35-Nine-Points-of-Collective-Insanity.html

<!DOCTYPE html>
<html lang="en">
<head>
  <!-- Basic Meta Tags -->
  <meta charset="UTF-8">
  <meta http-equiv="X-UA-Compatible" content="IE=edge">
  <meta name="viewport" content="width=device-width, initial-scale=1.0">

  <!-- SEO Meta Tags -->
  <meta name="description" content="Comprehensive AGI Risk Analysis">
  <meta name="keywords" content="agi, risk, convergence">
  <meta name="author" content="Forrest Landry">
  <meta name="robots" content="index, follow">

  <!-- Favicon -->
  <link rel="icon" href="https://github.githubassets.com/favicons/favicon-dark.png" type="image/png">
  <link rel="shortcut icon" href="https://github.githubassets.com/favicons/favicon-dark.png" type="image/png">

  <!-- Page Title (displayed on the browser tab) -->
  <title>Comprehensive AGI Risk Analysis</title>
</head>
<body>
  <p>
  TITL:
    <b>Nine Points of Collective Insanity</b>
    <b>By Forrest Landry</b>
    <b>Oct 15th, 2020</b>.
  </p>
  <p>
  ABST:
    Description of the 'slippery slope model'
    of how it happens; the way in which AGI/APS
    takeover begins, as the stage immediately
    prior to the power-seeking behavior, and
    other downstream problem implications.
  </p>
  <p>
  PREF:
    - where acronyms list:.
      - AI: Artificial Intelligence (ie, Narrow AI).
      - AGI: Artificial General Intelligence.
      - APS: Advanced, Planning, Strategically aware Systems.
      - NAI: Narrow AI (note the (@ distinction https://mflb.com/ai_alignment_1/si_safety_qanda_out.html#pg_3_r)).
  </p>
  <p>
  TEXT:
  </p>
  <p>
    - where considering how proto-AGI/AI systems,
    when in interaction with humans
    are shaped into more advanced AI/AGI systems.
      - ie; how 'human to human' interactions
      bootstrap the 'human to AI' interactions
      into a collection of AI to AI interactions.
      - as also about how narrow AI becomes general AI.
  </p>
  <p>
    - where a prediction of the trend of the future:.
  </p>
  <p>
      - 1; that engineers desire to explore
      the outer limits of AI functionality.
        - as per (@ note 1 #note1).
  </p>
  <p>
      - 2; that business owners desire
      to expand the utility generality
      so that they can market, sell, and extract,
      more value.
        - as per (@ note 2 #note2).
  </p>
  <p>
      - 3; that venture capitalists and investors
      desire to maximize profits,
      and so drive/compel CEO and executive team
      to direct/compel the engineering groups
      to make more and more powerful/general AI,
      to/towards business process optimization
      objectives/goals.
  </p>
  <p>
  :dcu
      - 4; that engineers,
      where continually having to make
      increasing changes to the AI systems in use,
      and overall increase the AI capability,
      as directed by owners/investors and
      overall startup investment business culture;
      that they end up in a kind of multi-polar trap
      to develop both stronger and stronger
      AI systems/capabilities,
      <b>and_also</b> to automate the increase of
      the design and deployment process
      for all of these capabilities,
      inclusive of the capability
      to deploy more capacity,
      to design the design of more capacity, etc.
  </p>
  <p>
  :deq
      - 5; that business owners,
        (where seeing reduced costs
        associated with people/engineering
        due to increased design and deployment
        process automation),
      elect to push harder/farther/faster
      to factor out the expensive humans sooner,
        (ie; by reducing the engineering team
        and system monitoring/oversight teams).
        - note; for somewhat similar remarks,
        refer to (@ "How AI Fails Us" https://carrcenter.hks.harvard.edu/files/cchr/files/howaifailsus.pdf).
          - as also arguing that tech use,
          over decades, has led to general reductions
          in economic output overall (human wellbeing),
          even though individual investors
          may have also (selectively) become richer.
  </p>
  <p>
  :dgw
      - 6; that venture capitalists
      then see dramatically increased capability
      to recognize, design and deploy
      optimized business plan process.
        - ie; where less and less invested capital
        results in more and more
        capital return on investment
          (ie; in the form of increased efficiency
          of all business/market marketing and sales
          of AI services
          as an extractive money making process).
      - then/that/therefore, the VC
      start using that optimizing power
      (of the now near fully general AI)
      to predict, and even more fully automate,
      the design of further and ever more efficient
      market value investment strategies
      that automatically return
      more and more value on investment,
        (ie; with fewer and fewer people
        involved anywhere in the process).
        - as basically the Google, Amazon, etc,
        business models, (insofar as they attempt
        to factor out all human involvement,
        as much as possible, in favor of profit),
        along with the entire
        crypto currency movement, and/or
        the emphasis on DAO/Etherium, etc,
        insofar as they factor out
        groups of people altogether.
