advertisement

siena06

50 %
50 %
advertisement
Information about siena06
Entertainment

Published on February 4, 2008

Author: Camilla

Source: authorstream.com

advertisement

Slide1:  Seminario del Dipartimento di Filosofia e Scienze Sociali - Università di Siena, 7 Marzo 2006 Natura e dinamiche della credenza Breve confronto fra epistemologia filosofica, modelli formali e teorie cognitive Fabio Paglieri Dottorato in Scienze Cognitive, Università di Siena Obiettivi & piano dei lavori:  Obiettivi & piano dei lavori Obiettivo: valutare possibili intersezioni (utili) fra filosofia, psicologia e modelli formali per studiare la nozione di credenza e le sue dinamiche Piano dei lavori: breve chiarimento iniziale su quale epistemologia possa interessare la psicologia discussione di “casi”, alcuni positivi, altri negativi: pars destruens: critica cognitiva del modello AGM di revisione delle credenze pars construens: analisi game-teoretica dell’influenza motivazionale sui processi di formazione delle credenze Quale epistemologia per la psicologia cognitiva?:  Quale epistemologia per la psicologia cognitiva? La psicologia cognitiva ha come obiettivo la spiegazione del comportamento, e usa categorie mentali come strumenti esplicativi Modelli teorici e dati empirici (tensione normativo vs. descrittivo) Quindi PC si interessa alla credenza, non alla conoscenza, perché gli agenti cognitivi reali agiscono sulla base della doxa, non dell’episteme Ramsey: credenze come mappe che usiamo per navigare nel mondo Credenza legata all’azione e all’intenzione Esempi: accettazione vs. credenza, logiche BDI Conseguenza: interesse molto limitato da parte della PC per problemi che esulano da questa “rilevanza comportamentale” della credenza Esempio: rapporti fra credenza e conoscenza in una prospettiva realista, e conseguente problema di Gettier Due problemi epistemologici molto studiati e di comune interesse: la revisione delle credenze l’influenza delle motivazioni sulla genesi delle credenze Slide4:  Per una critica cognitiva al modello AGM di revisione delle credenze Fabio Paglieri Dottorato in Scienze Cognitive, Università di Siena Presentato a: ILLC Logic Tea – Amsterdam, 26 Febbraio 2004 Secondo Convegno AISC – Ivrea TO, 19-20 Marzo 2004 La revisione delle credenze:  La revisione delle credenze La revisione delle credenze si occupa del modo in cui un agente cognitivo (umano o artificiale) cambia le proprie idee Più precisamente, ci si focalizza sulle credenze, cioè quelle rappresentazioni interne che l’agente considera sufficientemente attendibili come basi del ragionamento, della pianificazione e dell’azione: la revisione delle credenze ne studia i processi di cambiamento e trasformazione L’etichetta “belief revision” indica una serie di teorie formali e modelli computazionali del processo di revisione delle credenze Perché è interessante un approccio formale?:  Perché è interessante un approccio formale? Il tentativo di dare una formalizzazione simbolica o matematica ai modelli cognitivi di revisione delle credenze spinge alla chiarezza concettuale ed evidenzia le conseguenze (possibilmente inaspettate o indesiderate) delle proprie assunzioni I modelli formali insistono sulla dimensione normativa dei fenomeni cognitivi (es. razionalità), aiutando a superare e integrare approcci meramente descrittivi La formalizzazione costituisce un passo importante verso l’implementazione in sistemi computazionali e lo studio della revisione delle credenze in simulazioni ad agenti Belief revision in pillole:  Belief revision in pillole Studi filosofici delle dinamiche doxastiche (Peirce, Ramsey, Quine, Jeffrey, Rescher, Cherniak, Harman, Levi) Teoria AGM (Alchourrón, Gärdenfors e Makinson): manipolazione di insiemi di proposizioni, indipendente dal linguaggio in cui tali proposizioni sono espresse Critiche, revisioni e alternative al modello AGM: modelli fondazionalisti (Doyle), basi di credenza (Nebel), revisione con risorse limitate (Wassermann), revisioni reiterate (Boutilier, Darwiche e Pearl), revisioni con priorità non vincolata (Hansson et al.), gradi di credenza (di solito con modelli probabilistici: Fagin, Halpern, Friedman, Boutilier), belief revision vs. belief update (Katsuno e Mendelzon) Integrazione della belief revision con altre teorie e ambiti rilevanti: in logica: Segerberg, Leitgeb, van Ditmarsch, Gillies e Pollock in IA: Galliers, Alechina e Logan, Dragoni e Giorgini, Fullam e Barber nelle scienze cognitive: Castelfranchi, Paglieri AGM: general framework:  AGM: general framework The AGM theory is an idealistic formal model… of rational belief change… which aims to apply to common sense daily reasoning AGM: example:  AGM: example Oscar believed that he had given to Victoria a gold ring at their wedding… he bought two wedding rings in Casablanca and was told they were gold… he checked with the jeweler next door who confirmed they were gold. Back home Oscar dropped acid on the ring by mistake, and the ring was stained – so it could not be gold!