[FRIAM] Wolpert - discussion thread placeholder

[glen∉ℂ]

It's tough to resist the "larding". But I'll try.

- Static-dynamic: By "grow", I don't really mean "dynamic"... or not merely dynamic. I mean something more akin to a co-evolutionary, adversarial network accumulation. Time is necessary. But so is space. The contrast with specification targets "algorithms", compressions, and "finite sequences from finite alphabets". *Can* such adversarial accumulation be specified, formally or even informally?

- Ice9, cancer, et al: I think I disagree. I should avoid the word "intelligent" because I agree with Dave, that concept is one of the worst examples of homo-arrogant abstraction. I *do* consider cancer to be alive in some interesting sense, if only a pattern of behavior on top of a living substrate. My cancer, in particular, is really just a psychopathic immortal cell type. The runaway growth is really just that some cells refuse to politely commit suicide. I also think it's reasonable to call some of our AI workflows "alive" in some primitive sense, again even if only as a pattern of behavior on a living substrate.

- Scientism: But Wolpert's explicitly asking about our biases! That's the whole point of the paper, to object to that ethno-centrism. Y'all seem to be criticizing Wolpert for *trying* to steelman the Scientismists. You literally cannot make an effective objection without first demonstrating that you understand the details of your opponents' position. Calling him biased in this way *because* he's trying to build a good steelman seems a bit myopic. Of course, maybe we don't have to play the game he sets up. I tried to show I object to his steelman (but not his conclusion) because our proofs don't seem like *only* "finite sequences from finite alphabets". And I tried to object to that with List's mention of indexicality. But I think we could reformulate Wolpert's questions with that and it would be even more steely. *And* it would still agree with Wolpert's main point: that we are more limited/biased than we can possibly imagine.

In any case, don't read my comments the wrong way. I'm enormously grateful y'all have engaged.

