[FRIAM] more structure-based mind-reading

Marcus Daniels marcus at snoutfarm.com
Wed Sep 7 12:28:22 EDT 2022 

Meta, Google, etc. may have relatively surface behaviorist measures now, but eventually they'll be mining all the media streams they provide to peopple, all scholarly articles, everything.   It won't be shallow and behaviorist.
________________________________
From: Friam <friam-bounces at redfish.com> on behalf of glen <gepropella at gmail.com>
Sent: Wednesday, September 7, 2022 9:34 AM
To: friam at redfish.com <friam at redfish.com>
Subject: Re: [FRIAM] more structure-based mind-reading

But doesn't this answer your question of "why model some complex (if meaningless) phenomena if it can driven toward something less complex?" The answer is: because Meta, Google, etc. take only surface, shallow, behaviorist measures. Depth search can reveal (perhaps overly) complex structures the breadth search can't take. If the behaviorists are right, ontologically, and everything significant *is* written on the surface, then yes, the power is already gone.

But if the behaviorists are wrong, depth search, complex models, will be a power reserve not available to the identifiably Bad Faith behaviorists who think they can simulate invested community members at will.

Being a skeptical agnostic, my best guess (or hope) is that both depth and breadth have power. Perhaps what "we are for" is deep modeling. The whole rigmarole of denying the Hard Problem, Mysticism, the "inner life", etc. may well circumscribe the eventual value of the unique individual. It need not be any kind of metaphysical/biological thing, either. Once some future bot, a descendant of LaMDA, does possess "sentience", that bot will also be a deep/complex modeler.

Your question can be reformulated: Do we need deep/complex modelers? And it's just a restatement of the Hard Problem.

On 9/7/22 08:09, Marcus Daniels wrote:
> There's an energy efficiency benefit of biological systems over silicon ML, but we make up for it by driving around in cars and using huge amounts of energy crushing rock to make our roads and buildings.   I maybe know a hundred people, but these large-scale systems at Meta, Google, etc. can study millions of people over hundreds or thousands of hours of measurement.   It might be possible someday to grow a computing system from a seed, but for now I think you are right.   I don't see any point in talking about ceding power, it's already gone.   Best case scenario is joining a hive.
>
> -----Original Message-----
> From: Friam <friam-bounces at redfish.com> On Behalf Of glen
> Sent: Wednesday, September 7, 2022 8:00 AM
> To: friam at redfish.com
> Subject: Re: [FRIAM] more structure-based mind-reading
>
> Right. I agree with the sentiment that something like "agency" exists inside a trained LLM, and that the ontogeny of that LLM has historicity. That balled up pocket of "expertise" is not significantly different from the balled up expertise in an expert human like Murray Gell-Mann. But what *is* different is the asymmetric power held by the ontogenic process.
>
> Like GitHub or OpenAI literally profiting off the efforts of open-sourced code or scrapable manually generated art, you *cannot* withhold the source(s). The construction (and execution) of such models is and will be limited to those of us with the *power* to do so. I.e. Google, Amazon, Microsoft, etc. Spinning up T0pp <https://huggingface.co/bigscience/T0pp> on your own machine requires you to have enough "disposable" resources to instantiate that machine.
>
> Of course, it's reasonable to throw up one's hands and give people like Thiel a pass when he gives $15 mil to an evil-doer like JD Vance. Thiel has the right to do whatever he wants because he has the *power* to scrape all that data and use it to create a new Monarchy backed by Oligarchy. What ya gonna do?
>
> The above *should* demonstrate why the ontogeny of Dall-E is *not* the same as a horseman (or blacksmith or brewer or whatever) who continues to practice that obsolete artisanry. That's a false equivalence. The re-generation of Dall-E (into something like Stable Diffusion) is not like smithing or horsemanship. It's permanently a task that can only be done by those who have the power to do it.
>
> So, sure, we can cede the power to the Thiels as you argue. But let's be prepared to be the serfs we'll be when that happens.
>
> On 9/7/22 07:38, Marcus Daniels wrote:
>> It's like a horseman that insists on practicing his trade despite the existence of cars.    There's some market for it, but if that market got bigger due to some world event, people would easily relearn it.   It is funny to hear people preach about the importance of being literate and informed.  But, when a multi-billion parameter deep neural net reads many texts and can synthesize new ones, it is regarded I in a different way.    If someone says, "I wrote a novel", I can and do think "Ok, she offers to have me review the evidence of her latent state encodings."   The bigger question is what does human cognitive life look like after AI starts to outperform us?  What are we for?
>>
>> Deep fakes are going to get better and better, and really the only defense I can see is withholding the sources that make it possible.
>>
>> My point about prescriptive vs. descriptive models is about the appetite for them, rather than whether they are good or not.   There is little political or commercial value in aspiring to greater things if one can get by on demagoguery and repetitive advertising.
>>
>> -----Original Message-----
>> From: Friam <friam-bounces at redfish.com> On Behalf Of glen
>> Sent: Wednesday, September 7, 2022 7:05 AM
>> To: friam at redfish.com
>> Subject: Re: [FRIAM] more structure-based mind-reading
>>
>> As long as our metaphysical commitment is to an open system, I agree it's pragmatic to choose the model/explanation that's only as complex as it needs to be, not a bit more. But if we're wrong and the world is not open, or non-convex so that some regions of the space, once lost may be difficult to reach again, lost complexity might imply lost opportunity. "Use it or lose it", I guess.
>>
>> Personally, imagine trying to learn long-form literature or philosophy in these days of large language models:
>>
>> https://return.life/2022/07/26/conversation-stopper/
>>
>> I'm seeing more and more links like "I wrote a novel with GPT-3!" Ugh. Why would I read that? And what's the point of learning to write if you can knead GPT-3 into doing it for you?
>>
>> If we're all out here *driving* models to their least complex, we won't be able to detect deep fakes or resist manipulation by evil-doers like Palantir. Nobody'll be able to tell the difference. It'll be bronzer all the way down.
>>
>> On 9/6/22 17:54, Steve Smith wrote:
>>>
>>> On 9/6/22 6:17 PM, Marcus Daniels wrote:
>>>> Why model some complex (if meaningless) phenomena if it can driven toward something less complex?  I mean, jeez, isn’t DJT’s patchy and inconsistent use of bronzer proof that people don’t really care about detail?
