Math Expectations

FILE:
REVI:
  - [20_22/10/15;10:58:26.00]:.
    - draft document.
  - [20_22/10/16;13:23:53.00]:.
    - integrate updates/edits.
  - [20_22/10/17;09:33:36.00]:.
    - some typo fixes; notation edits.

TITL:
   *Math Expectations*
   *As Applied to X-Risk Research*
   *By Forrest Landry*
   *Oct 15, 2022*.

ABST:
   Review of a process collision
   between what is expected around
   the notion of 'formality in reasoning'
   around categories of AGI/APS x-risk assessments,
   and/or what sorts of people are actually needed
   to do this kind of work.

TEXT:

> What is your background?
   > How is it relevant to the work
   > you are planning to do?

Years ago, we started with a strong focus on
   civilization design and mitigating x-risk.
   These are topics that need and require
   more generalist capabilities, in many fields,
   not just single specialist capabilities,
   in any one single field of study or application.

Hence, as generalists,
   we are not specifically persons
   who are career mathematicians,
   nor even career physicists, chemists,
   or career biologists, anthropologists,
   or even career philosophers.
   Yet when considering the needs
   of topics civ-design and/or x-risk,
   it is very abundantly clear
   that some real skill and expertise
   is actually needed in all of these fields.

Understanding anything about x-risk
   and/or civilization means needing
   to understand key topics regarding
   large scale institutional process,
   ie; things like governments, businesses,
   university, constitutional law, social
   contract theory, representative process,
   legal and trade agreements, etc.

Yet people who study markets, economics,
   and politics (theory of groups, firms, etc)
   who do not also have some real grounding
   in actual sociology and anthropology,
   are not going to have grounding in
   understanding why things happen
   in the real world as they tend to do.

And those people are going to need
   to understand things like psychology,
   developmental psych, theory of education,
   interpersonal relationships, attachment,
   social communication dynamics, health
   of family and community, trauma, etc.

And understanding *those* topics means
   having a real grounding in evolutionary theory,
   bio-systems, ecology, biology, neurochemestry
   and neurology, ecosystem design, permaculture,
   and evolutionary psychology, theory of bias, etc.

It is hard to see that we would be able to assess
   things like 'sociological bias' as impacting
   possible mitigation strategies of x-risk,
   if we do not actually also have some real
   and deep, informed, and realistic accounting
   of the practical implications of, in the world,
   of *all* of these categories of ideas.

And yet, unfortunately, that is not all,
   since understanding of *those* topics themselves
   means even more and deeper grounding
   in things like organic and inorganic chemistry,
   cell process, and the underlying *physics*
   of things like that.
   Which therefore includes a fairly general
   understanding of multiple diverse areas of physics
   (mechanical, thermal, electromagnetic, QM, etc),
   and thus also of technology -- since that is
   directly connected to business, social systems,
   world systems infrastructure, internet,
   electrical grid and energy management,
   transport (for fuel, materials, etc), and
   even more politics, advertising and marketing,
   rhetorical process and argumentation, etc.

Oh, and of course, a deep and applied
   practical knowledge of 'computer science',
   since nearly everything in the above
   is in one way or another "done with computers".
   Maybe, of course, that would also be relevant
   when considering the specific category of x-risk
   which happens to involve computational concepts
   when thinking about artificial superintelligence.

I *have* been a successful practicing engineer
   in both large scale US-gov deployed software
   and also in product design shipped to millions.
   I have personally written more than 900,000
   lines of code (mostly Ansi-C, ASM, Javascript)
   and have been 'the principle architect' in a team.
   I have developed my own computing environments,
   languages, procedural methodologies, and
   system management tactics, over multiple
   process technologies in multiple applied contexts.
   I have a reasonably thorough knowledge of CS.
   Including the modeling math, control theory, etc.
   Ie, I am legitimately "full stack" engineering
   from the physics of transistors, up through
   CPU design, firmware and embedded systems,
   OS level work, application development,
   networking, user interface design, and
   the social process implications of systems.
   I have similarly extensive accomplishments
   in some of the other listed disciplines also.

As such, as a proven "career" generalist,
   I am also (though not just) a master craftsman,
   which includes things like practical knowledge
   of how to negotiate contracts,
   write all manner documents,
   make all manner of things,
   *and* understand the implications
   of *all* of this
   in the real world, etc.

For the broad category of
   valid and reasonable x-risk assessment,
   that nothing less than
   at least some true depth
   in nearly *all* of these topics,
   will do.

:d4e
   > You claim that you are going to do a proof.
   > So therefore, we will have the expectation
   > that you will construct your formal proof
   > in the conventional notation/language
   > that other specialized mathematicians
   > are familiar with.
   > And therefore also the *expectations*
   > that you will _look_like_ a career mathematician;
   > ie; one that has established PhD credentials
   > at an accredited university specifically
   > for something like advanced topos theory,
   > and who also knows about all of the latest
   > work in that one field (or in general etc),
   > someone with multiple published papers
   > demonstrating some original theorems,
   > that has also been reviewed and accepted
   > by the larger mathematics community, etc,
   > at least some people in which
   > can vouch for you, as being a respected
   > mathematician in that community, etc.

As mentioned, we do not, did not, have not,
   claimed to be career mathematicians.
   So therefore, some of your expectations
   do not seem all that relevant here.
   Just tacitly assuming that we 'should' be,
   or are 'just mathematicians',
   seems like a mistake.
   We are *generalists* --
   and that is actually what we
   actually should be.

Moreover, in the case of constructing a proof
   applicable to final conditions of the real world,
   we cannot just assume we can restrict ourselves
   to just the deterministic domains of mathematics
     (eg; algorithmic computability).
   Any initial conditions (and other premises)
   set at the start of any sort of proof claims
   must also correspond empirically and *soundly*
   with the dynamics that actually will show up
   in practice, in the real situations,
   not just in the model.
   We are attempting to describe something about
   the limits of modeling, *in this application*.
   One thus notices that more than *just*
   a principled understanding of formal deductive reasoning
   is required to derive an impossibility result
   regarding actual machine classes in the world:.

Since we are dealing with code
      that is learned through
      statistical approximation/optimization methods,
      as stored as abstraction layers
      of a software/firmware/hardware stack,
      and computed/routed as message transmissions
      from/to peripherals,
      we must have a sound understanding
      of relevant domains in computer science.

Given the side-channel-effects of
      and noise interference across
      the signal transmissions
      between AGI/APS internals
      and their connected physical surroundings,
      a solid grasp of fundamental laws and limits of physics,
      of entropy and of information theory,
      and of error detection and correction methods
      is also required.

Given signal feedback loops
      between AGI internals and surroundings,
      a principled understanding of cybernetics,
      and of nonlinear (chaotic) dynamics
      is also required.

Since (digital) code is necessarily embedded
      as part of an assembled molecular substrate
      and computed/expressed through that substrate,
      this necessitates a principled understanding
      of molecular chemistry.

Since the existence and continued computation
      (actual non-halting) of that code,
      over the long term,
      depends on the reproduction of
      a compatible molecular assembly
      and on the preservation of
      that assembly's functional integrity,
      this requires a understanding of
      basic manufacturing processes
      and molecular assembly theory.

Since the rate
      of reproduction and preservation/survival
      of (the learned variants of) code
      held within a substrate
      is subject to (feedback from)
      outside environmental conditions,
      this requires a principled understanding of
      evolutionary developmental biology
        (as including function co-option
        and extended phenotypes)
      and of eco(toxi)cology.

Since artificial substrate configurations of
      (self-learning, generally functional) code
      would need and could fulfill different conditions
      for continued existence and growth,
      their supply of and demand for resources
      across an extended artificial ecosystem
      would come to differ from the old ecosystem.
      Careful analysis is needed
      to model inter-ecosystem supply-demand differences,
      of equilibria in (the absence of)
      inter-ecosystem resource exchanges
      and game-theoretical interactions,
      and of the 'offense-defense balance'
      in available (info/biochem/physical)
      attack/leakage vectors
      and protective/containment barriers.

Only at this point --
   having carefully compiled work
   in the above domains (and more)
   over the preceding 15 years --
   that we do have a comprehensive enough
   *empirical basis*
   of where to even *begin* constructing
   a formal and symbolized construction regarding
   the long-term ecosystem-terminating feedback dynamics
   that we are describing.
   Jumping straight to utilizing existing models
   without having some sense of which and why
   (including of idealized computation)
   would be undisciplined, and ill-advised.

Where/if we wanted to construct something
   that is focused on *only* just validity,
     (by the rules of logic alone,
     as a basis of mathematics),
   then we could/would have constructed
   an overly simplified toy model.
   The only problem is that then
   everyone would simply discount
   all of the work as "not relevant" --
   ie, does not correspond to the real world
   ie; that validity without soundness/relevance
   is simply useless, in this case.
   If the point was only just to show
   that we could write out some ideas
   in conventional notations and forms,
   then we will instead simply 
   hire a specialized mathematician.

