Self-Destructing Models: Increasing the Costs of Harmful Dual Uses of Foundation Models
Open Access
- 8 August 2023
- conference paper
- Published by Association for Computing Machinery (ACM)
- p. 287-296
Abstract
A growing ecosystem of large, open-source foundation models has reduced the labeled data and technical expertise necessary to apply machine learning to many new problems. Yet foundation models pose a clear dual-use risk, indiscriminately reducing the costs of building both harmful and beneficial machine learning systems. Policy tools such as restricted model access and export controls are the primary methods currently used to mitigate such dual-use risks. In this work, we review potential safe-release strategies and argue that both policymakers and AI researchers would benefit from fundamentally new technologies enabling more precise control over the downstream usage of open-source foundation models. We propose one such approach: the task blocking paradigm, in which foundation models are trained with an additional mechanism to impede adaptation to harmful tasks without sacrificing performance on desirable tasks. We call the resulting models self-destructing models, inspired by mechanisms that prevent adversaries from using tools for harmful purposes. We present an algorithm for training self-destructing models leveraging techniques from meta-learning and adversarial learning, which we call meta-learned adversarial censoring (MLAC). In a small-scale experiment, we show MLAC can largely prevent a BERT-style model from being re-purposed to perform gender identification without harming the model’s ability to perform profession classification.Keywords
All Related Versions
This publication has 7 references indexed in Scilit:
- Dual use of artificial-intelligence-powered drug discoveryNature Machine Intelligence, 2022
- Recommendations on Export Controls for Artificial Intelligence2020
- Keeping Top AI Talent in the United States2019
- Bias in Bios2019
- Domain Generalization with Adversarial Feature Learning2018
- Deconfounded Lexicon Induction for Interpretable Social Science2018
- Hyperopt: A Python Library for Optimizing the Hyperparameters of Machine Learning AlgorithmsProceedings of the Python in Science Conference, 2013