Ways of Communicating A communication agent model
António Fonseca - NESS Conference
Summary ● ● ● ● ● ● ● ●
Collectives Collective Memory Agent model Parameters Parameterizing with Twitter Experiment Results Conclusions 01/27 António Fonseca - NESS Conference
Collectives Collective Consciousness - "The set of shared beliefs, ideas and moral attitudes which operate as a unifying force within society." (Durkheim, 1983)
Collective Identity - "The shared sense of belonging to a group." Collective Intelligence - "Shared or group intelligence that emerges from the collaboration and competition of many individuals" (Swarm Intelligence, Crowdsourcing, Mass Peer Review, Voting Systems, Consensus Decision Making, Stigmergy, ...)
Collective Memory - "The shared pool of information held in the memories of two or more members of a group."
Meme - "An idea, behavior, or style that spreads from person to person within a culture." (Wikipedia) 02/27 António Fonseca - NESS Conference
Collective Memory Influence
Collective Memory Individual Memory
Production 03/27 António Fonseca - NESS Conference
Collective Memory Memorialization - Collective memory is sustained through a continuous production of representational forms. (Halbwachs, 1950) - Memorials, Icons, Memorabilia.
Technology - Collective memory today differs much from the collective memories of an oral culture, where no printing technique or transportation contributed to the production of imagined communities where we come to share a sense of heritage and commonality.
Mass Media - Images, film scenes, news scenes, photographs, quotes, and songs, became very familiar to regular citizens and remained in their collective memory. Radio, TV and newspapers build a new kind of collective memory where various news events could be shown much quicker to wider audiences. As scenes, events and objects have become part of audiences' collective memory, references to these became also part of the common language of everyday person, knowing that a large audience will recognise and understand them without further explanation.
What about the Internet? 04/27 António Fonseca - NESS Conference
Collective Memory What about the Internet? ●
Individual’s memory only functions within a collective boundary and is formed through the sharing of different perspectives between people by social interactions. (Halbwachs, 1992)
●
Places like Jerusalem or a museum enforce memories in a collective body with space as the medium. This view can be applied to collective memory that is created from expressions of reminiscing memories, the place being the blogosphere, social networks, the web in general.
●
Factors of collective memory production on the Internet (Young Sik Kim et al., 2009): 1. 2. 3. 4. 5.
Content: The content expresses a form of personal experience in the past or recollection of memories. Openness: The memory is open and shared to the public. Trigger: A person reminisces past memories or remembers past memories when given a trigger, a stimulus. Sympathy: Rather than a memory that is extremely personal, it should be a memory that the majority of people can sympathize on. Interaction: A collective memory is formed only after there are interactions between people. 05/27 António Fonseca - NESS Conference
Agent Model Assumptions of the model : ●
Agents talk like: 1. 2. 3. 4.
“What?” “My name is John.” “The area of the circle is given by pi times its squared radius.” “350 5th Ave, New York, NY 10118, USA”
●
Sentence 1-3 in the above list represent expressions that are product of an agent (or another peer agent) mind, while sentence 4 expresses an entity in the world, what linguists call named entity or a proper name. More generally we are considering that any propositional form has its origin in the mind of agents, but any reference to the outside world has its origins on the environment.
●
We consider three distinct sources of meaning: 1. 2. 3.
The agent’s mind. Another agent’s mind. The World.
06/27 António Fonseca - NESS Conference
Agent Model Further assumptions of the model : ●
We consider that agents exchange information units and that each information unit represents an atomic repository of meaning that we call a token. Each agent processes tokens and thus produces, propagates and diffuses meaning within the community.
●
Each token we will encode the exact spatial, semantic and temporal location of an expression made by the agent. If we attribute an integer to each token, as there are infinite number of integers, we may consider integer codification to reasonably represent of the actual meaning event that was exchanged at certain point in time between agents. This meaning also includes a pragmatic content, although we are not worried about its actual form or its intended consequences.
●
This approach can be interpreted in the context of information representation in Situation Semantics Theory (Devlin, 1991, Barwise and Perry 1983) where agents exchange Infons: s |= > in situation s, the objects a1, . . . , an stand in the relation R 07/27 António Fonseca - NESS Conference
Agent Model Further assumptions of the model : Our agent’s model doesn't have a very sophisticated theory of the world. Specifically we're not taking into account more secondary effects in information flow as for example the Xerox Principle (Dretske, 1981): “If A carries the information that B, and B carries the information that C, then A carries the information that C” If for example an agent says that “Joe is at home in his office” she is not partially meaning that “Joe is at home”, but the statement “Joe is at home” carries the fact that Joe is either at his home or at his office. We are not concerned with ground truth values, only in meaning as conveyed by tokens, and also in the part of this meaning that is external and did not originate in the agent’s mind. The important distinctive information for the agent is its origin: 1. 2. 3.
