Dataset Information

An unethical optimization principle.

ABSTRACT: If an artificial intelligence aims to maximize risk-adjusted return, then under mild conditions it is disproportionately likely to pick an unethical strategy unless the objective function allows sufficiently for this risk. Even if the proportion ? of available unethical strategies is small, the probability p U of picking an unethical strategy can become large; indeed, unless returns are fat-tailed p U tends to unity as the strategy space becomes large. We define an unethical odds ratio, ? (capital upsilon), that allows us to calculate p U from ?, and we derive a simple formula for the limit of ? as the strategy space becomes large. We discuss the estimation of ? and p U in finite cases and how to deal with infinite strategy spaces. We show how the principle can be used to help detect unethical strategies and to estimate ?. Finally we sketch some policy implications of this work.

SUBMITTER: Beale N

PROVIDER: S-EPMC7428226 | biostudies-literature | 2020 Jul

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

An unethical optimization principle.

Beale Nicholas N Battey Heather H Davison Anthony C AC MacKay Robert S RS

Royal Society open science 20200701 7

If an artificial intelligence aims to maximize risk-adjusted return, then under mild conditions it is disproportionately likely to pick an unethical strategy unless the objective function allows sufficiently for this risk. Even if the proportion η of available unethical strategies is small, the probability p U of picking an unethical strategy can become large; indeed, unless returns are fat-tailed p U tends to unity as the strategy space b ...[more]

PMID: 32874640

Similar Datasets

Project description:BACKGROUND:Screening for Down's syndrome forms part of routine obstetric practice. Ethical considerations relating to genetic screening form a major part of the workload of research ethics committees. This study investigated the attitudes of research ethics committee members to several conditions varying in clinical severity and prognosis, including Down's syndrome. METHODS:The members of 40 randomly chosen research ethics committees were surveyed. A simple questionnaire comprising 19 clinical scenarios based around four "clinical" conditions was designed to review conditions that were potentially embarrassing, affecting life span but not mental ability, premature death, and intellectual impairment with a risk of neonatal cardiac defects (Down's syndrome). Screening tests with different degrees of effectiveness were described and the diagnostic test descriptions ranged from having no risk to an unaffected fetus to causing spontaneous abortion of two normal fetuses for each affected fetus identified. Replies were graded on a scale of 1 to 5. RESULTS:Seventy seven replies were received from 28 different research ethics committees. Screening was supported for treatment of a life threatening condition (95% in favour) but screening for conditions of a slight increase in premature death (14% in favour) or cosmetic features (10% in favour) were considered unethical. Views were ambiguous (49% in favour) about conditions involving significant shortening of lifespan. Down's syndrome screening was considered more ethical when described as a serious condition (56% in favour) than when the clinical features were described (44% in favour). Once increased rates of spontaneous abortion on confirmatory testing were added, 79% (21% in favour) and 86% (14% in favour) stated that screening was unethical (for "serious" and "clinical features" descriptions, respectively). CONCLUSIONS:Down's syndrome screening raises ethical concerns about genetic testing in general that need to be dealt with before the introduction of any prenatal screening test.

Dataset Information

An unethical optimization principle.

Publications

An unethical optimization principle.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure