A Computationally Grounded Framework for Cognitive Attitudes (extended version)
Tiago de Lima, Emiliano Lorini, Elise Perrotin, François Schwarzentruber · 2024
Read A Computationally Grounded Framework for Cognitive Attitudes (extended version) by Tiago de Lima, Emiliano Lorini, Elise Perrotin, François Schwarzentruber (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-14073.txt.
Introduction
Tiago de Lima, Emiliano Lorini, Elise Perrotin, and François Schwarzentruber wrote a paper in twenty twenty-four that sets out a formal language for the mental attitudes of agents, human or artificial. The language covers beliefs, what an agent holds to be true, and wants and likes, what an agent is drawn to or away from.
Meanings rest on belief bases, plain data structures a computer can hold, rather than on abstract possible worlds. Five modal operators state full belief, full attraction, full repulsion, and more realistic forms of the last two. The four give rules for the system and prove that none of the five can be swapped for the others. Mixed together, the operators can state mixed feelings, no preference, and being spurred on or put off. A further part of the language tracks how attitudes change.
Catalog record
| CREATOR | Tiago de Lima, Emiliano Lorini, Elise Perrotin, François Schwarzentruber |
|---|---|
| YEAR | 2024 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 84 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2412-14073
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
A Computationally Grounded Framework for Cognitive Attitudes (extended version) Tiago de Lima, Emiliano Lorini, Elise Perrotin, François Schwarzentruber Abstract We introduce a novel language for reasoning about agents’ cognitive attitudes of both epistemic and motivational type. We interpret it by means of a computationally grounded semantics using belief bases. Our language includes five types of modal operators for implicit belief, complete attraction, complete repulsion, realistic attraction and realistic repulsion. We give an axiomatization and show that our operators are not mutually expressible and that they can be combined to represent a large variety of psychological concepts including ambivalence, indifference, being motivated, being demotivated and preference. We present a dynamic extension of the language that supports reasoning about the effects of belief change operations. Finally, we provide a succinct formulation of model checking for our languages and a PSPACE model checking algorithm relying on a reduction into TQBF. We present some experimental results for the implemented algorithm on computation time in a concrete example. 1 CRIL, Univ Artois and CNRS, Lens, France 2 IRIT, CNRS, Toulouse University, France 4 National Institute of Advanced Industrial Science and Technology, Tokyo, Japan 3 ENS Lyon, LIP, France [email protected], [email protected], [email protected], [email protected] Code — https://gitlab.in2p3.fr/tiago.delima/cognitive-attitudes-source-code/ Introduction An agent’s cognitive state encompasses her epistemic attitudes (e.g., beliefs) and motivational (or conative) attitudes (e.g., desires and preferences). Their relationships as well as their influence on the agent’s behavior are objects of study in cognitive psychology ( Albarracín et al. 2018 ) and philosophy of mind ( Searle 2001 ; Humberstone 1992 ) . They play a prominent role in the explanation of others’ behaviours and of our own behaviours through the so-called intentional stance ( Dennett 1987 ) . Cognitive (or mental) attitudes have also been studied by logicians, both in philosophy and in AI. Several logics dealing with epistemic and practical reasoning of rational agents have been proposed. These include epistemic logics ( Hintikka 1962 ; Fagin et al. 1995 ) , logics of preferences ( Wright 1963 ; Wright 1972 ; Liu 2011 ; van Benthem, Girard, and Roy 2009 ) , logics of beliefs and preferences ( Boutilier 1994 ; Lorini 2021 ; Azimipour and Naumov 2021 ) , logics of desires and pro-attitudes ( Dubois, Lorini, and Prade 2017 ; Su et al. 2007 ) , logics of intention ( Shoham 2009 ; Icard, Pacuit, and Shoham 2010 ; Lorini and Herzig 2008 ) , BDI (belief, desire, intention) logics ( Cohen and Levesque 1990 ; Herzig and Longin 2004 ; Meyer, van der Hoek, and van Linder 1999 ; Wooldridge 2000 ) . The idea of describing rational agents in terms of their epistemic and motivational attitudes is shared with classical decision theory and game theory, according to which rational agents are assumed to make decisions on the basis of their beliefs and preferences. Most logics of cognitive attitudes rely on extensional semantics based on so-called Kripke models: sets of possible worlds (or states) supplemented with one or more binary accessibility relations for each agent representing, e.g., the agent’s epistemic uncertainty or preference ordering over the states. Multi-relational Kripke models are general and mathematically elegant. Nonetheless, they are limited from a modeling point of view as they are not succinct. Even a simple situation like a card game (e.g., Hanabi) with 4 players having 5 cards each among a set of 50 cards require a Kripke model with 5.5\times 10^{23} states to represent all alternatives on which an agent’s uncertainty and preferences bear. For this reason, it is hard, if not unfeasible, to implement model checking for a logic of cognitive attitudes using a Kripke semantics since the model cannot be explicitly constructed. Some alternatives to solve this succinctness problem have been proposed in the area of epistemic logic including semantics based on BDDs ( van Benthem et al. 2018 ) , Boolean formulas and programs ( Charrier and Schwarzentruber 2015 ; Charrier and Schwarzentruber 2017 ; Charrier and Schwarzentruber 2018 ; Charrier, Pinchinat, and Schwarzentruber 2019 ) , and the notion of visibility ( Cooper et al. 2021 ; van der Hoek, Troquard, and Wooldridge 2011 ) . However, to our knowledge no general succinct semantics for a logic of cognitive attitudes, combining epistemic ones with motivational ones, has been proposed up to now. In this paper, we tackle this problem by relying on a semantics for cognitive attitudes based on the notion of belief base that has been shown to provide an interesting computationally grounded alternative to Kripke models in representing epistemic concepts in the single-agent case ( Konolige 1986 ) as well as in the multi-agent case ( Lorini 2018 ; Lorini 2019 ; Lorini 2020 ; Lorini and Rapion 2022 ) . Unlike standard Kripke semantics for logics of cognitive attitudes in which possible states and agents’ accessibility relations are given as primitive, in our semantics they are defined from and grounded on the primitive concept of belief base. We will use belief bases to define three types of accessibility relation capturing, respectively, the states that an agent considers possible, those she finds attractive and those she finds repulsive. The first belongs to the epistemic sphere, while the second and the third belong to the motivational sphere. By means of this semantics, we will interpret a modal language including five modal operators for implicit belief, complete attraction, complete repulsion, realistic attraction and realistic repulsion as well as a dynamic extension of it that supports reasoning about the effects of belief change operations. We will show that in our semantics model checking can be formulated in a more succinct way than in the standard Kripke semantics, which opens up the possibility of using it in practice. We will provide a PSPACE model checking procedure relying on a reduction into TQBF and some experimental results for the implemented procedure on computation time. Related work Our belief base approach has some aspects in common with the body of literature on preference representation based on priority graphs ( de Jongh and Liu 2009 ; Liu 2011 ; Souza and Moreira 2021 ; Souza 2016 ) , that was recently extended to deontic logic in ( van Benthem, Grossi, and Liu 2014 ) . We share with these works the idea i) that there are two alternative approaches to the representation of preferences and more generally of mental attitudes, namely the extensional approach based, e.g., on multi-relational Kripke models and the syntactic approach based, e.g., on belief bases or priority graphs, and ii) that the two approaches must coexist and be integrated into the same logical framework. But there are also some important differences. These works are mostly focused on the single-agent case while our approach is multi-agent. Secondly, they mainly focus on preference representation, while our emphasis is on the bipolar aspect of motivational attitudes (i.e., attraction vs repulsion, motivation vs demotivation). Thirdly, our modal language of cognitive attitudes is profoundly different from existing ones, including from the one presented in ( Souza 2016 , Chapters 4 and 5) which, like ours, covers both epistemic and motivational attitudes. Souza’s language includes “betterness” (S4) and “strict betterness” modalities, while our language includes primitive modalities for (complete and realistic) attraction and repulsion. Our language