Nine Points of Insanity

FILE:
  - [20_22/10/15;20:14:44.00]:.
    - 1st draft.
  - [20_22/11/03;13:21:03.00]:.
    - seperate out as own post.
    - add own context lead and partial conclusions.
    - add section marks.
  - [20_22/12/27;00:43:15.00]:.
    - point edits.

TITL:
  *Nine Points of Collective Insanity*
  *By Forrest Landry*
  *Oct 15th, 2020*.

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.

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)).

TEXT:

- 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.

- where a prediction of the trend of the future:.

- 1; that engineers desire to explore
    the outer limits of AI functionality.
      - as per (@ note 1 #note1).

- 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).

- 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.

: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,
    *and_also* 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.

: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.

: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.

:dk4
    - 7; that this overall 'factoring out'
    of people has a few stages of its own:.

- 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).

- that then the engineers building
      the capability to build capability
      automate themselves away,
        (but *not* before also developing
        automation of maximally efficient
        optimized marketing and sales processes,
        since those sales/marketing people
        tend to also be *very* expensive
          (and opinionated temperamental
          narcissist artist types too)).

- 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).

: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.

: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?.

- 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.

: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 *not* 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.

:dwc
      - that so many succession plans
      turn out not to work:.
        - that the overall complexity
        of the overall process
        *and* 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,
        *and* 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.

:dzs
    - as leading to the sub-sequence of:.

- 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).

- 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.

- c; where after sufficient cycles of the above;
      that *only* the 'fully self managing' systems
      continue to endure in time.

- 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.

: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.

- as the "nine points of collective insanity".

~ ~ ~

:note1:
  - that engineers desire to explore, due to:.
    - geekery.
    - desire for product sales.
    - in-group prestige.
    - out-group conflicts.
    - drive R-and-D labs.

: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.

: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 *compose* 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.

: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'.

:menu:

If you want/need to send us an email,
   with questions, comments, etc, on the above,
   and/or on related matters, use this address:

ai@mflb.com

(@ Mode Switch com.op_mode_tog_1();) + (@ View Source com.op_notepad_edit_1();)

Back to the (@ Area Index https://mflb.com/ai_alignment_1/index.html).

LEGA:

This document will not be copied or reproduced
   outside of the mflb.com presentation context,
   by any means, without the expressed permission
   of the author directly in writing.
   No title to and ownership of
   this or these documents
   is hereby transferred.

The author assumes no responsibility
   and is not liable for any interpretation
   of this or these documents
   or of any potential effects and consequences
   in the lives of the readers of these documents.

ENDF: