Shaking the foundations: delusions in sequence models for interaction and control

The recent phenomenal success of language models has reinvigorated machine learning research, and large sequence models such as transformers are being applied to a variety of domains. One important problem class that has remained relatively elusive however is purposeful adaptive behavior. Currently...

Full description

Saved in:
Bibliographic Details
Published in:arXiv.org 2021-10
Main Authors: Ortega, Pedro A, Kunesch, Markus, Delétang, Grégoire, Genewein, Tim, Grau-Moya, Jordi, Veness, Joel, Buchli, Jonas, Degrave, Jonas, Piot, Bilal, Perolat, Julien, Everitt, Tom, Tallec, Corentin, Parisotto, Emilio, Erez, Tom, Chen, Yutian, Reed, Scott, Hutter, Marcus, Nando de Freitas, Legg, Shane
Format: Article
Language:English
Subjects:
Error signals
Machine learning
Shaking
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The recent phenomenal success of language models has reinvigorated machine learning research, and large sequence models such as transformers are being applied to a variety of domains. One important problem class that has remained relatively elusive however is purposeful adaptive behavior. Currently there is a common perception that sequence models "lack the understanding of the cause and effect of their actions" leading them to draw incorrect inferences due to auto-suggestive delusions. In this report we explain where this mismatch originates, and show that it can be resolved by treating actions as causal interventions. Finally, we show that in supervised learning, one can teach a system to condition or intervene on data by training with factual and counterfactual error signals respectively.
ISSN:2331-8422