The individual (its mind) The other individuals (communication) The environment (the world, culture, mass media, ...)
08/27 António Fonseca - NESS Conference
Agent Model
09/27 António Fonseca - NESS Conference
Parameters 1. 2. 3.
θm - degree of expression about self θi - degree of expression about others θe - degree of expression about the outside world
6.
αo - degree of self awareness αi - degree of awareness of others αe - degree of awareness of the outside world
7.
β
4. 5.
- degree of general forgetting
10/27 António Fonseca - NESS Conference
Parameters In terms of Shannon's Communication Theory the mutual information between interfaces can be made directly proportional to the α , β and θ parameters:
Mutual Information between interfaces
It suffices to drop a fraction of k tokens to reduce by k
log2(k) the mutual information.
11/27
António Fonseca - NESS Conference
Parameterizing with Twitter In order to parametrize our model we devised an experiment taking advantage of the communication properties of the online microblogging network Twitter. This way we apriori reduced the number of free variables. Between 2013-02-25 to 2013-04-31 we've collected a total amount of 4.7 million tweets from a network of the most active 5000 portuguese users. After applying statistics to the stream of tweets we've estimated the following values for a general individual:
θm = 0.5789932 ~ 58%
(degree of expression about self)
θe = 0.2382742
~ 24%
(degree of expression about the outside world)
θi = 0.1827326
~ 18%
(degree of expression about others)
12/27 António Fonseca - NESS Conference
Parametrizing with Twitter 27 modes of communicating with followers
= presence of an URL or hashtag (#)
13/27 António Fonseca - NESS Conference
Parametrizing with Twitter
14/27 António Fonseca - NESS Conference
Parameterizing with Twitter ● ● ●
People like to talk to the public when they're talking public matters. People like talking private matters when they're talking just to another. People talk to one they also like talking to some but not many.
The statistics for expression broadcasting were the following: to_all 0.6926911 ~ 70% to_one 0.2615209 ~ 25% to_some 0.04578795 ~ 5% The high value of to_all should be due to the context of a social network.
15/27 António Fonseca - NESS Conference
Experiment Our experiment was implemented with statistical averages over the results after 100 iterations, repeated 10 times of 500 agents. The key parameter was progressively varied between 0 and 1, the other parameters were given a normal distribution ~ N(0.5, 0.25). Each agent communicates with the community agents, with to_one, to_some, to_all probability of transmitting a token to another agent; to some others or to all the agents. There are two distinct types of tokens : 1.
2.
500 different Inside tokens representing at an initial point in time the individual expression of each of the 500 agents Outside tokens that represent the further expression of every agent during the simulation. 16/27 António Fonseca - NESS Conference
Results Forgetting
High levels of forgetting tend to reduce memory size but do not affect much expressions
17/27 António Fonseca - NESS Conference
Results Degree of self awareness
High self awareness increases slightly memory size
18/27 António Fonseca - NESS Conference
Results Degree of awareness of others
Awareness of others increases individual expressions but also memory size.
19/27 António Fonseca - NESS Conference
Results Degree of awareness of the outside world
Low levels of awareness of the outside world increase individual expression but also lowers memory.
20/27 António Fonseca - NESS Conference
Results Degree of expression about self
Expression about self does not change much memory but is essential to propagate individual memory
21/27 António Fonseca - NESS Conference
Results Degree of expression about others
Expression about others does not change individual memory.
22/27 António Fonseca - NESS Conference
Results Degree of expression about the outside world
High levels of expressions about the outside world are contrary to individual expression.
23/27 António Fonseca - NESS Conference
Results Forgetting and Awareness and Entropy Density High levels of forgetting favor low memory but also high information density.
High levels of awareness are the exact opposite of forgetting favor high memory but also low information density.
24/27 António Fonseca - NESS Conference
Results Expression of self and others and of the outside world versus entropy density
Low levels of outside world expression (privacy) favor low memory but also high information density.
25/27 António Fonseca - NESS Conference
Conclusions ● ● ●
A new model of communication agency. Interpreting collective memory as a communication problem. The role of individual memory.
Work to follow: ● ●
Validating the model against more realistic situations Analysing temporal aspects of communication.
26/27 António Fonseca - NESS Conference
Thank You!
The Observatorium - ISCTE-IUL Lisbon University Institute http://theobservatorium.eu
[email protected],
[email protected] 27/27 António Fonseca - NESS Conference
