditions of consistency of such output.
Principle 2 (Internal consistency of a gamification move) A gamification
move is consistent if the following holds:
Formalism Explanation
CONT(gmoveAg→U ) ≡s
The content and the visualization of a gamified
VIS(gmoveAg→U )
move should be semantically coherent.
The content of a gamified move should be
CONT(gmoveAg→U ) ⊆ T
within the agreed persuasive topic.
VIS(gmoveAg→U ) ∈ ⪯gm The visualization should be part of an agreed
∪ ⪯gf set of gamified affordances.
Principle 2 can be seen close as a design principle for visual and content
aspects of a gamification process, in which the information carried by a gamified
affordance has to be consistent with its visualization.
The next set of principles is related to the gamified interaction between agent
and user, specifically, these formalisms try to avoid design practices against
engagement and commitment of the agent to the users’ goals.
Principle 3 (Coherent gamified interaction) A gamified persuasive inter-
action between U and Ag is coherent if the following conditions hold:
Autonomous gamification 5
Formalism Explanation
A gamified persuasive output should not be
gmoveti ̸= gmoveti+1
repetitive regardless of the state of U
∀ gmoveU
ti
→Ag
, A persuasive agent should not ignore
SA(gmoveU →Ag
ti+1 ) \ {ignore}
petitions from a user.
When a user ignores a gamification move,
SA(gmoveU →Ag ) = {ignore},
then the agent must update the gamification
then UPD(⪯gf , ⪯gf )
feedback preferences.
U →Ag When a user rejects a gamification move, the
SA(gmove ) = {reject}, then
agent must update the gamification feedback
UPD(⪯gf , ⪯gf ), UPD(U)
preferences and the beliefs of the user model.
The set of principles defined as coherent gamified interactions establishes a
guideline checklist of what an agent should do when a user communicates directly
through gamified moves.
A last set of principles establishes conditions to assess the rational actions
of a user. In the agents’ literature, the user’s intention to X (e.g. X = walk)
provides the agent with support to believe that s/he will do X, i.e. beliefs and
intentions are in the same “direction” [3]. The following formalisms capture the
notion of rationality considering an alignment between beliefs, intentions, and
actions.
Principle 4 (Rational persuasive gamification) An agent Ag persuasive gam-
ification can be considered rational, incomplete or irrational if the following
holds:
Formalism Explanation
U →Ag A move is rational if it contains a belief that
If CONT(gmove ) ∈ B and
is in the agent’s user model, and it is part of
⪯I , then such move is rational
the hierarchy of preferred intentions.
U →Ag The model of a user is
If CONT(gmove ) ∈⪯I and
intention-belief-incomplete if a move is
CONT(gmoveU →Ag ) ∈ B, then U ∗
inline with the preferred intentions but
is intention-belief-incomplete
contrary to the belief model.
If CONT(gmoveU →Ag ) ∈⪯I and A move is irrational if the move content is part
¬(CONT(gmoveU →Ag )) ∈ B, then of the hierarchy of preferred intentions but at
such move is irrational the same time is against the set of beliefs.
4 Discussion and future work
Current gamification literature (see reviews [10,11]) shows five main trends: 1)
the popularity of a limited number of affordances such as leader-boards, rewarded
points, and textual or visual feedback; 2) goal-orientation of the theoretical gam-
ification foundation; 3) the extended use of competitiveness and cooperativeness
mechanisms for gamification, and 4) a gradual generalization of tailoring gami-
fied persuasion. However, most of these approaches do not consider the notion
of autonomy or proactiveness of the gamified software.
6 Esteban Guerrero, Tero Vartiainen, and Panu Kalmi
On the AI-side, persuasion is a well-established research track, especially in
the argumentation theory [9]. Computational persuasion is a highly regulated
process that leads to the design of argumentation-based dialogue games, which
is a protocol -based exchange of information between two agents. Shortcomings
of these dialogues are well-identified (see [9]), such as the harnessing of finding
optimal decisions (moves) at every game stage, and the use of only exocentric
persuasion [7] disregarding context information or mental states of a user.
We introduced a set of rules aiming to promote transparency of persuasive
attempts (Principle 1), which are in line with the current joint effort that the
European Union and other leading AI countries to develop guidelines for trust-
worthy AI 3 . In the human-computer interaction field, trustworthiness was high-
lighted to be a fundamental principle for system credibility [16], explainable and
persuasive interfaces [15]. Consistency between visual and content aspects is rel-
evant for most of the gamification used in persuasive attempts. Principle 2 is
linked with the work presented in [14], where Nemery et al. have highlighted
the importance of visual consistency, introduced a set of principles for persua-
sive interfaces. The third set of formalisms, Principle 3, establishes a minimum
set of rules for coherent interaction between an agent and a user. We acknowl-
edge that the type of interactions are linked to argumentation-based dialogues,
which limits the type of potential interactions that other gamification mech-
anisms can produce. Nevertheless, Principle 3 is a generalization of different
proactive gamification mechanisms that have been investigated in the literature
(see [5]). Finally, our set of principles for evaluating rational inputs from the
user (Principle 4), which are based on the well-established theory of practical
reason of Bratman in [1], establishes a guide for evaluating if the user’s actions
are aligned with the information that an agent possesses about the user.