>>>
>>> I am often perplexed by this...   I can't tell if DJT cluelessly/arrogantly bronzed up with his own tiny fists or if he had a professional makeup person go to the effort to make him look that bad...   my father's comment about Rodeo Clowns was: "you have to be really good to look that bad"...
>>>
>>> There are other features of DJTs behaviour that suggests it really is arrogant cluelessness, but then there is *also* clearly a "method to his madness" on many levels...  He is the ultimate "tool" which is fascinating because he has created (or groomed) so many "tools" himself...  if one must grant him "genius" it is rooted somehow in his ability to play both ends against the middle in so many dimensions...
>>>
>>>> I watched this black comedy last night.  _Killer Joe_.  It predated MAGA.  It nicely captures how low-dimensional culture can be.  What’s needed in these circumstances is a complete deconstruction and deletion of empathy.  Ask what rats would do.   Oh it takes me back.
>>> I do like me a good "black comedy"...  I recently enjoyed Woody Harrelson as _The Man from Toronto_...  McConaughey also rarely disappoints.
>>>>
>>>>> On Sep 6, 2022, at 11:15 AM, Steve Smith <sasmyth at swcp.com> wrote:
>>>>>
>>>>> 
>>>>>
>>>>> I can't find/recall the exact quote, but you made something of a convert of me when we were discussing whether creativity/learning was *anything more* than complex/elaborate mimicry.
>>>>>
>>>>>
>>>>> Crypto-anythings (closeted "whatevers") have worked this in a similar way to spies, but where there is a little more complicity by the non-cryptos who may well be collaborating in the "closeting", in the spirit of "don't ask, don't tell"...
>>>>>
>>>>>
>>>>> "I/he/she/it/ze can pass" is the bar...   it is OK if some/many of the observers "suspect" the true nature but the community shares the consequences of a community member proving to be "less than fully-compliant".
>>>>>
>>>>>
>>>>> Whitelash supremacists' dog-whistles are a good example.   I don't want to think that my neighbor is part of that movement, so some of the slightly "off color" things she might say across the fence, I am inclined to give the benefit of the doubt to...  so if she notice I don't respond to her dog whistles, she continues to whistle them under her breath now and then, just to soothe her inner racist/mysXinist and maybe keep checking if I maybe have been "converted", and I continue to (hopefully) ignore it and keep bringing her casseroles (laced with xanax) when her husband is recovering from his latest self-inflicted gunshot wound...
>>>>>
>>>>>
>>>>> In this case, we are *all* "acting as if"...   until someone gets converted to "radical honesty" and that just adds another level of indirection of (self/other) deception.
>>>>>
>>>>>
>>>>>
>>>>> On 9/6/22 8:37 AM, Marcus Daniels wrote:
>>>>>> I had to do some cybersecurity training and it was set up so that all the choices one could make led to the same outcome.   The point was to understand the properties of the paths, not the outcome.
>>>>>> While that wisdom might be of some value in some other situation, often there is no discernable difference between the nuance in a social rule and variation that arises due to novelty or ambiguity of circumstances. The signal to noise ratio just isn't high enough to justify the extra precision.   The actors in this training could have been interpreted as quietly demonstrating concern rather than neglect.   One could imagine a cartel boss would not want to wait for a reasonable number of outliers before taking action. After all the cartel boss is a criminal and not concerned with fairness.  An experienced undercover cop knows she needs to mimic the expected distribution very carefully, and that even if she does mimic it very carefully her life is still in danger.
>>>>>>
>>>>>> Marcus
>>>>>> ------------------------------------------------------------------
>>>>>> -
>>>>>> ------------------------------------------------------------------
>>>>>> -
>>>>>> ------------------------------------------------------------------
>>>>>> -
>>>>>> ------------------------------------------------------------------
>>>>>> -
>>>>>> ------------------------------------------------------------------
>>>>>> -
>>>>>> ------------------------------------------------------------------
>>>>>> -
>>>>>> ------------------------------------------------------------------
>>>>>> -
>>>>>> ------------------------------------------------------------------
>>>>>> -
>>>>>> ------------------------------------------------------------------
>>>>>> -
>>>>>> ------------------------------------------------------------------
>>>>>> -
>>>>>> ------------------------------------------------------------------
>>>>>> -
>>>>>> ------------------------------------------------------------------
>>>>>> -
>>>>>> ------------------------------------------------------------------
>>>>>> -
>>>>>> ------------------------------------------------------------------
>>>>>> -
>>>>>> ----------------------------------------------------
>>>>>> *From:* Friam <friam-bounces at redfish.com> on behalf of glen
