Skip to content

Latest commit

 

History

History
346 lines (254 loc) · 17.5 KB

README.md

File metadata and controls

346 lines (254 loc) · 17.5 KB

rio: Reasoner for Input/Output Logics

Author: Alexander Steen

You may cite the tool as: DOI

Input/Output (I/O) logics have been devised by D. Makinson and L. van der Torre [1] as a class of formal systems for norm-based deontic reasoning. Intuitively, they formalize the detachment process of obligations (referred to as the output) from a given set of conditional norms and a specific situation (called the input). It differs from other deontic logics (e.g., based on modal logics or extensions of it) in the sense that the norms themselves are not part of the object logic and hence do not carry truth values. Constrained I/O logic [2] extend plain I/O logics by restricting the output set to be consistent with respect to a given set of formulas, addressing deontic paradoxes such as contrary-to-duty situations and other dilemmas.

In the standard formulation, there are four different principal output operators called out1, out2, out3 and out4. out1 allows for basic detachment, out2 adds reasoning-by-cases principles, and out3 and out4 augment out1 and out2, respectively, with (cumulative) transitivity. For each operator there exists a so-called throughput variant outi+. Constrained I/O logic filter excess output that is inconsistent wrt. a set of constraints. Preferred output further filters the set of consistent outputs such that only preferred sets (wrt. to given preference relation) are returned. For details, cf. we refer to [1].

The Tool

rio provides a TPTP-aligned [3] automated reasoning system for unconstrained and constrained I/O logics based on the operators outi and outi+, 1 ≤ i ≤ 4. It is implemented as a Scala application and based on the scala-tptp-parser [4].

In short, the system allows you to specify a set of conditional norms and a number of inputs (each of which describing aspects of the current situational context), and provides automated means for inferring whether given obligations (also encoded as formulas) can be derived. rio can also be used to infer the set of detached obligations instead of checking detachment of given ones. See Usage below for details.

It is inspired by the decision procedures introduced in [5], though the actual procedures differ.

Installation

The simplest method of obtaining rio is to download the provided .jar file of the most recent release from GitHub, see https://github.com/aureleeNet/rio/releases. You can also download the sources and build rio yourself.

Using a pre-built built

When using a pre-built .jar file, you are already done. rio is run by (assuming that the .jar file is called rio.jar and in your current working directory) ...

> java -jar rio
usage: rio [-o <operator>] [-p <parameter>] [-c <constraint>] <problem file>
[...]

Building from sources

In order to build rio from sources, you need the following software/packages:

  • Make
  • sbt (scala build tool)
  • JDK (≥ 11)
  • g++
  • zlib
  • must (installation described below)

Unpack the source archive of rio. Download must from https://github.com/jar-ben/mustool/archive/refs/heads/master.zip and extract the contents of the archive into a folder called mustool (important!) as top-level directory within the rio folder. Your rio folder should now look like this (only directories shown):

.
├── contrib
├── examples
├── mustool
├── project
└── src

Run make. If no error occurs, there should be a bin directory containing the rio.jar file. Also, there is a convenience executable rio that just wraps the jar file with shell run script.

Usage

rio requires a Java runtime environment (JRE), version ≥ 11, to be run. Older java versions might also work, though this has not been tested. Additionally, rio is currently only tested on Linux as it uses a shared PicoSAT library only provided for linux-x86-64 architecture. This will be addressed soon (hopefully!).

> java -jar rio
usage: rio [-o <operator>] [-p <parameter>] [-c <constraint>] <problem file>

 <problem file> can be either a file name or '-' (without parentheses) for stdin.

 Options:
  -o <operator>
     If <problem file> does not contain a logic specification statement, explicitly set
     the output operator to be assumed to <operator>.
     This option is ignored, if <problem file> contains a logic specification statement.
     Supported <operator>s are: out1, out2, out3, out4

  -p <parameter>
     Set additional parameters to the output operator.
     This option is ignored, if <problem file> contains a logic specification statement.
     Multiple parameters can be passed.
     Supported <parameter>s are: throughput, constrained-credulous, constrained-skeptical

  -c <constraint>
     Pass constraints to the reasoning system. <constraint> must be a valid THF formula
     (without annotations). Multiple constraints can be passed by passing multiple
     respective -c options.
     This option is ignored, if <problem file> contains a logic specification statement.
     Constraints are ignored if neither -p constrained-credulous nor -p constrained-skeptical
     was passed.