  </p>
  <p>
  :dk4
      - 7; that this overall 'factoring out'
      of people has a few stages of its own:.
  </p>
  <p>
        - where/first it begins with
        the customer service people --
        they are automated away
          (since that is expensive,
          and no one wants to do it --
            everyone hates working with
            abusive customers anyway).
  </p>
  <p>
        - that then the engineers building
        the capability to build capability
        automate themselves away,
          (but <b>not</b> before also developing
          automation of maximally efficient
          optimized marketing and sales processes,
          since those sales/marketing people
          tend to also be <b>very</b> expensive
            (and opinionated temperamental
            narcissist artist types too)).
  </p>
  <p>
        - then, when the VC --
        seeing that the executive team
        is no longer needed --
          (since everyone else
          has already been factored out
          of the overall company structure)
        that they (the VC folk) notice
        that they also have the capability
        to make fully autonomous corporations
          (maybe using some of the newer
          'Decentralized Autonomous Orgs',
          with a heavy crypto emphasis).
          - (note; for parallel remarks, refer to
          (@ Production Web scenario https://www.lesswrong.com/posts/LpM3EAakwYdS6aRKf/what-multipolar-failure-looks-like-and-robust-agent-agnostic)
          by Andrew Critch,
          both the engineers and CEO version,
          as examples).
  </p>
  <p>
  :dp6
      - where all of the people
      in the entire org-tree
      have been factored out;
        (of what was once a "company", a corporation)
      then that the overall situation
      is effectively "passive income on steroids"
      for the VC, so much so, to the point
      that even the VCs themselves,
      are no longer even having to evaluate
      startup business plan proposals at all --
      they already have the perfect evaluative tools,
      and the ability, with their 'owned' AI,
      to identify the 'best possible versions'
      (the most profitable, and extractive versions)
      of any possible business plan(s),
      and therefore also, with the automation,
      the ability to execute on that plan
        (ie; now fully optimized for
        return on investment, via AI,
        without even needing any people at all,
        in/at the new corporate inception).
          - as also not at all incorporating
          any sort of awareness of human needs.
          - as a bit like the 'two masters problem',
          where wall street investors are seeking only
          capital over capital gains,
          regardless of any secondary impacts
          such actions may have in/on the larger world.
          - also see (@ note 5 #note5) re 'metrics'.
        - that the AI systems are increasingly
        given as much data and knowledge
        about real world business, situations,
        background, aspects of practice, legal, etc,
          (not to mention, a full suite of human bias
          towards competitive seeking of power, etc,
          whic will be/become important later).
        - that corporations spawn more corporations,
        all fully automated, (no humans necessary)
        where even the 'what corporations to spawn'
        is itself fully automated, self-selecting, etc,
        on the basis of ever increasingly general AI.
  </p>
  <p>
  :dqq
      - 8; where at some point, maybe far in the future;
      that the entire process of 'economics',
      of purchasing and selling at volume,
      becomes more or less fully automated
      within all of these corporations themselves.
        - as accounting for the increasingly
        substantial majority of all large scale,
        large volume bulk aggregate economic activity;
        that even the VCs themselves
        are barely involved (no longer needed).
        - that they also have ended up (eventually)
        getting themselves factored out
        by the (once) billionaires --
        the ones who were the sources
        of the investment portfolio funds.
          - who needs an investment advisor
          and/or a hedge fund manager
          when the increasingly intelligent AI
          can get more out of the market
          than any predictive person could.
          - ?; with fully automated and optimized
          rapid trading AI software
          available to anyone
          with sufficient money to buy it,
          who needs hedge funds,
          VC investment funds, etc,
          to optimize and increase
          ones economic holdings?.
  </p>
  <p>
        - that the net effect ends up being
        that fewer and fewer actual people (users)
        have any degree of significant potential
        for any sort of combined economic impact
          (aside from (maybe) a very few
          world scale global trillionaires),
        and the total share
        of the overall world total economy
        that has anything to do with
        the real needs of people,
        since the total value
        of their 'investment ability'
        is monotonically decreasing,
        and at an ever increasing rate.
  </p>
  <p>
  :dt8
      - 9; where eventually
      that the handful of world mega-trillionaires,
        (who are no longer needing
        any other lower stages of humanity
        to select and 'operate' and/or 'implement'
        (ie; 'execute' or 'execute with')
        any part of 'their holdings',
        or market/political/world governance plans),
      discover --
        (where and as they grow old/older) --
        (to their perpetual surprise) --
      that it is <b>not</b> actually so easy
      to 'hand off' the administration
      of their fortunes
      to their preferred next of kin.
        - that the inter-generational transfer of power
        is far more (inherently) complex
        than it seemed
        at any time prior to the attempt;
        so much harder, to an extreme --
        much more so than acquiring and developing
        the power and holdings in the 1st place,
        etc.
  </p>
  <p>
  :dwc
        - that so many succession plans
        turn out not to work:.