Instead, we need to set up a careful claims
   that *soundly correspond with*
   the causal dynamics in the real world --
   requiring us to be disciplined about
   selecting only for properties and laws
   that are generally applicable
   and well-established through induction
     (ie; through multiple layers of
     iterative empirical observation/falsification).

:d9e
   > If you do not have the skill of
   > a formal mathematician,
   > then you cannot claim to be
   > "doing a proof", insofar as the
   > absence of complete knowledge of
   > all of the kinds of things that can
   > go wrong with reasoning,
   > have not been accounted for.

> People have noticed that even
   > having lots and lots of evidence
   > for a single specific claim
   > can sometimes have established
   > unexpected counterexamples
   > far deep out into the number field.
   > So therefore, if you do not do things
   > the way we expect, your work *will*
   > be discounted, without inspection,
   > as 'inherently incorrect' --
   > ie; not worth my time to review.

What we can do, and what we *are doing*,
   involves careful, formal reasoning,
   with lots of interacting key details,
   that all need to be tracked,
   lots of complexity with clear attention,
   tracking of all manner of assumptions,
   all types of combined complexity to organize.
   In that way, our work is more in the space
   of _formal_reasoning_ than anything else --
   so we describe it as "like math"
   to those people who are _asking_us_
   to *describe* our current and planned work.

When attempting to describe ourselves,
   and/or certain elements of our work,
   we will have to "borrow" whatever terms
   are available in the language of the person
   we happen to be talking to in that moment,
   to attempt to convey, for them, in their
   language and metaphors of understanding,
   as best as possible, what we are attempting
   to convey to them, at their request.

Just because we are not -- do not happen to be
   *just* career mathematicians (only and exclusively)
   does *not* mean that we are not
   exactly the right people
   to be doing *this* project --
   it may be the case
   that very few others, in the world,
   (in any single specialist discipline)
   will actually have the complete and specific
   total *range* of skills necessary
   to the actual depth necessary, in each,
   for them to actually have any hope at all
   of fully encompassing the needed thinking.

So therefore,
   not being just a mathematician
   and looking like one that has done just that
   is *not* a failing on our part.
   It is actually an indication of our rightness
   for this work.

So, yes, I am claiming that I do actually have
   *some* real depth of knowledge in *each*
   of the above mentioned fields of study
   (and a bunch of others I forgot to mention),
   also and inclusive of various topics in math.
   And there can be no doubt that having
   *all* of this is actually relevant,
   particularly when attempting to assess
   the x-risk implications of planned/future
   AGI/APS/superintelligence deployments.

:dcu
   > What/which university did you study at?
   > What were your specific focus topics of study?
   > Which branches of math did you specifically
   > study, and with who, and did you eventually
   > contribute original work to that field?

There is a lot that we could share with you
   detailing this.  Can you be a bit more specific
   as to what you actually need?

> Where are your co-authored papers detailing
   > some of your prior results/work published?
     > (Note, if it is not already in arxiv.org,
     > I will be less interested -- the best work
     > tends to go there, and I should be able
     > to find it using a quick internet search
     > that I am doing right this moment now.
     > Also, If you do not have a sufficient number
     > of such papers, I am letting you know now
     > that will not have a favorable impression).

> Also, has your work been peer reviewed
   > and was it also recognized and accepted
   > in a recognized math related journal?
   > Does anyone else cite your work?
   > What are the views of other mathematicians
   > also working in your chosen topic of focus,
   > regarding your work, ideas, proofs, etc?
   > Can at least someone in the math community
   > vouch for you, as being a respected
   > mathematician in that community, etc?

We are not claiming to be mathematicians,
   though we do recognize that we need to do
   careful formal work, and that it will maybe
   (probably) involve generally accepted types
   of math symbolism, in some form, eventually.
   The emphasis is expressive conceptual clarity,
   not the explicit terms or symbols used.

And we do "get", (understand and accept, etc),
   that you are needing at least *some* basis
   to establish some realistic assessment
   that we are reasonable, know something about
   what we are attempting to do, and are actually
   using the right sort of tools and techniques,
   in the right sorts of ways, and moreover,
   that we are more likely than most to actually
   be able to accomplish our stated goals, etc.

We do see that before listening to and/or
   attempting to be understanding our work,
   (and/or maybe recommending it or not),
   you will want to know that we have handled
   the difference between induction arguments
   in physics and the deductive arguments of math,
   can correctly apply formal reason,
   have examined all relevant priors,
   recognized and processed exceptions, etc.