>>>>>> <gepropella at gmail.com>
>>>>>> *Sent:* Tuesday, September 6, 2022 7:57 AM
>>>>>> *To:* friam at redfish.com <friam at redfish.com>
>>>>>> *Subject:* Re: [FRIAM] more structure-based mind-reading Well,
>>>>>> Steve's targeting of "feeling included" does target "understanding". I'd argue that the spies don't understand the communities they infiltrate. Even deep undercover or method acting doesn't provide understanding. I argue that any bad faith actor like a spy or "acting while cynical" has a reductive objective as their target. What was interesting about the concept of bad faith was Sartre's suggestion that the deep undercover operator who finally *does* begin to identify with the community they've infiltrated is the interesting edge case. That's the cusp of understanding.
>>>>>>
>>>>>> I suppose I'm making a similar argument to EricC's argument for "belief", which I call "dispositional". If you don't act on your belief, then you don't actually believe that thing. So, an undercover cop who infiltrates a drug cartel but refuses to Necklace a local do-gooder just doesn't understand what it means to be in the cartel. They can't understand. And they shouldn't understand. The spy is there for a more specific objective, not understanding.
>>>>>>
>>>>>> One of those more specific objectives might be *prediction*. In simulation and [x|i]ML, there's a stark distinction between predictive versus explanatory power. Ideally, strong explanatory power provides predictive power. But practically, 80/20, reductive prediction is easier, faster, and more important. The excess meaning is swept under the rug of variation or noise. At raves, a reductive objective is harm reduction. Sure, it would be fantastic to teach all the kids pharmaco[kinetics|dynamics] and chemistry ... as well as psychology and neuroscience. But the harm reduction tent is not really there to get into the kids' minds. The objective isn't understanding. It's a reductive focus on dampening the edge cases.
>>>>>>
>>>>>>
>>>>>> On 9/3/22 08:47, Marcus Daniels wrote:
>>>>>>> The claim is that there is all this diversity in subcultures and that the only way to understand them is to participate in them. If it is possible to fake it, and I think it is, then that raises doubts about the claim.   That is what spies specialize in.
>>>>>>>
>>>>>>>> On Sep 2, 2022, at 7:17 PM, Steve Smith <sasmyth at swcp.com> wrote:
>>>>>>>>
>>>>>>>> I have spent most of my life avoiding "acting while cynical"... I have *felt* cynical about a lot of things, and Marcus' description of a lot of things speaks to my "inner cynic" but I haven't spent much time being *harmed* by engaging in "performative activities while feeling cynical about them".    If I dig a hole it is either because *I* need a hole, or someone else *needs* a whole, and only rarely do I help someone dig a hole as a team/trust/affinity building exercise unless the   There are too many holes in the world that *want* digging to spend much effort en-performance.
>>>>>>>>
>>>>>>>> I've never felt particulary "included" in any social circle and I have seen that a little bit of "Performative Grease" might have helped this square peg fit more-better in the round holes it encountered, but generally I simply avoided those activities and drifted further and further out.  That is not to say I haven't *tried* to be a good sport and do what others were doing on the off chance that it would actually be something that worked for me, but generally not.
>>>>>>>>
>>>>>>>> BTW... there seems to be some inverted general usage of "square-peg/round-hole", drilling a round hole and then driving a square(ish) peg into it guarantees a good tight fit... it is preferred to round peg-round hole in traditional joinery.
>>>>>>>>
>>>>>>>>> On 9/2/22 8:17 AM, glen wrote:
>>>>>>>>> OK. But the affinity and "inner self" alluded to by the phrase "faking it" is nothing but a personality momentum, a build-up of past behaviors, like a fly-wheel spun up by all the previous affinities and faking of it. We faked it in our mom's womb, faked it as babies, faked it as children on the playground or in class, etc. all the way up to the last time we faked it digging ditches or pair programming in Java.
>>>>>>>>>
>>>>>>>>> The only difference between feeling an affinity and engaging in a new faking it exercise is the extent to which the new collaboration is similar to the previous collaborations. As both Steve and Dave point out, spend enough time living in a world and you'll grow affine to that world (and the world will grow affine to you).
>>>>>>>>>
>>>>>>>>> I suppose it's reasonable to posit a spectrum (or a higher dim space) on which some people have particularly inertial fly-wheels and others have more easily disturbed things that store less energy. Of the Big 5, my guess would be neuroticism would be most inertial. Perhaps openness and agreeableness would be the least inertial.
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>> On 9/2/22 05:35, Marcus Daniels wrote:
