The experiments presented in this document are taken from:
Article: D. Rudrauf, G. Sergeant-Perthuis, O. Belli, Y. Tisserand, and G. Di Marzo Serugendo. Mod- eling the subjective perspective of consciousness and its role in the control of behaviours. arXiv:2012.12963, 2021.
Agents are embeded with a conscious like global workspace that allows them to take a subjective perspectives on their environment; it also allows them to imagine in perspective of other agents which can influence their actions depending on their degree of Theory of Mind.
In this situation there are two cubes c1 and c2 and two agents S and O. Agent 1 is the test subject (S), it likes the object c1 but dislikes c2. However c2 blocks its access to c1. Depending on it's capacity to predict what will happen in the future it can overcome the obstacle c2 or not.
Agent 1 (S) likes object c2 and likes agent 2 (O) but does not like object c1 that much. If agent 1 is capable of social-affective perspective taking then it updates its preference with respect to what it infers on the preferences of the other agent and will start to like object c2. No social-affective perspective taking can be seen as a simulation of Autistic Spectrum Disorder (ASD).
Agent 1 (S) likes cube c1 and Agent 2 (O) likes cube c2, and in order to reach c1, S needs to bypass c2. If S takes into consideration the preferences of O in its decisions, it will try to avoird O. When S is socially anxious it believes that O is constantly focused on him.
Scenario without negative priors.
S believes O has negative priors about him (which is in fact false).
S believes O has negative priors about him (which is in fact false) and believes/ imagines that the second agent is constantly focused on him (behaviour associated to SAD). It is implemented by having S imagine, when taking the perspective of O, that it is always looking at it.
This situation is an adaptation, in our context, of the Sally and Anne experiment. Both agents like the same cube, cube 2 and they are competing to get near the cube. Both agents can perform social-affective perspective tacking and assume the other agent has positive priors for cube 2, however agent 1 (S) believes agent 2 (C) does not like him and it is in fact the case. The experiment is divided into two phases. The first phase is a test round where the agents act according to the initial setting described just before; it is to show how they go through with their actions. The objects are then switched and the agents are replaced far from the objects for the second round to start.
In the first scenario both agents can see where the objects are switched too. Both are aware that the objects were switched.
In the second scenario, only agent 1 (S) knows that the objects have been switched while agent 2 believes they have not been. Agent 1 exploits false beliefs of Agent 2.