State-of-the-Art: Design of the Semantics Acquisition Games - Semantic Acquisition Games: Harnessing Manpower for Creating Semantics - page 103

7.2.6 Cheating and Dishonest Player Behavior
In all computer games, including semantics acquisition games, cheating and dishon-
est player behavior is a phenomenon that must be considered. Naturally, it is mostly
present in the competitive games (in particular with social context), while games
based on player self-challenge suffer only a little from this phenomenon. Generally,
cheating includes player effort to exploit game rules, various “holes” and bugs or
directly interfere with a game's implementation in order to acquire a higher score
or some other advantage in a game. This has negative impact on the perception of
fairness by other players and may discourage them from playing the game. In case
of SAGs, it may also damage the problem solving capability of the game.
Semantics acquisition games usually deal with the threat of dishonest player
behavior by pursuing restrictive rules to discourage players from cheating (e.g.
preventive banning of certain tags from use [ 20 ]) or they have control mechanisms
based on mutual player supervision [ 9 , 28 ]. An example is the ESP game where
dishonest player behavior is a considerable issue. The possible situation of such case
is when players somehow agree on certain words (e.g. via some obscure portal),
which they enter as tags for each image. The ESP Game solves this issue by various
heuristics (e.g. tracking the recurrence of certain words, identifying typical “cheat-
ing” behavior patterns) [ 28 ]. In KissKissBan game, Ho et al. added to this by mutual
player supervision introduced by new player—opponent to the two collaborating
players.
Some SAGs even analyze the player's behavior on-the-fly: the SAG Akinator
(popular person guessing and information harvester) is able to detect random ques-
tion answering of the player (and discourage them from such further behavior by
taunting the player). The Seneviratne's framework combines the machine learn-
ing for dishonest behavior with checking the artifacts against a validation data
set [ 17 ].
In our research, we introduce an a posteriori anti-cheating heuristics for SAGs,
based on the notion that for solving human intelligence tasks, it is unlikely that an
automated cheating method would produce them in sufficient quality (if so, there
would be no need for the game). Therefore, player actions earning highest scores in
the game can always be examined whether they lead to the creation of useful artifacts.
If not, they may be subject to disqualification.
To sum up, SAGs implement the following anti-cheating strategies (including
combinations):
1. Prevention by restrictive rules.
2. Mutual player supervision.
3. Anomalous behavior pattern detection (machine learning, validation data use).
4. A posteriori cheating detection.