  --version
     Prints the version number of the executable and terminates.

  --help
     Prints this description and terminates.

Input and output formats

The input files for rio are plain text files according to the TPTP TFF language (typed first-order logic), though only the propositional fragment of TFF is currently supported. The semantics of formula roles differs from the TPTP standard (see below), and the problem needs to contain a logic specification of the form

tff(<name>, logic, $$iol == [<parameter> == <value>, ...]).

Alternatively, semantics parameters can also be passed using command-line arguments.

rio gives SZS status results as follows:

  • SZS status Success if the problem does not contain conjectured outputs (formulas of role conjecture). In this case, a SZS result of type ListOfFormulae containing the finite base of the output set is given.
  • SZS status Theorem if the problem contains a conjectured output (a formula of role conjecture) and the conjectured output is indeed in the output set.
  • SZS status CounterSatisfiable if the problem contains a conjectured output (a formula of role conjecture) and the conjectured output is not in the output set.
  • If the problem contains multiple conjectures, one output (Theorem or CounterSatisfiable) is given for each conjecture.

Semantics specification

A problem's semantics is specified using a dedicated logic formula, as foreseen in the currently explored extension of the TPTP library, cf. http://tptp.org/TPTP/Proposals/LogicSpecification.html for details.

Parameter Value Description
$$operator One of $$out1, $$out2, $$out3, $$out4 Select the output operator (required parameter)
$$throughput $true or $false Enable/Disable throughout (default: $false)
$$constrained $$credulous or $$skeptical Enable constrained output operators with credulous or skeptical net output function, respectively (default: unset)
$$constraints [<list of formulas>] Set the constraints for constrained output. Ignored of $$constrained is not set (default: unset).
$$preference [<list of name tuples>] A comma-separated list of tuples of the form [name1,name2,...], where each identifier refers to the name of a norm in the problem file. Earlier norms are preferred over later norms, norms in the same tuple are equally preferred. Tuple symbols may be dropped if they are singletons. Ignored of $$constrained is not set (default: unset).

Problem contents

Formula role Description
axiom TPTP formulas of role axiom specify the conditional norms. Every axiom formula needs to be an expression of the form {$$norm} @ (left,right) where both left and right are formulas. left is the body of the norm (the condition) and right is the head of the norm (the obligation). The can be arbitrarily many formulas of role axiom in a problem file (zero, too).
hypothesis TPTP formulas of role hypothesis denote the inputs and are regular formulas. There can be arbitrarily many formulas of role hypothesis in the problem file (zero is ok).
conjecture TPTP formulas of role conjecture denote conjectured outputs. For each conjecture, rio will check whether the formula is indeed in the output set and return a corresponding SZS result. There can be arbitrarily many formulas of role conjecture in the problem file (zero is ok).

Examples

Use case 1: Check whether given obligations can be detached

In order to check whether a number of obligations can actually be derived from the given set of norms and the situational context (the input), the obligations in question are specified as THF formulas of role "conjecture":

Let's say this content is stored in a file named out1.p:

tff(semantics, logic, (
    $$iol == [ $$operator == $$out1 ] )).

tff(norm1, axiom, {$$norm} @ ($true, helping) ).
tff(norm2, axiom, {$$norm} @ (helping, telling) ).
tff(norm3, axiom, {$$norm} @ (~helping, ~telling) ).

tff(input1, hypothesis, ~helping).

tff(c1, conjecture, ~telling).
tff(c2, conjecture, ~helping).

Then you run rio via:

./rio out1.p

... producing the following output:

% SZS status Theorem for src/test/resources/out1.p: c1
% SZS status CounterSatisfiable for src/test/resources/out1.p: c2

As can be seen, only one of the conjectures (only c1) is indeed detached. As a consequence, SZS status Theorem is given for c1 but SZS status CounterSatisfiable is given for conjecture c2.

Use case 2: Generate set of outputs

If no conjectures are contained in the problem file, e.g., as in ...

thf(semantics, logic, (
    $$iol == [ $$operator == $$out3 ] )).

thf(norm1, axiom, {$$norm} @ ($true, helping) ).
thf(norm2, axiom, {$$norm} @ (helping, telling) ).
thf(norm3, axiom, {$$norm} @ (~helping, ~telling) ).

thf(input1, hypothesis, ~helping).

rio will instead output a complete finite basis of detachable formulas and SZS status Success:

% SZS status Success for out3.p
% SZS output start ListOfFormulae for out3.p
$false
% SZS output end ListOfFormulae for out3.p

A finite basis is a finite set of formulas A such that every detachable output x follows from A, i.e., A ⊧ x. This allows to have a finite representation for the infinite set of outputs. In the above case, every output x is in the output set as {⊥} ⊧ x for every x.

Use case 3: Constrained outputs

In the previous example (use case 2), the output set becomes too large as the input and the set of norm entail an inconsistency, known as Contrary To Duty situations in the Deontic logic community.

In the context of I/O logics, the output set can be constrained to remedy this. This is reflected in rio by enabling constrained I/O operations via the logic specification:

thf(semantics, logic, (
    $$iol == [ $$operator == $$out3,
               $$constrained == $$credulous,
               $$constraints == [~helping] ] )).

thf(norm1, axiom, {$$norm} @ ($true, helping) ).
thf(norm2, axiom, {$$norm} @ (helping, telling) ).
thf(norm3, axiom, {$$norm} @ (~helping, ~telling) ).

thf(input1, hypothesis, ~helping).

In this case, the output is constrained to be the credulous output aggregate of all maximally consistent norm families wrt. detachment via out3 and the constraint ~helping. See [2] for the theoretical details.

This problem file gives a consistent output set as follows:

% SZS status Success for out3-constrained.p
% SZS output start ListOfFormulae for out3-constrained.p
~ telling
% SZS output end ListOfFormulae for out3-constrained.p

Use case 4: Preferred output

In the constrained case, a preference ordering can optionally be imposed to further restrict the generated outputs.

tff(semantics, logic, (
    $$iol == [ $$operator == $$out3,
               $$constrained == $$credulous,
               $$preference == [norm1, norm2, norm3] ] )).

tff(norm1, axiom, {$$norm} @ (heatingOn, ~windowOpen) ).
tff(norm2, axiom, {$$norm} @ ($true, windowOpen) ).
tff(norm3, axiom, {$$norm} @ ($true, heatingOn) ).

In this case, the output is constrained to be the credulous output aggregate of all maximally consistent norm families wrt. detachment via out3 (and the empty set of constraints). Additionally, the preference ordering name1 > name2 > name3 is declared. See [6] for details.

This problem file gives a consistent output set as follows:

% SZS status Success for examples/preferred.p
% SZS output start ListOfFormulae for examples/preferred.p
windowOpen
% SZS output end ListOfFormulae for examples/preferred.p

Usage as library/API

rio's reasoning procedure can also be invoked by adding the .jar file to your project's classpath and calling the Reasoner.apply method directly.

License

rio is distributed under the BSD 3-Clause license (see LICENSE file), and uses third party libraries that are distributed under their own terms (see LICENSE-3RD-PARTIES file).

In particular, rio uses

References

[1] Makinson, D., van der Torre, L.W.N.: Input/Output Logics. J. Philosophical Logic 29(4), 383–408 (2000). https://doi.org/10.1023/A:1004748624537

[2] Makinson, D., van der Torre, L.W.N.: Constraints for Input/Output Logics. J. Philos. Log. 30(2): 155-185 (2001). https://doi.org/10.1023/A:1017599526096

[3] Sutcliffe, G, and Suttner, C.: The TPTP Problem Library for Automated Theorem Proving. http://tptp.org.

[4] Steen, A.: scala-tptp-parser (Version v1.2). Zenodo, 2021. DOI: http://doi.org/10.5281/zenodo.4468959

[5] Steen, A.: Goal-Directed Decision Procedures for Input/Output Logics. In 15th International Conference on Deontic Logic and Normative Systems (DEON 2020/2021, Munich), Fenrong Liu, Alessandra Marra, Paul Portner, and Frederik Van De Putte (Eds.), College Publications, London, 2021. (to appear). See: http://www.collegepublications.co.uk/DEON/?00003

[6] Parent, X.: Moral particularism in the light of deontic logic. Artif. Intell. Law 19(2-3): 75-98 (2011)