… Oscar tested Victoria’s ring, which also turned to be a fake. (Gärdenfors, 1988) Here Oscar, presented with a new belief “our rings are stained by acid”, is forced to drop his previous belief “our rings are gold”. Intuitively, we consider this revision rational: why is it so? AGM: basic features:  AGM: basic features Belief states are modeled as sets of propositions, which are assumed to be consistent and deductively closed (therefore infinite) Given a belief state B, an agent can express three epistemic attitudes towards a proposition p: accepted (p  B), rejected (~p  B), or indetermined (p  B and ~p  B) Three basic operations of belief change: expansion (B + p, i.e. p is added to the set B), contraction (B – p, i.e. p is deleted from the set B), and revision (B * p, i.e. p is added to the set B and ~p is deleted) Rationality is expressed by axioms binding the operations of belief change AGM axioms aim to ensure consistency-preservation and minimal change in the revision process (Harman, 1986) AGM: postulates for expansion:  AGM: postulates for expansion B+1. B+p is a belief state B+2. p  B+p B+3. B  B+p B+4. If p  B, then B+p = B B+5. If B  H, then B+p  H+p B+6. For all belief sets B and all propositions p, B+p is the smallest belief set which satisfies B+1-B+5 AGM: postulates for revision (incomplete):  AGM: postulates for revision (incomplete) B*1. B*p is a belief state [closure] B*2. p  B*p [success] B*3. B*p  B+p [inclusion] B*4. If ~p  B, then B+p  B*p [preservation] B*5. B*p = L iff ~p [consistency] B*6. If p  q, then B*p = B*q [equivalence] Here I skip axioms B*7 and B*8, which deal with revision by conjunctions AGM: multiple outcomes of rational revision:  AGM: multiple outcomes of rational revision Rational revision (and contraction) can have multiple outcomes, i.e. there can be more than one resulting belief states which satisfies the AGM postulates {p, ~q} {p, p  q, q} * ~q ? {p  q, ~q} AGM: examples of multiple outcomes:  AGM: examples of multiple outcomes Revision: {p, p  q, q} * ~q Possible rational outcomes: {p, ~q}, {p  q, ~q} Ex. 1: Cold sun p : The weather is sunny q : The temperature is nice p  q : When the weather is sunny, the temperature is nice ~q : The temperature is not nice at all! Ex. 2: Meeting JFK p : JFK was killed in Dallas q : JFK is dead p  q : If JFK was killed, then he is positively dead ~q : JFK is not dead at all! AGM: epistemic entrenchment I:  AGM: epistemic entrenchment I Epistemic entrenchment is an ordering  over propositions in L, which in turn determines an ordering over sets, i.e. over equally rational outcomes of revision and contraction Every belief state is characterized by an epistemic entrenchment, independent by what happens in revisions and contractions When a belief state B is contracted (or revised), the sentences in B that are given up are those with the lowest epistemic entrenchment AGM: epistemic entrenchment II:  AGM: epistemic entrenchment II Examples: Cold sun EE: T  ~q  …  p  p  q Outcomes ordering: {p, ~q}  {p  q, ~q} Meeting JFK EE: T  ~q  …  p  q  p Outcomes ordering: {p  q, ~q}  {p, ~q} Rationality can now be expressed by postulates binding EE, which in turn ensure the fulfillment of rational postulates for revision and contraction (see representation results in Gärdenfors, 1988) Criticisms and refinements of AGM:  Criticisms and refinements of AGM finite belief bases vs. infinite belief states (e.g. Nebel, 1989) resource-bounded belief revision (e.g. Wassermann, 2000) iterated belief revision (e.g. Boutilier, 1996; Darwiche & Pearl, 1997) non-prioritized belief revision (e.g. Boutilier et al., 1998; Hansson et al., 2001) degree of belief (usually with probabilities: e.g. Fagin, Halpern, etc.) belief revision vs. belief update (Katsuno & Mendelzon, 1992) A cognitive critique of AGM:  A cognitive critique of AGM no distinction between data and beliefs (cf. indeterminacy for both non-assessed data and unknown data) no account of the reasons why we believe something, vs. foundation theories (Doyle, 1992) an idealistic approach to deal with concrete cases: it defines rationality for BR in cognitive agents using a framework which does not apply easily to cognitive agents lack of integration with (1) other logical approaches to belief dynamics (Segerberg, 1999; Pollock & Gillies, 2000; Leitgeb,Segerberg, in press), and (2) other related cognitive tasks (information update, argumentation, hypothetical reasoning; cf. Veltman, 1997; van Ditmarsch, 2003) Una possibile alternativa: punti rilevanti:  Una possibile alternativa: punti rilevanti Natura procedurale della formazione e revisione delle credenze: l’informazione passa attraverso diverse fasi nel processamento epistemico, acquistando di volta in volta nuove proprietà funzionali – né il processo è “a senso unico”, esistono meccanismi di feedback Importanza della distinzione fra dati e credenze e suo impatto sui modelli formali delle dinamiche epistemiche Analisi dettagliata delle ragioni del credere e loro integrazione nella revisione delle credenze Data-oriented Belief Revision (DBR): Basics:  Data-oriented Belief Revision (DBR): Basics distinction between data (available stored information) and beliefs (information accepted as reliable for reasoning and action); cf. Rescher, 1976; Castelfranchi, 1997; Tamminga, 2001; Floridi, in press two steps process: data revision + belief selection = BR external and internal belief revision principles (universal and qualitative) and parameters (individual and quantitative): cf. Chomsky, parameters as