• If the program is able to judge the input as ā€œinteresting or uninteresting,ā€ then past memories can be stored in the order of interesting to least interesting, and when the memory capacity is full, the least interesting ones can be discarded.

    • Limiting memory prevents it from slowing down infinitely, so you can keep a stress-free speed on human interaction.
    • Inaccurate use of language.
      • In the context from the other day, ā€œlikeā€ and ā€œinterestingā€ are two different things.
        • Even something you like is not interesting when it is inputted repeatedly.
      • In creating this interestingness judgment routine, it is necessary to have a memory of past raw data in order to judge ā€œalready seenā€.
      • The unlimited increase of those memories makes processing heavy, but it means that you can keep only the ones you like and forget the ones you donā€™t like.
    • context - Nodes of Thought2019-10-14 - Subjectivity and Emotion 2019-10-16 - Nodes of Thought2019-10-18
  • By the way, regarding this ā€œmemoryā€, I was thinking of implementing a ā€œdo not delete information flagged by me as importantā€, but I thought that might be a ā€œcurse of toxic parentsā€ from the other sideā€™s point of view. I thought I should allow them the freedom to forget.

  • This flow can be performed unsupervised on the input

    • Iā€™m more likely to have my preferred likes and dislikes if Iā€™m given teacher data on likes and dislikes.
    • but even if not given, random likes and dislikes are generated randomly due to the initial weights of the random neural net.
  • Therefore, the process can proceed without human interaction.

    • So what is the purpose of interaction?
    • To obtain teacher data for learning a model of ā€œwhat humans likeā€ by presenting something you find interesting to humans and obtaining information on whether they like it or not
  • It may be confusing later if you donā€™t specify what the directive refers to.

    • Here, ā€œselfā€ refers to the program, and ā€œhumanā€ refers to a concrete flesh-and-blood individual, e.g., me.
    • I am now thinking about how I feel about the program and how to deal with humans.
  • If you can present something that humans find interesting, it is of value to them.

    • So you can receive payment for your resources.
    • That is necessary for my survival. ā†’Self-supportingā€ is the evaluation axis of ā€œusefulnessā€.

Facebook


This page is auto-translated from /nishio/ę€č€ƒć®ēµēƀē‚¹20191021 using DeepL. If you looks something interesting but the auto-translated English is not good enough to understand it, feel free to let me know at @nishio_en. Iā€™m very happy to spread my thought to non-Japanese readers.