>>>>>>>>>> There are many common tasks that parties could direct their attention toward.   This happens at companies, prison cafeterias, and churches.   That it is grounded in a particular way doesn't make it any truer, or anyone more committed to it.   We are often forced to participate in cultures we don't care about, and faking it is an important skill. Just because someone sweats or gets calluses or tolerates others' inappropriate emotions in some circle of people, doesn't mean there is any affinity toward that circle. Oh look, he dug a hole.  I dug a hole.    Sure, I'd do those kind of performative activities if I were a politician, as I bet there are people who think this way.
>>>>>>>>>>
>>>>>>>>>> -----Original Message-----
>>>>>>>>>> From: Friam <friam-bounces at redfish.com> On Behalf Of glen
>>>>>>>>>> Sent: Friday, September 2, 2022 12:06 AM
>>>>>>>>>> To: The Friday Morning Applied Complexity Coffee Group
>>>>>>>>>> <friam at redfish.com>
>>>>>>>>>> Subject: Re: [FRIAM] more structure-based mind-reading
>>>>>>>>>>
>>>>>>>>>> And, of course, there is no such thing except appearance. What could it possibly mean to say that an appearance of a bond exists, but no actual bond exists?
>>>>>>>>>>
>>>>>>>>>> On September 1, 2022 7:29:45 PM PDT, Marcus Daniels <marcus at snoutfarm.com> wrote:
>>>>>>>>>>> If you want to create the appearance of a bond where none exists, get to work.   Once one recognizes the nature of work it is easy.
>>>>>>>>>>>
>>>>>>>>>>> On Sep 1, 2022, at 6:25 PM, Prof David West <profwest at fastmail.fm> wrote:
>>>>>>>>>>>
>>>>>>>>>>> 
>>>>>>>>>>>     From glen: "If you want to share values with some
>>>>>>>>>>> arbitrary shmoe, then get to
>>>>>>>>>>>           *work*. Build something or cooperate on a common task.
>>>>>>>>>>> Talking,
>>>>>>>>>>>           communicating, is inadequate at best, disinfo at worst."
>>>>>>>>>>>
>>>>>>>>>>> This is kinda the whole point of Participant Observation at the core of cultural anthropology. The premise is you cannot truly understand a culture until you live it.
>>>>>>>>>>>
>>>>>>>>>>> Of course, there is still a boundary, a separation, between the anthropologist and those with whom she interacts, but sweat, calluses, blood, and emotions go a long way toward establishing actual understanding.
>>>>>>>>>>>
>>>>>>>>>>> davew
>>>>>>>>>>>
>>>>>>>>>>> On Thu, Sep 1, 2022, at 12:30 PM, Steve Smith wrote:
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> On 9/1/22 11:21 AM, glen wrote:
>>>>>>>>>>> Inter-brain synchronization occurs without physical
>>>>>>>>>>> co-presence during cooperative online gaming
>>>>>>>>>>> https://www.sciencedirect.com/science/article/pii/S0028393222
>>>>>>>>>>> 001750
>>>>>>>>>>>
>>>>>>>>>>> There's a lot piled into the aggregate measures of EEG. And the mere fact of the canalization conflates the unifying tendencies of the objective (shared purpose) with that of the common structure (virtual world, interface, body, brain). But overall, it argues against this guru focus on "sense-making" (hermeneutic, monistic reification) and helps argue for the fundamental plurality, openness, and stochasticity of "language games".
>>>>>>>>>>>
>>>>>>>>>>> If you want to share values with some arbitrary shmoe, then get to *work*. Build something or cooperate on a common task. Talking, communicating, is inadequate at best, disinfo at worst.
>>>>>>>>>>>
>>>>>>>>>>> I agree somewhat with the spirit of this, however a recent writer/book I discovered is Sand Talk<https://www.harpercollins.com/products/sand-talk-tyson-yunkaporta?variant=32280908103714> by Tyson Yunkaporta and more specifically his references to "Yarning" in his indigenous Australian culture offered me a complementary perspective...
>>>>>>>>>>>
>>>>>>>>>>> I definitely agree that the "building of something together" is a powerful world-building/negotiating/collaborative/seeking experience.   The social sciences use the term Boundary Object<https://en.wikipedia.org/wiki/Boundary_object> and Boundary Negotiation Artifact.    Jenny and I wrote a draft white-paper on the topic of the SimTable as a "boundary negotiating artifact" last time she visited (2019?).    A lot of computer-graphics/visualization products provide fill this role, but the physicality of a sand-table with it's tactility and multiple perspectives add yet more.   The soap-box racer or fort you build with your friend as a kid provides the same.   The bulk of my best relationships in life involved "building something together" whether it be a software system or a house...
>>>>>>>>>


--
ꙮ Mɥǝu ǝlǝdɥɐuʇs ɟᴉƃɥʇ' ʇɥǝ ƃɹɐss snɟɟǝɹs˙ ꙮ

-. --- - / ...- .- .-.. .. -.. / -- --- .-. ... . / -.-. --- -.. .
FRIAM Applied Complexity Group listserv
Fridays 9a-12p Friday St. Johns Cafe   /   Thursdays 9a-12p Zoom https://bit.ly/virtualfriam
to (un)subscribe http://redfish.com/mailman/listinfo/friam_redfish.com
FRIAM-COMIC http://friam-comic.blogspot.com/
archives:  5/2017 thru present https://redfish.com/pipermail/friam_redfish.com/
  1/2003 thru 6/2021  http://friam.383.s1.nabble.com/
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://redfish.com/pipermail/friam_redfish.com/attachments/20220907/26a2775f/attachment.html>

More information about the Friam mailing list