How a lie can become the truth

Arnoud van der Meulen s1901974
Ivo Bril s2001179
Ben Wolf s1879227

Introduction

  One thing everyone learns when they grow older is that a lot of things are uncertain. As we students learn more and more about research and develop ourselves as smart individuals, we learn to see that there can be more sides of the box than the side we can see. Sometimes we rely on others to tell us how one of these sides looks like. This relies on us trusting the information someone delivers. However, we do not trust everyone we meet, not even people we see on a more regular basis. So what do we do when we are confronted with two conflicting stories? When two persons tell us a different story about another side of the box? In this project we have tried to formalize the balance between having beliefs and discarding them when one is presented with new, conflicting information.


Project explanation

  The outline of this project stems from a prank that was pulled on a fellow student. He is quick to jump in on juicy details about someone's personal life, so the joke was about some personal matter of another fellow student. The gossip student thought he was in on a personal matter that was made up. To enjoy the joke, the initial group of pranksters decided to tell the secret about this prank to people they trusted a lot. The catch here was that more people knowing about the truth of this prank increases the likelihood of the gossiper finding out the truth. The gossiper can find out about it because the people that the pranksters trust are also good friends with the gossiper.
This project tries to emulate this joke, but makes it more interesting by making the gossiper spread the lie around as if it were the truth. So on one hand, there are the pranksters telling the truth about the prank to people, and on the other hand there is the gossiper who thinks the prank is real and tells his version of the story to everyone he knows and trusts. While trying to formalize lying we got inspiration from Van Ditmarsch (2014), who speaks of different agents. In this project we have tried to simulate sceptical agents as they do not accept all beliefs blindly. More about how the rules framework tries to bring forth this sceptical nature will be explained in the next section.


Approach and Design

  The program language used for this project is Java. Although it does not provide any logical reasoning methods, it is a language that everyone felt comfortable with. It also allowed us to easily create an agent class and start building a framework for the agents to communicate and tell each other their beliefs.

Rules
Communication
The rules used for how agents communicate with one another are as follows:

Beliefs
The belief phi is represented by a conviction value, along with a validation to address the fact if they believe the prank or not. The conviction of the belief ranges from zero to ten. The validation can be either 'true', which means the agent believes the prank to be true, or 'false', which means the agent knows the truth about the prank. The rules used for the beliefs of agents are as follows: The conviction of a belief is (re)calculated as follows:
Using these rules, the system can show how differing convictions about one belief influence one another. In the next section will be discussed how the initial conviction values of the prankster and the gossiper influence the results, as they determine how strongly one side of the story is brought over to the, initially, beliefless agents.


Results

 To produce results we performed a parameter sweep using the initial values of conviction of both the prankster and the gossiper (going from one to ten). All possible combination were ran three times as the random-factor of our system demands more runs to say something about convergence to a balanced state. The amount of socialization rounds was 40. This number is chosen as it represents a decent amount of socializations, while still staying true to not dragging out the social process too long.
The results of the parameter sweep showed that the pranksters side of the story prevailed most often. The gossipers side of the story only remained present within the group if the conviction of the prankster was low (ranging from 1 to 3). Interestingly enough, this result remains constant over all conviction values of the gossiper.
The reason that every run with the prankster having a value of more than 3 converges to everyone knowing the truth about the prank is due to the fact that the prankster never changes his conviction. This leads to a source that cannot be convinced of the opposite and keeps saying things with the same conviction value over and over again. Due to this, a very convincing prankster will always have everyone believing his side of the story after a certain amount of steps, making the truth about the prank common knowledge within the group.
The final implementation also allows for the analysis of multiple facts, but this was not researched for the results.


Conclusions

 In this course we have learned how to reason about beliefs and knowledge. The project tries to show how a heavily debated part of this field, namely lies, can be simulated and how there can be a balance between the truth and a lie. It also shows that there are complex dynamics in this balance, and that this balance is very fickle. There is also a lot to conclude about socialization: the process can become as complex as possible. During the project, we have noticed that there is many potential for further implementation, such as reciprocatary communication (not merely telling facts, but also adjusting ones beliefs if the social partner does not agree with the fact teller) or adjusting the trust in a person based on multiple wrong gossips.


References

  1. Van Ditmarsch, H. (2014). Dynamics of Lying. Synthese, 191(5), 745-777.