[FRIAM] Selective cultural processes generate adaptive heuristics
Steve Smith
sasmyth at swcp.com
Tue Apr 12 22:07:02 EDT 2022
On 4/12/22 5:10 PM, glen wrote:
> Ha! 8^D
>
> But neither the ANN clone, nor the *stereotyped* heuristics generated
> by an autonomous car capture the high-dimensional opportunity I
> believe meat organisms experience. Yes, the subsequent evolution of
> the ANNs and the stereotyped out-group are more concrete than most
> synthetic minds. But my claim, were I to actually hold it and try to
> state it more clearly, is that meat, living in meat space, is more
> open than those 2 examples. It's the openness that provides the meat
> with the opportunity. The ANNs and autonomous car are more fixed, more
> closed.
>
> However, I do believe machine intelligence *will* reach meat
> intelligence. But it'll have to look a lot more like meat intelligence
> to do so. It's already looking a lot more like meat intelligence than
> it was even 10 years ago. And if we stay at this supralinear rate (or
> higher), it'll happen sooner than I, this meat bag, thinks.
As a "meat bag" driving a car, I find myself *regularly* imagining what
a driverless car would do in the current/instantaneous moment as I
interrupt my own automatic driving instincts to run a supervisory
analysis consciously to make sure that my instincts are doing "the right
thing" which usually yields something like an infinite regress of second
guessing which only ends/completes/halts because the clock runs out...
the race condition converges between my
reactions/instincts/conscious-review and the events-of-the-world.
As I've had *very* few (but not-zero) out-of-control moments in a
vehicle and as I write this, I feel like I should (if I could) collect
up my most authentic memories of those moments and do some kind of
meta-analysis of them. What I have done in the moment (after the
chaotic events returned to somewhat linear in my apprehension) is
usually to peg some overly conservative heuristic over the top of my
more regular suite of heuristics that seem to keep me on my own side of
the road and maintaining appropriate speeds and yields in most
contexts. With time, the precedence of those heuristics fade or get
papered over by "yet other" heuristics for various other competing goals
(hurrying, staying entertained, testing my abilities, etc.).
In any case, I have no trouble imagining the car I'm driving (with the
mildest of safety features like sonar proximity alarms in bumpers for
parking and "your turn-signal is on, dipshit" and automatic
driving/headlamps) incrementally getting better and better at helping me
drive until *I* am either letting it do the driving and running my
"supervisory heuristics" as I observe and consider intervening, OR even
just putting on my VR goggles and watching a movie or playing a game or
hacking lamely at some code while the car drives me off a cliff.
Chances are the ability of my car will exceed my ability as a
highway/urban driver sometime before *I* become a radical hazard as a
driver (NM DMV just allowed me to renew my DL for 8 years sight unseen).
If my car emulates *my* style of driving (because it trains on my actual
driving?) somewhere down the road (pun recognized but not intended), I
can't even imagine what it would mean for it to have a subjective,
conscious experience like my own, even if somehow there is a monitor
running that layers in heuristics uncannily like the ones *I* paste up
in my imagination. Of course, I *know* that a proper self-driving
vehicle will be trained on some maximal ensemble of *good drivers* and
not on my personal ideosyncratic style. It might be different if it
were training for Stock Car racing on a dirt track where my
ideosyncracies might be somehow useful (or more likely entertaining).
I don't know if anyone else has these kinds of self-reflective episodes
about who we might (already) be becoming as the boundaries between our
phenotypic selves and our techno-extended selves blur. I have been
riding a bicycle since I was about 4, a motorcycle since I was about 15,
and driving a car since soon after that. My entire CN system all the
way out to the fine-structure is co-evolved with those activities, even
though those devices didn't even exist until about 200 and 130 years
respectively. Yesterday I fumbled my newer, more manageable bicycle (as
of 4 months ago) into a spill and though it has been a couple of decades
since I fell off of a bike, it seems my CNS/muscle memory remembered how
to fall yet better than it remembered how to not-fall. Neolithic (or
even mesolithic) hunters probably had an even more acute merging with
their various tools than I have with my own (esp. vehicles, keyboards,
tracking devices, hammers, chainsaws, alexa, siri ... ).