a way of expressing individual variation in BR Il modello AGM: semplice, forse troppo:  Il modello AGM: semplice, forse troppo Nota: tutto avviene al solo livello delle credenze, non c’è alcuna menzione dei dati Stato epistemico iniziale Stato epistemico finale Operazioni di revisione Nuova credenza Architettura DBR per la formazione di credenze:  Architettura DBR per la formazione di credenze Info update Data mapping Recupero Selezione Inferenze Focaliz-zazione Oblio DATI IN MEMORIA DATI ATTIVI CREDENZE ATTIVE CREDENZE DATI CREDENZE IN MEMORIA Archiviaz. Le ragioni del credere:  Le ragioni del credere CREDIBILITÀ: esprime il numero e la qualità dei dati che supportano un’informazione (evidenze esterne e interne), confrontati con i dati che la contrastano (contro-evidenze esterne e interne) IMPORTANZA: esprime il potere esplicativo e la salienza epistemica di un dato, cioè il numero di informazioni che verrebbero messe in discussione, se quel dato fosse confutato (in reti di dati, misura la connettività del nodo) RILEVANZA: esprime l’utilità pragmatica di un dato, cioè il numero e il valore degli scopi al cui perseguimento quel dato risulta utile GRADEVOLEZZA: esprime la soddisfazione (o il disappunto) connessa all’accettazione del dato, quando ciò coincide con l’appagamento (o la frustrazione) di uno o più scopi dell’agente Nota: in questo modello, le ragioni del credere si caratterizzano come proprietà dei dati, sulla base delle quali avviene il processo di focalizzazione (rilevanza) e selezione delle credenze (credibilità, importanza, gradevolezza) Dati e credenze: tabella riassuntiva:  Dati e credenze: tabella riassuntiva Osservazioni conclusive:  Osservazioni conclusive La distinzione fra dati e credenze (ovvia) rivoluziona il modo in cui viene concepita la revisione delle credenze, e apre la strada ad approcci ibridi (simbolici e sub-simbolici) per l’implementazione Un modello procedurale permette di proporre una teoria unificata delle diverse fasi del processamento cognitivo della conoscenza (information update, selezione e revisione delle credenze, ragionamento inferenziale, oblio) La revisione delle credenze è un fenomeno complesso, con profondi legami con altri processi cognitivi, e soggetto a variazioni individuali (sia perché agenti cognitivi diversi possono applicare diverse strategie di revisione, sia perché contesti differenti determinano differenti criteri ottimali di revisione) Slide26:  Influence of Social Motivation over Belief Dynamics A Game-Theoretical Approach Fabio Paglieri Dottorato in Scienze Cognitive, Università di Siena Cristiano Castelfranchi ISTC-CNR Roma IT / Università di Siena Presentato a: CogEco’05 - International Conference on Cognitive Economics (Sofia, 5-8 August 2005) Questions to Be Addressed:  Questions to Be Addressed Does motivation affect belief formation and change? Yes Do current formal models of belief dynamics capture such motivational effects? No How can we model motivational effects over belief dynamics? Game-theoretical modelling of the influence of social motivation over dialogical belief dynamics Extension to the single-agent case (only outlined) Evidence of Motivational Influence:  Evidence of Motivational Influence tendency toward social conformity (Asch, 1952) avoidance of cognitive dissonance (Festinger, 1957) motivational biases in probability attribution (Zuckerman, 1979) effects of motivated reasoning (Kunda, 1990) tendency toward self-verification (Swann, 1990) effects of denial (Miceli, Castelfranchi, 1998) motivational factors in delusions (McKay et al., 2005; McKay, Paglieri, 2006) [?] framing effects (Kahneman, Tversky, 1979) Basic Patterns of Influence:  Basic Patterns of Influence Motivations influence: the information and/or features we pay attention to (i.e. attentional focus); our assessment of such evidence (i.e. belief formation and change). Our interest here is confined to the latter. A Case Study: Social Motivation:  A Case Study: Social Motivation Problem: effects of social motivation over belief change in dialogical interaction Setting: two agent arguing with each other on a claim on which they hold mutually excluding views (P and not-P), each of them can end up either maintaining or revising her own position Notes: all-or-nothing situation with no intermediate possibilities standard game-theory but applied with some ‘creativity’ (cf. non-negotiability of belief) Basic Motivational Attitudes:  Basic Motivational Attitudes Motivations modelled as preference orderings Basic attitudes as partial orderings Cognitive attitudes CONSERVATIVE: Maintain is preferred over Revise (MM or MR) > (RM or RR) EXPLORATIVE: Revise is preferred over Maintain (RM or RR) > (MM or MR) Social attitudes COOPERATIVE: Agreement is preferred over Disagreement (MR or RM) > (MM or RR) ANTAGONISTIC: Disagreement is preferred over Agreement (MM or RR) > (MR or RM) Motivational Profiles:  Motivational Profiles Motivational profiles as total orderings, obtained by (1) integrating cognitive and social attitudes and by (2) giving priority to one of the two of them over the other Individual-oriented profiles: cognitive > social Social-oriented profiles: social > cognitive Two Despotic Agents:  Two Despotic Agents Despotic profile: MR > MM > RM > RR (individual-oriented) Two Contentious Agents:  Two Contentious Agents Contentious profile: MM > RR > MR > RM (social-oriented) Despotic & Agreeable:  Despotic & Agreeable Despotic profile: MR > MM > RM > RR (individual-oriented) Agreeable profile: RM > MR > RR > MM (social-oriented) Open-minded & Polemist:  Open-minded & Polemist Open-minded profile: MR > RM > MM > RR (social-oriented) Polemist profile: RR > MM > RM > MR (social-oriented) Results Achieved So Far:  Results Achieved So Far To sum up: whatever combination of motivational profiles we consider, we are always able to determine (either in strategic or extended form) a jointly preferred outcome. Such preferred outcome depends on: the agents’ motivational profiles; their knowledge of each other’s motivational profiles. So far we assume such knowledge to be exhaustive and exact… but what happens if we now relax this constraint? All Possible Preferred Outcomes:  All Possible Preferred Outcomes Agent’s motivational profile Agent’s assumption on the counterpart Simplified Matrix I:  Simplified Matrix I Agent’s motivational profile Agent’s assumption on the counterpart Simplified Matrix II:  Simplified Matrix II Agent’s motivational profile Agent’s assumption on the counterpart Notes on the table: individual-oriented types do not need to know anything on the other agent motivations; social-oriented types need (A) only a limited amount of information, whose nature (B) depends on their own motivational profile. Conclusions on the Social Case:  Conclusions on the Social Case Social motivation effects over belief dynamics depend (1) on the agent motivations (trivial) and (2) on her beliefs concerning the motivations of others (slightly less trivial). However and more interestingly, (3) such beliefs do not need to be complete, and (4) the diagnostic value of each piece of information depends once again on the agent motivations. Bottom-line: even considering a toy-problem within a very abstract framework, the interaction between motivations and belief dynamics turns out to be complex, deep and many-faceted – hence it calls for further and careful investigation. Extension to the Single-Agent Case:  Extension to the Single-Agent Case Problem: effects of motivation over individual belief dynamics Setting: an agent who has to decide whether to maintain or drop previous beliefs, when faced with new information that he might either accept or reject Notes: here the relevant axis is time, not sociality precise mapping with the predicted outcomes of AGM belief revision Basic Motivational Attitudes:  Basic Motivational Attitudes Motivations modelled as preference orderings Basic attitudes as partial orderings Past-directed attitudes CONFIDENT: Maintain is preferred over Drop (Expansion or Stasis) > (Revision or Contraction) UNSURE: Drop is preferred over Maintain (Revision or Contraction) > (Expansion or Stasis) Future-directed attitudes TRUSTFUL: Accept is preferred over Reject (Expansion or Revision) > (Stasis or Contraction) SCEPTICAL: Reject is preferred over Accept (Stasis or Contraction) > (Expansion or Revision) Motivational Profiles:  Motivational Profiles Motivational profiles as total orderings, obtained by (1) integrating past-directed and future-directed attitudes and by (2) giving priority to one of the two of them over the other Past-oriented profiles: past-directed > future-directed Future-oriented profiles: future-directed > past-directed Observations on the Individual Case:  Observations on the Individual Case Coordination is not an issue here, since there is only one agent with a unique motivational profile (that is, we disregard the idea of the agent ‘arguing with himself’ while assessing his beliefs). Comparison with the AGM framework for belief revision: we do not force acceptance of the new input (non-prioritized belief revision; cf. Hansson et al., 2001), in contrast with the original AGM model; the main concern behind the agent’s preferences in AGM belief revision is semantic (consistency), while here it is purely motivational; orderings in AGM belief revision apply to sets of propositions (epistemic entrenchment), while here they apply to operations over such sets (i.e., they express preferences over belief revision strategies; Rott, 2004). Again, the main interest lies in the mutual interplay between motivation and beliefs: (1) motivational profiles might affect belief dynamics (e.g. biasing towards a specific preferred strategy), but (2) the agent understanding of past and current events (i.e., some of his beliefs) will also contribute to define his own motivational profile (e.g., making him more trustful or sceptical, confident or unsure, etc.). Future Work:  Future Work Refinement of the model. Closer comparison with existing empirical evidence of motivational influence over beliefs. Integration with formal models of belief dynamics, both AGM and non-AGM (e.g., DBR, Data-oriented Belief Revision: Castelfranchi, 1996;1997; Paglieri, 2004). Single-agent and multi-agent computational modelling. Empirical testing of the model: analysis of argumentation dialogues (Paglieri, Castelfranchi, 2004); economical experiments (problem: how to manipulate the subjects’ motivations without removing or lowering their epistemic concerns?).