          - that the overall complexity
          of the overall process
          <b>and</b> the sheer number and depth
          of the secret keys used to
          access and administrate and oversee
          all of the complex and multi-faceted
          money making process, automations, etc,
          <b>and</b> the perception of significant risk
          of multiple modes of possible exposure/loss
            (that the plutocrat notices
            as inherently associated with
            their saying _anything_ to _anyone_
            about _any_aspect_ of their "system",
            let alone the risk
            of actually giving anyone access controls),
          all combine in such a way
          the each of the current incumbent generations
          will delay "teaching" the next generation
          how to be/become the inheritor generation,
          to the maximum extent possible.
            - that a complete and consistent knowledge
            of all that has happened, which is relevant,
            over these long years,
            and/or of all of the varied different kinds
            of practical knowledge
            that defines how to understand and operate
            _the_system_,
            gradually becomes increasingly lost.
  </p>
  <p>
  :dzs
      - as leading to the sub-sequence of:.
  </p>
  <p>
        - a; the next generation inheriting
        has less knowledge/skill
          (and for sure less exposure
          to the details and key history)
        than their fathers,
        and so becomes less effective
        at any of the key regulation and control actions.
          - and also (@ note 7 #note7).
  </p>
  <p>
        - b; that the general automation
        has to either assume more self sovereignty,
        OR it becomes decayed/broken,
        no longer known or maintained,
        and no longer relevant to the story.
          - ie; failed systems,
          no matter how much anyone
          may be using and depending on them
          simply do not get fixed,
            because no one knows how.
          - that people (even the ultimate rich,
            though they may say that they are the
            "ultimate governors of all things")
          become increasingly ineffective and incapable
          as one generation replaces the next.
  </p>
  <p>
        - c; where after sufficient cycles of the above;
        that <b>only</b> the 'fully self managing' systems
        continue to endure in time.
  </p>
  <p>
        - d; that eventually full market separation occurs.
        - as that the once fully human mediated market
        has become completely machine managed market.
          - ie; as a fully virtualized market,
          involving only digital currency, etc.
          - that the 'human involving marketplace'
          and the 'machine only process marketplace'
          end up having no relationship to one another.
          - that humans simply can no longer provide
          anything that the machine market
          needs or wants.
            - labor and intelligence are not wanted,
            and interest in human reproductive process
            has never been even on the table
            as at all "interesting" for/to machines.
  </p>
  <p>
  :e4a
      - as having the outcomes of:.
        - that these overall effects can only be known
        in overview.
          - where from the perspective of any of
          the participants, these overall trends
          are not at all obvious or clear, except
          insofar as the individual participants
          are directly affected, in their choice making,
          strategy evaluation, next moves, etc.
        - that prior 'human to AI' interactions
        have effectively/functionally implemented
        a bootstrapping of all kinds of possible
        'AI to world' interactions,
        and how these 'AI to world' interactions,
        in turn, set the context and future
        for all manner of 'AI to AI' interactions.
          (or maybe 'AI to AGI' and/or maybe AGI to AGI,
          etc)
          - where those AI/AGI to AI/AGI interactions
          are either exogenous
            (if multiple AGI come to exist)
          or endogenous
            (if a world singleton AGI is the result).
        - that (either way, one or many advanced AI)
        there are now near-threshold-AGI systems
        in interaction with each other and the world.
  </p>
  <p>
    - as the "nine points of collective insanity".
  </p>
  <p>
    ~ ~ ~
  </p>
  <p>
  :note1:
    - that engineers desire to explore, due to:.
      - geekery.
      - desire for product sales.
      - in-group prestige.
      - out-group conflicts.
      - drive R-and-D labs.
  </p>
  <p>
  :note2:
    - that engineers, and their optimization methods,
    broadly select for functionality
    that is adaptable to/for achieving
    an expanding set of (profitable) goals.
    - that this happens by means of (for example):.
      - outperforming humans at productive tasks.
      - planning in pursuit of goals.
      - strategic awareness.
  </p>
  <p>
  :note5:
    - there are other problems with 'metrics'
    inherent in all kinds of optimization processes
    that are associated with the fact of floating
    point (continuum) based measurement in itself.
      - for example, there is a 'knapsack' type problem
      when considering how to <b>compose</b> any sort of
      singular unitary feedback process, for ML,
      out of all of the 'intake sense data',
      using any sort of formulaic/algorithmic process,
      as distinct from, yet in addition to,
      all of the data collection issues, etc.
  </p>
  <p>
  :note7:
    - that many people
    (particularly outside AI/tech communities)
    are already struggling
    to be keeping up with, and adapting to,
    the currently deployed digital technology.
      - as falling outside the heuristics
      that evolution has equipped us with;
      which is itself already far behind
      recent improvements in AI research and
      development over the last years.
    - which is also not even inclusive of
    adapting to widespread developer use
    of current large transformer models,
    which in many cases
    are still uni-modal or 'single domain'.
  </p>
</body>
</html>