Considering the Risk of Deceptive Artificial Intelligences

27 July 2023

• What is Artificial Intelligence
  • Artificial Intelligence is an umbrella term
    • LLMs
    • Genetic Algorithms
    • Optimization Problems
    • Games and Game Theory
  • Current Artificial Intellengeces are not autonomous
    • Require training and/or supervision
    • limited domain of action
  • Theory of Mind
    • What we call “AI” currently implements functions of a human mind
    • No current AI system exhibits a theory of mind
    • Humans are excellent at theorizing about internal mental states
• What is deception
  • Deception is an interaction in which one party (the deceiver) attempts to create a counterfactual mental state in another party (the deceived)
  • Current AI systems are fundamentally unable to deceive
    • Lack a theory of mind
    • Lack autonomy of action
• SO what are the real risks here?
  • Humans attribute human attributes to current AI systems
    • autonomy lies with the trainer/operator of the machine intelligence
    • Humans atribute meaning to interaction, machine intelligence treats interaction as raw data
  • Psychology of human-computer interactions
    • Fictional accounts and the myth of machine intelligence
    • The myth of measurement
    • Past performance is no indication of future returns
    • The wisdom of crowds and social media
• Natural Stupidity Beats artificial Intelligence every time
  • The data sources are hidden by the AI interface
  • Unlike algorithms, AIs are not self-explanatory
  • Human organizations have hidden autonomy and agendas
  • The medium is the message
    • AI systems provide a method of communication of complex results to simple people
    • humans tend to trust human-like things
    • AI as a tool of corporate feudal plutocracy
    • user-friendly is not the same as “friendly human”
  • People like to fool and flatter themselves

This document was translated from L^AT_EX by H^EV^EA.