>
> On 4/12/22 15:58, Marcus Daniels wrote:
>> Now it is entirely possible to take a massive pre-trained neural net
>> like GPT3 and run it in two places at once or have different
>> instances use a baseline and take divergent paths from different
>> training.
>> None of that is possible for humans, at least yet. Some autonomous
>> cars even know enough to be afraid of the police! (Regarding
>> concreteness.)
>> https://electrek.co/2022/04/10/gm-cruise-autonomous-taxi-pulled-over-by-police-in-san-francisco-without-humans-bolts-off-u-cruise-responds/
>>
>> -----Original Message-----
>> From: Friam <friam-bounces at redfish.com> On Behalf Of glen
>> Sent: Tuesday, April 12, 2022 3:47 PM
>> To: friam at redfish.com
>> Subject: Re: [FRIAM] Selective cultural processes generate adaptive
>> heuristics
>>
>> Exactly. Both of these (low turnover wisdom propagation & "flat"
>> infoscape) fail in my conception because they lack the concrete
>> (definit) particulars. Even if we have one 400 year old vampire
>> telling funny stories to a 30 year vampire about a now-exploded
>> vampire from 700 years ago, the sheer *number* of anecdotes required
>> to capture a 400 year lifespan *forces* some abstraction ... some
>> leaving out of important detail.
>>
>> And even if the concrete details of why, say, Galileo was such an OCD
>> journaling nerd can be found in biographies or whatnot, actually
>> reading and learning about all the persnickety nonsense that was
>> *crucial* to the arrival at, emergence of, any given inflection
>> point, ... even if that concrete detail is logged/documented out
>> there somewhere, nobody can learn it all. Each learner is forced to
>> take an abstracted slice through it.
>>
>> What the commitment to meat space interactions is, is a way to ensure
>> that the concreteness remains ... at least within *some* small "open
>> ball", you're getting a high-dimensional opportunity. I think of it
>> in terms of the space vs time tradeoff and (yes, broken record) the
>> parallelism theorem. Sure, a sequential system can simulate a
>> parallel one perfectly, but only if you give it the time to do so ...
>> and the amount of time it takes to do it is related to the amount of
>> space the parallel system uses. Another way to think of it is the
>> project management triangle: cheap, fast, or good. But those are
>> low-dimensional. The space being balanced by organisms in the world
>> is high-dimensional.
>>
>> On 4/12/22 14:19, Steve Smith wrote:
>>> Generations past (and under-mobile near-subsistence cultures today)
>>> have more intergenerational households and neighborhoods providing
>>> the heterarchical/holarchical connection/communication you
>>> suggest. Or so my "just so" story relates.
>>>
>>> The expansive breadth offered by global (near-instantaneous, global)
>>> communication/publication/relationship connections possibly makes up
>>> for that in the large, a major refactoring of problems and solutions.
>>>
>>> I personally suffer from the lack of cross-cultural, cross-class
>>> experience of frequenting a neighborhood "watering hole"
>>> (pub/tavern/saloon) in the way Glen seems to enjoy (cultivate). My
>>> oldest regular drinking-philosophy buddy would be over 110 today (he
>>> died over 20 years ago from alcohol-related illness) and until about
>>> 5 years ago I had a small cohort of 30ish imbibing interlocutors. I
>>> blame COVID, but the reasons are probably larger and more nefarious.
>>
>>
>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: OpenPGP_0xFD82820D1AAECDAE.asc
Type: application/pgp-keys
Size: 3122 bytes
Desc: OpenPGP public key
URL: <http://redfish.com/pipermail/friam_redfish.com/attachments/20220412/d586732c/attachment.bin>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: OpenPGP_signature
Type: application/pgp-signature
Size: 840 bytes
Desc: OpenPGP digital signature
URL: <http://redfish.com/pipermail/friam_redfish.com/attachments/20220412/d586732c/attachment.sig>
More information about the Friam
mailing list