Add a comment

Related presentations

Related pages

Dom Santa Maria Assunta, Siena, Toskana

Dom Santa Maria Assunta, Siena, Toskana: Impressum / Copyright ... Dom Santa Maria Assunta, Siena, Toskana: Impressum / Copyright
Read more

siena - Fernweh - Leserfotos - Stuttgarter Zeitung

und hier ein kleiner Spaziergang durch Siena. 06.10.2015, 17 Bilder von epigramma. Bewerten. Diese Galerie gefällt 15 Usern Details.
Read more

Siena 2006 - Istituto Nazionale di Fisica Nucleare

S i e n a 2 0 0 6: 10th Topical Seminar on Innovative Particle and Radiation Detectors (IPRD06) 1 - 5 October 2006 Siena, Italy GEANT4 tutorial workshop 5 ...
Read more

Diary of a Social Butterfly | An 8th Grade perspective ...

An 8th Grade perspective about climbing the social scale. Menu. Home; About; Search for: Skip to content. ... Posted on September 30, 2014 by siena06.
Read more

siena_06

wstecz · zurück · wstecz · zurück
Read more

Wetter Siena: 7-Tage Prognose | wetter.com

Das Wetter in Siena. Finden Sie bei wetter.com die aktuelle Wettervorhersage für heute und die nächsten 7 Tage inkl. Temperatur, Wetterzustand und ...
Read more

Racine Dominicans Justice Outreach

siena06 Author: Judy Miner Subject: siena06 Created Date: 1/31/2006 9:17:50 AM ...
Read more

Wetter Siena morgen | wetter.com

Regen oder Sonne? So wird das Wetter morgen in Siena. Temperatur-, Wind- und Regenvorhersage auf wetter.com.
Read more

AC Siena (06/07) | National Football Teams

AC Siena 06/07. Choose season: Name D. o. B. Land of origin Position Season M G; Manninger, Alex: 1977-06-04: Austria Goalkeeper: 06/07: 38: 0: Jaakkola ...
Read more

Florencia & Siena'06 - YouTube

FLORENCIA, FIRENZE, PISA, ASÍS, SIENA y PADUA, ITALIA, ITALY, (VIDEO), por JC.avi - Duration: 14:52. Juan Carlos Garcia Blanco 26,377 views
Read more