We do have some prior formal work in the area
   of x-risk, which involved some technical analysis,
   and which was published to
   a number of academic forums.
   The main one which is closest to what you are
   specifically asking for is the (@ Dark Fire https://authorzilla.com/xzvZz/putting-out-the-dark-fire-constraining-speculative-physics-disasters.html) paper,
   which was a consideration of
   another class of x-risk entirely,
   as better understood in the area of physics,
   though some aspects of it were (and probably are)
   at least still somewhat controversial,
   (though only in the physics community).

The argument in that paper is a complex one,
   and it involves a lot of specific understandings
   in high energy particle physics and cosmology
   so as to constrain the probability of a *maybe*
   possible (world ending) collider incident.
   In regards to 'is this a good exemplar'
   of our type and category of thinking,
   I hesitate to recommend too strongly,
   due to that complexity (typical of that world).
   Insofar as it is based on my concepts,
   and insofar as we collaborated extensively
   while that paper was written, exploring idea, etc,
   then yes, this is a good example of my work.

Most physicists that I suggest this paper to
   tend to reject it for a variety of reasons.
   One of the more recent and unexpected reasons
   is due to my co-author, Anders Sandberg,
   although he is highly respected in the x-risk field,
   as it happens, his PhD (from Stockholm University),
   is in 'computational neuroscience', not physics,
   and thus he was not considered 'legit enough'
   by some, far too strongly opinionated people.
   Having my name on it is no help at all,
   at least in that respect, unfortunately.

I do not see this as a problem as Anders is also
   a generalist, in many of the same relevant ways,
   and moreover, someone whom I personally recognize
   as being a very careful, disciplined thinker,
   whose opinions I value, with multiple clear
   and insightful ideas.  The work is valuable,
   and insofar as I contributed content to that,
   I will send that over with these disclaimers.

We can also send over other examples of
   some of my prior formal reasoning work,
   in the areas of limits of epistemic methods,
   though I would rather not,
   given you are clearly looking for
   something specifically math oriented,
   and the topics tend to involve a lot of
   fairly specialized technical definitions,
   and outside of that context,
   are very likely to be misunderstood.

:dfn
   > If I happen to get around to reading
   > the papers you send over and suggest, *and*
   > it happens that I do not also recognize
   > and understand (@ note 1 #note1) the specific symbolism used,
   > I *will* judge you as being not a mathematician,
   > and therefore will reject any/all claims
   > of your formally proving anything at all,
   > *particularly* regarding AGI/APS x-risk.

> Also, it will not matter if your arguments are
   > provided with clear/exacting definitions, etc --
   > it has to actually be formalized in equations,
   > using generally the accepted symbolisms --
   > else I will not recognize it as 'actual math',
   > and therefore I will also judge that
   > you do not have a proof of your claims,
   > and that therefore your arguments
   > are invalid, inapplicable to AGI/APS, etc.

> And no, you cannot send over computer program code,
   > nor can any of your "formal" arguments be rendered
   > in any sort of 'technical language' other than
   > what I will recognize as (judged to be) math.
   > Though I am totally sure that I will be able
   > to understand anything that you send (@ note 2 #note2), so
   > no worries about that (I am really smart).

Sorry, we cannot help you.
   Nor do we even want to, anymore.

~ ~ ~

:djg
   > What happened?
   > What is the overall assessment?

The pattern seems to be the following sequence:.

- 0; we have made well structured observations
     which we think we could probably formalize.

- 1; we presented our work as towards doing a 'proof',
     as a kind of deductive argument with inductive relevance,
     which implied, to most people, some kind of 'math';
     and so;.

- 2; where as based on our own given descriptions
     of what we were trying to do,
     some people who needed/wanted to evaluate
     our "suitability" for this sort of work,
     elected to send someone
     with an educational background in mathematics
     to evaluate us and our claims.
       - this would have made sense, of course,
       *if* such a person had the more relevant skills
       to make such an assessment,
       even though in this case,
       it did not help.

- 3; that person, asked to evaluate us,
     had their own very strong expectations
     that we would be 'just like' every other
     practicing career mathematician who was
     doing formal proof work.
       (As such, she asked about our basis of argument,
       our prior work in formal papers, etc).

- 4; And where insofar as we did not look like
     a long term practicing career mathematician
     (who has the time to specialize in that topic only);.

- 5; that/therefore; we failed out,
     with respect to their assessment,
     insofar as we did not match the projected expectations
     and prior strong opinions of the assessor.

