Does Mind Matter? Reputation for Governance

1
Does Mind Matter? Reputation for Governance

• Rosaria Conte
• LABSS/ISTC-CNR
• http//labss.istc.cnr,it
• Lorentz Center, Leiden
• 12-16 January 2009

2
Outline

• Aspects and levels of cognition
• The role of reputation in governance
• Modelling Supraindividual Entity Reputation (SER)
• Conclusions

3
Aspects and levels of cognition
4
Mental states

• Cognitive systems have mental states
• Belief
• Goal
• Emotion
• To act on the world they must accomplish also
  mental operations, by manipulating their and
  others mental states
• Memorizing
• Learning
• Reasoning
• Planning
• Taking decisions
• Imitating
• Etc.

5
Doesnt mean they are conscious
6
Mental manipulation

• Includes
• Confrontation qualitative differences may be
  observed among mental states p and q
• Reasoning (assert/denycounterfactual reasoning)
• Nesting representations of representations
  (meta-cognition)

7
Nesting

• Given n types of mental states, f.i.
• Beliefs
• Goals
• Emotions
• All sorts of combinations are possible, whoever
  the holder of the mental states is

• B
• Be Ego believes that
• Bo Other believes that
• G
• Ge
• Go
• E
• Ee
• Eo

8
Different from ToM

• Meta-cognition is broader than Theory of Mind,
• Not only social beliefs, but also
• Social goals
• Social emotions
• Mental states about own mental states

9
One level nesting

• Mental states belong to the same agent (Ego)
• To others (Other)

10
Social mental states
Bo Go Eo
Be I believe you are an atheist I know what she likes I understand your feeling
Ge I make him believe I got her to want I want you to feel ashamed!
Ee I hate your creed! I fear his intention I like her shyness
11
Self-representation and manipulation
Be Ge Ee
Be I believe I know I dont know what I want, but I do know what I dont! I dont think I am in love
Ge I desperately want to believe in God! Ill never do it again. I promise. I decided Ill stop suffering
Ee Im happy to learn that I wish I didnt want to I cherish my longing for him
12
Multiple levels of nesting

• Too many combinations
• Interesting with mixed holders
• You want me to believe that you are in love with
  me, but
• I want you not to realize how I really feel about
  you

13
Where to stop ?
14
Depends on what you are accounting for

• Essential for communication
• Suppose we had only one level nesting it would
  not account for the difference between
• Manipulation
• Communication

I Want
Want
15
Why bother?
MC
16
Some properties of nesting

• Truthvalue is not extensible from one level to
  another
• In BeBe,BeBo, BoBe, BoBo
• Both may be true or false
• But y may have different rutvalues
• nested belief may be true, but not the
  metabelief and
• Viceversa
• In particular,
• Meta-cognition does not inherit the turthvalue of
  cognition
• No commitment of the meta-beliefs holder on
  truthvalue of nested beliefs.

17
Evaluation

• Evaluation (from Conte and Paolucci, 2002)
• a belief about the power of a given entity wrt
  someones goal(s)
• This tool is good to chop wood
• Entity ei (tool) has the power to achieve agent
  ais (the speakers) goa to have wood chopped.
• Truncated evaluationThis is a good ax.
• Good for what?
• The entity is evaluated against the goal/function
  it has been made for,
• which it incorporates

18
Social evaluation Image

• People are often evaluated against others goals,
  
• which the target of evaluation may adopt
• Lola is a good company for shopping
• Or not
• Carola is a good partner for gossip tell her a
  news, and soon it will spread throughout the
  office!
• Image can be truncated
• Walter is a good chap

19
Images characters

• Three characters
• Target (T)
• Evaluator (E)
• Beneficiary (B)
• Conceptually distinct,
• Emprically may overllap

20
Self-image E T

• T E B
• need not be realistic -)

21
Whos B?

• B E
• T is evaluated against a goal of Es
• evaluation is self-interested (and prudent
• T ? E B candidates selection for a job
• When T is evaluated against one of Ts goals (T
  B), evaluation is tutorial
• T B ? E educational evaluation
• With no overlapping, evaluation is neutral
• T ? E ? B standard peer review in science

22
Reputation meta-evaluation

• Belief about how T is evaluated by E (often
  indefinite or implicit)
• 4 characters
• T
• E
• B
• G (gossiper)
• The former 3 may overlap, but not all!
• G will always (pretend to) report on others
  evaluation
• G does not necessarily share the evaluation
• G does not take responsibility over its
  consequences

23
Reputation and Image

• Complex interplay
• G may report on the reputation of x
• G may report on his image of x
• G may report on both
• These may coincide I have the worst possible
  opinion about that guy, and I know him to be
  ill-reputed
• Or not
• I dont care what people say I think Diego is a
  sweetie
• Do you know what they say about Fatimas last
  date? I mean I think hes great, buti found out
  hes known to be a rogue!