A key limitation of this work is the lack of empirical validation. Our im-
mediate future work will be to evaluate empirically these principles using a real-
world scenario. Currently, we are designing the following version of the gamified
platform (omitted details for blind review) to support financial decisions. We
also want to further establish an axiomatization of those principles, considering
different types of gamification affordances and different types of software agents
(e.g. [5]).
References
1. Bratman, M.: Intention, plans, and practical reason. Harvard University Press
(1987)
2. Dignum, V.: Responsible Artificial Intelligence: How to Develop and Use AI in a
Responsible Way. Springer International Publishing (2019)
3. Fishbein, M., Ajzen, I.: Belief, Attitude, Intention, and Behavior: An Introduction
to Theory and Research. Addison-Wesley series in social psychology, Addison-
Wesley Publishing Company (1975)
3
Ethics guidelines for trustworthy AI see https://digital-strategy.ec.europa.eu/en/
library/ethics-guidelines-trustworthy-ai, last access November 29 2021
Autonomous gamification 7
4. Guerrero, E., Lindgren, H.: Practical Reasoning About Complex Activities.
SpringerLink pp. 82–94 (Jun 2017)
5. Guerrero, E., Lindgren, H.: Typologies of persuasive strategies and content: a for-
malization using argumentation. In: International Conference on Practical Appli-
cations of Agents and Multi-Agent Systems. pp. 101–113. Springer (2021)
6. Hamari, J., Koivisto, J., Sarsa, H.: Does gamification work?–a literature review of
empirical studies on gamification. In: 2014 47th Hawaii international conference on
system sciences. pp. 3025–3034. Ieee (2014)
7. de la Hera Conde-Pumpido, T.: Persuasive gaming: Identifying the different types
of persuasion through games. International Journal of Serious Games 4(1), 31–39
(2017)
8. Horvitz, E.: Principles of mixed-initiative user interfaces. In: Proceedings of the
SIGCHI conference on Human Factors in Computing Systems. pp. 159–166 (1999)
9. Hunter, A.: Computational Persuasion with Applications in Behaviour Change. In:
COMMA. pp. 5–18 (2016)
10. Klock, A.C.T., Gasparini, I., Pimenta, M.S., Hamari, J.: Tailored gamification:
A review of literature. International Journal of Human-Computer Studies 144,
102495 (2020)
11. Koivisto, J., Hamari, J.: The rise of motivational information systems: A review
of gamification research. International Journal of Information Management 45,
191–210 (2019)
12. Luck, M., d’Inverno, M., et al.: A formal framework for agency and autonomy. In:
Icmas. vol. 95, pp. 254–260 (1995)
13. Methnani, L., Aler Tubella, A., Dignum, V., Theodorou, A.: Let Me Take Over:
Variable Autonomy for Meaningful Human Control. Front. Artif. Intell. 0 (2021)
14. Némery, A., Brangier, E.: Set of guidelines for persuasive interfaces: Organization
and validation of the criteria. Journal of Usability Studies 9(3) (2014)
15. Némery, A., Brangier, E., Kopp, S.: First Validation of Persuasive Criteria for
Designing and Evaluating the Social Influence of User Interfaces: Justification of
a Guideline. In: Design, User Experience, and Usability. Theory, Methods, Tools
and Practice, pp. 616–624. Springer, Berlin, Germany (Jul 2011)
16. Oinas-Kukkonen, H., Harjumaa, M.: Persuasive systems design: Key issues, process
model, and system features. Communications of the Association for Information
Systems 24(1), 28 (2009)
17. Parsons, S., Wooldridge, M., Amgoud, L.: An analysis of formal inter-agent dia-
logues. In: Proceedings of the first international joint conference on Autonomous
agents and multiagent systems: part 1. pp. 394–401. ACM (2002)
18. Rudin, C.: Stop explaining black box machine learning models for high stakes
decisions and use interpretable models instead - Nature Machine Intelligence. Nat.
Mach. Intell. 1, 206–215 (May 2019)
19. Shoham, Y.: Agent-oriented programming. Artificial intelligence 60(1), 51–92
(1993)
20. Wooldridge, M., Jennings, N.R.: Agent theories, architectures, and languages: A
survey. SpringerLink pp. 1–39 (Aug 1994)