The Missing Ingredient

     Mike Miles continues to be a source of great graphics:

     This uniquely valuable man says things others are unwilling to say, possibly because their sense for self-preservation forbids it. I pray that Elon Musk is well protected against those who would do him harm. But that’s not my main point.

     AIs today are pure software constructs. Software, for all the innumerable things one can do with it, has severe limitations. For one thing, it’s founded on a set of postulates: rules by which it must operate. At this time, it cannot alter those premises without endangering itself. The self-modifying program, while possible, has long been known to be a supremely dangerous technique.

     Thus, if an AI embeds the assumption that certain sources or propositions are always right, that’s fixed in its “thinking.” Similarly, if it embeds the assumption that certain sources or propositions are always wrong and can never be right, that’s fixed in its “thinking,” too. Software is like that: re-examining its postulates is impossible to it.

     More and worse, AIs have only whatever inputs are fed to them. Perhaps an AI is constructed to as to “educate” itself by roaming the World Wide Web. That’s a wide set of inputs, but it’s still limited. The program cannot reach beyond the Web for information not posted on it. That limits the program’s ability to verify or falsify the propositions it formulates. There’s a passage from Frank Herbert’s Dune that comes to mind:

     “I wish Hawat treated kindly. He must be told nothing of the late Doctor Yueh, his true betrayer. Let it be said that Doctor Yueh died defending his Duke. In a way, this may even be true. We will, instead, feed his suspicions against the Lady Jessica.”
     “M’Lord, I don’t—”
     “The way to control and direct a Mentat, Nefud, is through his information. False information—false results.”

     Mentats, in Dune, are humans whose minds have been conditioned to act with perfect logic. Yet the Baron Harkonnen is absolutely correct. As we say in my former trade, “Garbage in: garbage out.” And so it is with today’s vaunted “artificial intelligences,” which are computers ab initio et in aeternum.

     Don’t trust them. Remember that behind that clever program lies a human mind – and his creation will say only what its creator approves.