24
Why bother?

• Exchange and cooperation
• Partner selection
• Identify and isolate cheaters
• Norm-enforcement
• Group formation and maintenance

25
Image and Reputation A Comparison

• What are the respective effects?
• Hypotheses
• In partner selection and norm-enforcement
• Image insufficient, since social knowledge is
  acquired only via direct experience
• But what about image exchange? In NormSim
  (Castelfranchi et al., 1998), it was shown to
  allow norm-compliant to compete with cheaters in
  the same population
• What is the difference between image exchange and
  reputation transmission?
• In social and cultural evolution reputation
  allowed the enlargement of human settlements
  (Dunbar, 1998)

26
The role of reputation in governance
27
The system Repage

• The main element of the Repage architecture is
  the
• memory, which is composed by
• a set of predicates. containing either
• a social evaluation, either image, reputation,
  shared voice, shared evaluation or
• valued information,
• evaluation related from informers, and
• outcomes.
• contain a tuple of five numbers representing the
  evaluation plus a strength value that indicates
  the confidence the agent has on this evaluation.
• conceptually organized in a network of
  dependencies, specifying which predicates
  contribute to the values of others
• each predicate has a set of antecedents and a set
  of consequents.
• If an antecedent is created, removed, or its
  value changes, the predicate is notified,
  recalculates its value and notifies the change to
  its consequents.
• REPAGE runs on a JADE-X platform.

28
The market scenario

• Simulations were run (Paolucci et al., 2007
  Quattrociocchi et al., 2008) with fixed number of
  sellers and buyers (respectively to 100 and 15).
• Goods are represented by a 1-100 valued utility
  factor (interpreted as quality, but, at this
  level of abstraction, could as well represent
  other utility factors as quantity, discount,
  timeliness).
• Results were explored with an instrument
  (Dimensional Fact Modelling) to extract from
  simulation findings factors, also in their
  combinations and interactions with others,
  relevant for our reputation theory.

29
Main findings
30
Further directions

• Governance
• Opinion manipulation and political choice
• Reputation in industrial clusters
• Institutional reputation

Beyond the individual
31
Monitoring institutions The case of last
Italian election(from Quattrociocchi et al., in
preparation)
32
Media informational cheating
33
Social perception
34
What has happened?

• Quality collapses people buy (vote for) security
  champions, which they dont need (i.e., lemmons)
• At what price? Probably giving up
• Welfare state, especially RD investment
• Conflict-of-interest legislation
• Pursuit of bribery.
• Hence
• high information cheating through the media
• feeds gossip
• In turn reinforcing information cheating

35
A puzzle?

• Social perception follows the trend of lies
• People buy lemmons, assuming that it is gold.
• ???

36
Questions for policy makers

• How reduce media informational cheating on social
  perception?
• How reduce the effects of media on peoples
  private communication?

37
Question for the social scientist

• When and why informational cheating produces a
  self-fulling prophecy?

38
(e-)Governance

• The role of reputation networks in industrial
  clusters (see SOCRATE, http//socrate.istc.cnr.it/
  socrate )
• Reputation technology on the Internet (see eREP,
  http//megatron.iiia.csic.es/eRep/?qtracker )
• Evaluation by results reputation of institutions
  and public administration

39
Reputation on the Internet main results

• (From eREP first deliverable, at
  http//megatron.iiia.csic.es/eRep/?qnode/37 )
• Image and reputation imply different degree of
• Commitment
• Repsonsibility on consequences of communication
• Hence, conditions of communication may force one
  or the other with different outputs
• When
• no reputation report is allowed, and one or more
  of the following conditions holds
• Feedback is target accessible
• Choice of recipinet is unfeasible
• Feedback is not anonymous
• Underprovision and overrating are likely
  (courtesy equilibrium)
• In the opposite conditions, provision and
  underrating can be expected (prudence
  equilibrium)

40
Modelling Supraindividual Entity Reputation (SER)
41
Superindividual entity reputation (SER)

• How
• model
• Implement
• Quantify it
• Why bother

42
Modelling SER

• (From Conte and Zaccaria, 2009, forthcoming)
• Internal (the entitys)
• Global (before its members)
• Distributed (of its members)
• External (before external agencies)
• global (of entity on the market, before users or
  clients etc.)
• distributed (reputation of members in the
  external world, public employees, FIAT workers,
  parliamentary, etc.).

43
Some suggestions

• Take into account
• difference between I and R allow for choosing
  between
• direct and
• reported on feedback,
• With and without explicit source

44
Feedback provision the role of structural
modalities

• Activate
• Users networks for exchanging info about
  personal experiences
• Accessible Vs protected feedback
• Broadcast Vs narrowcast feedback,
• With and without choice of recipient,
• Target-accessible only to target
• User-accessibile soltanto a un subset di utenti
• Anonymous Vs signed feedback
• etc.

45
Internal and external SER

• What do members believe and report on compared to
  users?
• Multiple external and internal sources
• from other istitutions, users, pribate or public
  competitors.
• feedback from social network (accompanying
  person, family, etc. but how aggregate them?).
• How do they interact?

46
Global and distributed SER

• Is global SER a sum of members reputation?
• How do they interact?
• Members inherit SER, but also
• Affect it
• To what extent?
• Are there critical thresholds?

47
The circuit of external reputation

• If SE are funded on the grounds of their global
  reputation
• External global reputation affects external
  distributed reputation
• External distributed reputation of members
  affects external individual reputation
• Members manage SER external reputation to manage
  their own reputation
• SER manages (selects) members to manage external
  reputation.

48
Conclusions

• Social evaluation is crucial for governance
• But evaluation and meta-evaluation have different
  impact
• They have different advantages and disadvantages
  (see REPAGE results)
• Which must be understood
• Conditions favouring either are to be found out
• Impact of either need to be assessed
• But to understand them we need to model the
  properties of complex mental states