Documenting the Coming Singularity

Wednesday, July 11, 2007

Overcoming Public Misconceptions About AI

We are all, to one degree or another, influenced in our beliefs by the popular culture in which we exist. For example, what the public may believe, or think it knows, about artificial intelligence, has been influenced and shaped by science fiction in both movie and book form. For this reason it seems to be important, for the sake of informed and intelligent discourse, that laymen become disabused of the more egregious and even absurd ideas that have embedded themselves into our consciousness.

To that end, I am reposting here a short piece from a very detailed analysis of "Creating Friendly AI," which was posted on an excellent reading list at Accelerating Future. The short piece is titled "Movie Clichés about AIs." I think you'll find it entertaining as it points out some of the more untenable features of AIs as portrayed in movies.

Movie cliches about AIs

  • All AIs, no matter how primitive, can understand natural language.
    • Corollary: AIs that comically or disastrously misinterpret their mission instructions will never need to ask for help parsing spoken English.
  • No AI has any knowledge about blatant emotions, particularly emotions with a somatic affect (tears, frowns, laughter).
    • Corollary: AIs will always notice somatic affects and ask about them.
    • Double corollary: The AI will fail to understand the explanation.
  • AIs never need to ask about less blatant emotions that appear in the course of ordinary social interactions, such as the desire to persuade your conversational partner to your own point of view.
    • Corollary: The AI will exhibit the same emotions.
    • Double corollary: An evil AI will feel the need to make self-justifying speeches to humans.
  • All AIs behave like emotionally repressed humans.
    • Corollary: If the AI begins to exhibit signs of human emotion, the AI will refuse to admit it.
    • Corollary: Any evil AI that becomes good will gradually acquire a full complement of human emotions.
    • Corollary: Any good AI that becomes evil will instantly acquire all the negative human emotions.
    • Corollary: Under exceptional stress, any AI will exhibit human emotion. (Example: An AI that displays no reaction to thousands of deaths will feel remorse on killing its creator.)
  • AIs do not understand the concept of "significant digits" and will always report arithmetical results to greater-than-necessary precision.
    • Corollary: An AI running on 64-bit or 128-bit floats will report only four more digits than necessary, rather than reciting fifteen or thirty extra digits.
  • AI minds run at exactly the same rate as human minds, unless the AI is asked to perform a stereotypically intellectual task, in which case the task will be performed instantaneously.
    • Corollary: The reactions of an overstressed AI undergoing an Awful Realization will be observable in realtime (the Awful Realization will not take microseconds, or a century).
Clichés that are actually fairly realistic:
  • A newborn AI can take over the entire global computer network in five minutes. (Humans stink at network security - it's not our native environment.)
  • A spaceship's on-board AI can defeat any crewmember at chess. (The amount of computing power needed for decent AI makes Deep Blue look sick.)
Singularity & The Price of Rice is updated daily; the easiest way to get your daily dose is by subscribing to our news feed. Stay on top of all our updates by subscribing now via RSS or Email.