Google's AI Principles

Source: https://www.blog.google/technology/ai/ai-principles/

Principles

  1. Be socially beneficial
    • Only proceed when likely benefits (in terms of "a broad range of social and economic factors") substantially exceed "the foreseeable risks and downsides"
    • "make high-quality and accurate information readily available using AI, while continuing to respect cultural, social and legal norms in the countries where we operate"
      • Interesting: the norms to respect are then dependent on the norms of the host
      • So what happens in China, or when those norms actually class with Google's own norms?
    • Evaluate when it is appropriate to make technologies available on a non-commercial bases (e.g. open source code)
  2. Avoid creating or reinforcing unfair bias
    • Interesting formulation "unfair bias": bias is in itself not necessarily unfair!
    • "We recognize that distinguishing fair from unfair biases is not always simple, and differs across cultures and societies."
    • "We will seek to avoid unjust impacts on people, particularly those related to sensitive characteristics such as race, ethnicity, gender, nationality, income, sexual orientation, ability, and political or religious belief."
  3. Be built and tested for safety
    • "avoid unintended results that create risks or harm"
    • testing in constrained environments and monitoring after deployment
  4. Be accountable to people
    • AI systems with "appropriate opportunities for feedback, relevant explanations, and appeal."
    • "subject to appropriate human direction and control"
  5. Incorporate privacy design principles
    • "give opportunity for notice and consent, encourage architectures with privacy safeguards, and provide appropriate transparency and control over the use of data."
  6. Uphold high standards of scientific excellence
  7. Be made available for uses that accord with these principles (evaluate if potential uses of the AI application accords with above principles)
    • What's the primary purpose
    • Is the technology unique or will it be generally available?
    • Scale: will this technology have significant impact?
    • Is Google "providing general-purpose tools, integrating tools for customers, or developing custom solutions"

Applications that Google does not pursue

  1. Technologies that "are likely to cause overall harm"
    • "Where there is a material risk of harm" (my emphasis)
    • unless "we believe that the benefits substantially outweigh the risks" (also interesting: so here harm is accepted after all, curious if there's examples )
  2. Weapons etc. whose primary purpose is to harm people
    • Google does work with military and is explicit about this: "cybersecurity, training, military recruitment, veterans’ healthcare, and search and rescue"
  3. "Technologies that gather or use information for surveillance violating internationally accepted norms."
    • So what type of surveillance is okay then?
  4. Tech that is in conflict with "widely accepted principles of international law and human rights"