Inferring Capabilities from Task Performance with Bayesian Triangulation

Fuente: arXiv
Enregistré dans:
Détails bibliographiques
Auteurs principaux: Burden, John, Voudouris, Konstantinos, Burnell, Ryan, Rutar, Danaja, Cheke, Lucy, Hernández-Orallo, José
Format: Preprint
Publié: 2023
Sujets:
Accès en ligne:
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866915537157619712
author Burden, John
Voudouris, Konstantinos
Burnell, Ryan
Rutar, Danaja
Cheke, Lucy
Hernández-Orallo, José
author_facet Burden, John
Voudouris, Konstantinos
Burnell, Ryan
Rutar, Danaja
Cheke, Lucy
Hernández-Orallo, José
contents As machine learning models become more general, we need to characterise them in richer, more meaningful ways. We describe a method to infer the cognitive profile of a system from diverse experimental data. To do so, we introduce measurement layouts that model how task-instance features interact with system capabilities to affect performance. These features must be triangulated in complex ways to be able to infer capabilities from non-populational data -- a challenge for traditional psychometric and inferential tools. Using the Bayesian probabilistic programming library PyMC, we infer different cognitive profiles for agents in two scenarios: 68 actual contestants in the AnimalAI Olympics and 30 synthetic agents for O-PIAAGETS, an object permanence battery. We showcase the potential for capability-oriented evaluation.
format Preprint
id arxiv_https___arxiv_org_abs_2309_11975
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Inferring Capabilities from Task Performance with Bayesian Triangulation
Burden, John
Voudouris, Konstantinos
Burnell, Ryan
Rutar, Danaja
Cheke, Lucy
Hernández-Orallo, José
Artificial Intelligence
As machine learning models become more general, we need to characterise them in richer, more meaningful ways. We describe a method to infer the cognitive profile of a system from diverse experimental data. To do so, we introduce measurement layouts that model how task-instance features interact with system capabilities to affect performance. These features must be triangulated in complex ways to be able to infer capabilities from non-populational data -- a challenge for traditional psychometric and inferential tools. Using the Bayesian probabilistic programming library PyMC, we infer different cognitive profiles for agents in two scenarios: 68 actual contestants in the AnimalAI Olympics and 30 synthetic agents for O-PIAAGETS, an object permanence battery. We showcase the potential for capability-oriented evaluation.
title Inferring Capabilities from Task Performance with Bayesian Triangulation
topic Artificial Intelligence
url https://arxiv.org/abs/2309.11975