The unfortunate aspect is that
   the assessment is clearly wrong,
   not insofar as being narrowly correct
   about our not being career mathematicians,
   but in the space of whether we can do
   meaningful work regarding AGI/APS
   and inherent terminal x-risk 'proofs'.
   It was based on expectations/opinions
   that simply (mistakenly) do not apply,
   and moreover, are not the right sort of
   expectations/opinions needed for assessment.
   Better discernment is regarded regarding
   that which is inherently generalist work,
   vs that which is inherently specialist work,
   and we are definitely the former,
   which is much harder to validly assess.

This is in itself to be expected,
   since there are very few generalists
   in the world today, given the ongoing emphasis
   of nearly every academic in every institution,
   to specialize, as designed to make "progress"
   (in the modern world, in STEM, market, etc).
   Anyone attempting to evaluate our work
   is going to have a hard time
   finding the right mix of discernments
   to be able to tell that what we are doing
   is just inherently better than that for which
   they can presently have a concept for.

:note1:
   ...and understand...

The risk here is that 'understanding'
   can easily, especially in adverse opinion,
   be conflated with 'agreement' and thus with
   issues like 'validity' and 'applicability'.
   These are not the same, and yet she has given
   no assurance, and in fact, many dis-assurances
   that she would actually be "reasonable"
   and thus actually, in practice, distinguish
   personal opinion and bias from logic,
   despite claims to the contrary, etc.

Maybe she does not actually *both* understand,
   *and* agree with all of the claims made?.
   Is that a true failing of the work itself,
   or simply a limit of the reader capability?

Maybe she could find some easily correctable
   nit-picks -- how are we to know that she
   was not simply creating them, as a way of
   discounting opinions/ideas that she does not
   want to see established as 'proven'?
   How are we to know if the reviewer 'observations'
   are fair, reasonable, applicable, correct,
   if she simply does not mention them at all,
   but simply rejects the entire work
   without comment, as it seems very likely
   that she is predisposed to 'just reject',
   without any other valid reason, anyway.

Also, just about anyone can come up with
   infinite synthetic reasons for
   disagreements with definitions, etc.
   There may be specific reasons why a given
   definition is constructed the way it is,
   that is not obvious on 1st inspection.
   What if it happens that this disagreement
   is based on their failing to understand,
   or meet their own priors, expectations, etc.

:note2:
   The implied "vast familiarity with math",
   and that this reviewer later indicates
   that they "had read analytical philosophy
   papers of a some other specific author"
   with the implication that they would therefore
   "for sure correctly read and understand"
   a much different analytical philosophy work,
   despite our repeated warnings about
   the time and care needed to actually understand
   the actual specific meanings of the terms used,
   (as based on also a number of preceding papers),
   seemed a bit aggressive and overly presumptive.

We ended up with the very distinct impression,
   (via numerous video gestures, face expressions, etc)
   that they were clearly implying/claiming, to us,
     "that they could not (would not ever) possibly
     fail to understand, completely and totally,
     absolutely anything and everything
     that we could/would even possibly do, ever" --
   (ie; how could "three men" ever have anything
   original to say to someone like them).
   This goes far afield of issues of agreement,
   let alone of simple logical validity.

This strong feeling clear absence of any sense
   of any type or aspect of intellectual humility
   and/or of actual emotional communicative honesty
   (ie; our experience of the review) leads us to have
   significant doubts as to the prevalence of bias,
   and thus, of the clear in-applicability of
   this particular reviewer, has having as any kind of
   valid and justified opinion of our intended work.

Unfortunately moreover, as we later discovered,
   once we had a chance to review the reviewer bio,
   it turned out to clearly be the case that
   this reviewer themselves does not even "fit"
   the 'specialized/career mathematician' persona
   that they were expecting that we should have,
   and that they ware projecting for themselves.
   This frankly makes their self presentation,
   _as_if_ knowing and being able to evaluate
   and thus *judge*, all manner of 'math stuff',
   not to mention any disciplined philosophy
   and/or policy recommendations for AGI work
   look even very much more presumptive and biased.

The bio profile shows little mention of math,
   and much more around *entrepreneurship*,
   machine learning, and public communication --
   all topics that indicate clearly entangled interests
   and the strong possibility of motivated reasoning.
   Thus, our initially somewhat charitable impression
   of this reviewer *maybe* having a "somewhat less"
   judgemental bias, for illegitimate reasons,
   became very much stronger, and less good.

Overall, the clearly presumptive attitude
   and the level of intellectual entitlement
   left us with a very strong impression of
   a very strong and illegitimate negative bias
   and of a prejudgement of distaste/distrust,
   for reasons having nothing to do with
   the work itself and/or of our capability
   as persons.

:menu

If you want/need to send us an email,
   with questions, comments, etc, 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: