IEEE Technology and Society Magazine - December 2019 - 58

Therefore, according to law 11 in The Fédération
Internationale de Football Association's (FIFA) "Laws
of the Game" [15] - a player in an offside position is
only penalized if, at the moment the ball touches or
is played by one of his team, s/he is, in the opinion
of the referee, involved in "active play," either by
direct participation, interfering with an opponent, or
gaining an advantage. Leaving the rule to be decided
based on the referee's opinion makes it subjective,
open to interpretation, and hence the source of considerable controversy.
SmartFlag [17] is an automated offside detection and
decision system. By the definition of the offside rule, a
player can be in an offside position, but would not necessarily be committing an offside foul unless the ball is
played to him and he displays intent to play the ball.
SmartFlag combines a ball and player localization subsystem based on sensor networks5 with a decision-making subsystem based on machine learning. The
intention is to distinguish between an offside situation
(based on raw sensor data) and offside offense (based
on interpretation of that data).
By detecting and calling offsides independently,
SmartFlag would be not just be a corrective measure
(like VAR), but a definitive one (leaving aside the difficulties of "explainable AI" to justify its decisions, in systems such as SmartFlag [18]). Although this approach
could lead to consistency and preclude incorrect offside
calls (by definition), taking such a "code is law"
approach [16] to the administration of rules has a number of implications.
Nevertheless, these advantages are predicated on
the assumption that the efficiency and consistency of
automated decision-making (oblivious of any "outsidetext") is necessarily preferable to, or more acceptable
than, labor-intensive and potentially inconsistent human
decision-making based on a rule-set that itself contains
the need for interpretation and opinion. The rest of this
article is concerned with considering the parameters of
that assumption.

Implications of Algorithmic Definition in Sport
There is inherent ambiguity of some rules, particularly
offside, that must be adjudicated by referees in different
sports. While technological solutions are being developed that can provide decision support, further developments can provides innovative systems for decision
making. A particular feature of such systems is that it
would reduce "fuzzy" situations that are uncertain,
hypothetical, and open to interpretation (like offside) to
a "crisp" binary decision like in/out of a tennis ball.
5

High-quality sports digitization and performance monitoring equipment is
commercially available, e.g., Kinexon (https://kinexon-sports.com) provides
centimeter-accurate position and movement data in three dimensions.

58

This provides a new perspective on applying rules in
sport. One approach to definition is to write down a set
of rules (i.e., a specification), and then classify an object
or an event according to whether it fits the definition or
not.6 An alternative approach to definition is ostensive
definition, where one points to an object and defines it
by direct demonstration. The system proposed here
allows for what might be called algorithmic definition: a
player is offside (by definition) if the algorithm implementing the rule to define the situation (rather than
interpreting the situation according to the rule) outputs
"the player is offside."
The rest of this article is concerned with this
proposition: While we could define offside this way,
is it preferable or even acceptable to do so? We raise
this concern because the outcome could have a significant impact on the acceptability of algorithmic
governance,7 whereby the computer defines a space
of legal or valid actions and the user is limited to a
choice within that space, and the more general use
of data-driven algorithms in law enforcement and
judicial proceedings [19]. The issue is "technology
creep": if it is acceptable to use AI and algorithms
for definition, decision-mak ing, and governance
in sport, then once that becomes normalized, it
becomes more acceptable to use AI in other domains
as well.
The following discussion is structured from three perspectives: firstly, social implications, i.e., is algorithmic
definition a preferable means of decision-making; secondly, juridical implications, i.e., is algorithmic definition an appropriate means of administering justice and
enforcing rules; and finally, what we call marginalization, the potential impact of supplanting human decision making by algorithms.

Social Implications
To assess the question of preference, we consider each
of the stakeholders in the sport: the team managers and
player, the referees, the spectators, journalists, and society "at large."
For team managers and players, one of their primary
complaints is lack of consistency in the application of
rules, not just between different games refereed by different officials, but even within the same game refereed
by the same officials. Given that SmartFlag would be the
same system trained on the same dataset and used in
6

Allegedly, when Plato gave the definition of man as "featherless bipeds,"
Diogenes the Cynic plucked a chicken and brought it into Plato's Academy,
saying, "Behold! I've brought you a man," and so the Academy added "with
broad flat nails" to the definition. See https://en.wikipedia.org/wiki/Diogenes.
7
This is in contrast to algorithmic self-governance, which provides algorithmic decision-making support for self-determination, i.e., to enable
those who are affected by a set of rules to participate in their selection,
modification, enactment, application, and enforcement.

IEEE TECHNOLOGY AND SOCIETY MAGAZINE

∕

DECEMBER 2019
https://en.wikipedia.org/wiki/Diogenes https://www.kinexon-sports.com
IEEE Technology and Society Magazine - December 2019

Table of Contents for the Digital Edition of IEEE Technology and Society Magazine - December 2019
