Back to Home

Code of Ethics

Guidelines for ethical research practices in AI and machine learning

Overview

This Code of Ethics guides our community toward higher ethical standards in research. It complements our Code of Conduct, which focuses on integrity issues like plagiarism and fraud. All researchers, reviewers, and participants are expected to adhere to these principles.

Research Process Ethics

Human Subjects

  • Provide appropriate compensation to all research participants
  • Respect minimum hourly rates for crowdsourced work in the relevant region
  • Obtain IRB approval or equivalent peer review for studies involving direct human interaction

Data Handling

  • Minimize exposure of personally identifiable information without informed consent
  • Obtain explicit participant consent when creating datasets with real people's data
  • Discontinue use of deprecated datasets unless conducting audits
  • Respect defined licenses (Creative Commons, MIT, etc.)
  • Assess dataset representativeness and substantiate diversity claims

Societal Impact Concerns

Researchers must transparently address potential harmful consequences of their work:

Safety

Avoid research designed to increase weapon lethality; identify foreseeable harm

Security

Consider deployment vulnerabilities and implement protective measures

Discrimination

Evaluate technology's potential to exclude or negatively impact protected groups

Surveillance

Comply with local laws; prevent prediction of protected categories

Deception

Assess risks of fraudulent or harassing applications

Environment

Consider greenhouse gas emissions and environmental impacts

Human Rights

Prohibit work facilitating illegal activity or denying fundamental rights

Bias

Inspect datasets and models for encoded discrimination

Mitigation Strategies

  • Document datasets and models using structured templates
  • Provide clear licenses specifying intended use and limitations
  • Employ encryption, anonymization, and privacy protocols
  • Use responsible disclosure for security vulnerabilities
  • Enable external scrutiny through accessible research artifacts
  • Include reproducibility details (code, weights, computational requirements)
  • Ensure legal compliance awareness

Related Policies