On 9/15/22 10:02, Steve Smith wrote:
> glen∉ℂ wrote:
> 
>> Great question. 
> I also appreciate the specificity of the question, despite wanting to tease it into 3 parts: A) convincing evidence; B) superior intelligence; C) cultural inheritance .
>> I agree with Dave's emphasis against "finite sequences from a finite alphabet" as being central to our SAM. *If* Wolpert's actually relying on that as crucially as he seems to be, then the "grow vs. specify" accusation isn't a strawman.
> Static (specification) vs dynamic (growth) is an important and I think fundamental distinction.  A genome *is* a finite specification while the embryology of it's earliest expressive development and the "cultural embedding" it continues to form within are not precisely finite (maybe finite-huge in scale but not finite in pre-stateability?).
>> But the question Wolpert wants to ask remains; and your concise phrasing nails it. If there is an "effective computing" artifact that demonstrates maximal intelligence with minimal cultural grounding, what is it? One valid answer is there is no such thing.
> I do think the question is on the same as "what is art" and "what is pornography" and the answer "I know it when I see it" isn't fully responsive but possibly as good as it gets?
>> All forms of "intellignece" are not abstract, are embedded-embodied-concrete, tightly grounded to context. (Where I'm probably relying on my definition of "concrete" more than Dave's.)
> In pursuit of an abstract definition of B) above it is tempting to gesture toward "fitness for survival" but with a *larger* sense of "self" and a long-now sense of "time".   Ice9, Cancer and grey-goo have high fitness by some measure but in both cases most would be loathe to call them "intelligent".   An expansive fitness with an arbitrarily broad sense of "what means self" might be the most abstract way of thinking of "superior intelligence"?
> 
>> But I think that answer, however valid, is unsound. There are ways of behaving that *translate* across contexts. The berserker physicists who take that to the extreme notwithstanding, anyone who travels experiences this. As Wolpert explicitly mentions, perhaps the "level" at which this occurs is our bodies? As long as the society I visit on Alpha Centauri was built by homonid-similars, I think some set of my behaviors will translate, however small that set.
> I think you are arguing for the definition of "self" in this case to be confined to the contents of our skin-bag (torus really), and maybe on a good day some of the cells recently shed from it's surface or expelled from  one end of it's digestive canal or the other?
>> But maybe there's a lower level, perhaps capturing less concrete detail than a homo-built society, of water and carbon based life? I.e. any society built by water and carbon based life will allow some translation of behaviors to our society?
> It is familiar to define it as carbon-based life, but seems like a coincidence of history and awareness (if perchance there are non-carbon based life-forms we are unaware of within our light-cone)?
>>
>> I don't grok Dave's antipathy, though. It seems to me like Wolpert is *asking* these questions and challenging our berserker Scientismists and Mathematicians in the very same gist as Dave does. Wolpert wouldn't write (and distribute) papers like this if he *weren't* a bit skeptical of the universality of our SAM.
> Speaking for my inner DaveW, I think *my* antipathy is not really specifically to Wolpert's specific questions/formulation, but the *larger* expanse of Wolperts-at-large whose biases are (naturally) ethno-centric or more accurately human-chauvanistic and contemporary-western-civilization centric?   I am more acutely antipathic in this regard *because* I often *am one*...  there is no anti-smoker at large than a former smoker, especially one who perchance sneaks a guilty fag in private now and then?
>>
>> On 9/14/22 22:29, Marcus Daniels wrote:
>>> What would be convincing evidence of a superior intelligence independent of cultural inheritance?
>>>
>>>> On Sep 14, 2022, at 7:34 PM, Steve Smith <sasmyth at swcp.com> wrote:
>>>>
>>>> 
>>>>> On 9/14/22 7:31 PM, Marcus Daniels wrote:
>>>>> ML gets better every day because it learns more like a newborn child than a university student.   This isn't 1970s AI anymore. It all seems like a strawman argument, whether you know it or not.
>>>>
>>>> And as I have referenced watching a puppy and a kitten grow together from 3 and 4 months respectively, I believe that broadly, contemporary ML is learning like they are. Current fetishes for NLP to drive NLG and Visual Art misses a *lot* that animals (even one's domesticated by us for millenia) do so well as they express what their genes and gestation already prepare them for.
>>>>
>>>> I'd claim the puppy knows a modest vocabulary of human utterances/gestures already, though to a dog, I think human language is very tonal to animals, to the point that maybe I can say "YES" in the same tone I say "NO" and vice versa and the tone, not the phoneme would dominate.
>>>>
>>>> The kitten is (as I feel all cats are) almost entirely disinterested in our *intentional* communications and *much more* aware of the implications of our *actions* than in our words. The puppy does seem to have a much stronger sense of anticipating our interests and seeking our approval.  The cat is more interested in her interests and treating us as facilitators or constraints to obtaining those.
>>>>
>>>> Paw prints of either species qualify as "art" in our house anytime they get involved in a painting project or the setting of plaster, cement, or clay.   Our appreciation of same reflects *our* training more than *theirs*.
>>>>
>>>>>
>>>>> -----Original Message-----
>>>>> From: Friam <friam-bounces at redfish.com> On Behalf Of Prof David West
>>>>> Sent: Wednesday, September 14, 2022 5:54 PM
>>>>> To: friam at redfish.com
>>>>> Subject: Re: [FRIAM] Wolpert - discussion thread placeholder
>>>>>
>>>>> Regarding Wolpert's first four questions:
>>>>>
>>>>> In my opinion, all four reflect a kind of arrogance that I have accused Scientists and Mathematicians of many times in the past—an attitude that modern formal and abstract math and science are a kind of ultimate achievement of our species. Any and all other forms/means of understanding are discounted or denied. This is analogous to the arrogance of Simon and Newell (mentioned previously) that a machine that thought like a university professor was necessarily intelligent.
>>>>>
>>>>> Ignored in the AI instance is the learning ability of a new born child. Ignored in the case of SAM is the very real Science and Mathematics exhibited by our species beginning in the Neolithic. Metallurgy, agriculture, animal husbandry, pottery, weaving, cooking, food preservation, etc.
>>>>>
>>>>> Levi-Strauss writes extensively of two different kinds of science: concrete and abstract; the former grounded in perception and imagination, the latter divorced from same. The object of all science is connections and explanations and based on experimentation and empirical evidence, but "concrete science" relies far more heavily on sensible intuition and not formal "proof."
>>>>>
>>>>> SAM, for Wolpert, seems to be restricted to the that which came into being the past few hundred years. This fetish makes questions like—"Why do we have that cognitive ability despite its fitness costs?"—somewhat nonsensical. What fitness costs? Mutually assured destruction with nuclear weapons?" Certainly there were no evolutionary fitness costs; and, in fact, those cognitive abilities were essential and the prime mover of our species out of the neolithic.
>>>>>
>>>>> A more reasonable question is what caused a small subset of our species to 'go beserk' and take a subset of the SAM that served our species so well for so long, to such abstract extremes? An answer might be found, and is argued, in the Ian McGilchrist works on recent  "left-brained" dominance. [left-brain is such a limited shorthand for what McGilchrist argues in some 700 pages of prose, that I am trepedatious using it lest it evoke the wrong headed popularization of the notion.]
>>>>>
>>>>> If we ignore the aberrant contemporary SAM and ask if we can find evidence that other species, e.g., cephalopods and cetaceans, have an equivalent to the concrete SAM that was widespread among our own species as far back as the neolithic. The answer is yes. Tool making, modification of environment, herding, even quasi-domestication of other species can be found.
>>>>>
>>>>> The cognitive abilities of dolphins and octopi (et. al.) are well documented and include language, reasoning, knowledge of spatial relationships, planning, and even (when given LSD (famously the research by John Lilly with dolphins and more recently with octopi), altered states. There is little, or no, reason not to assume them to be SAM-sufficient for their environments and needs, just as humans were prior to, roughly, the Renaissance.
>>>>>
>>>>> to be continued ...
>>>>>
>>>>> davew
>>>>>
>>>>>
>>>>>> On Mon, Sep 12, 2022, at 6:29 AM, glen∉ℂ wrote:
>>>>>> My question of how well we can describe graph-based ... what? ...
>>>>>> "statements"? "theorems"? Whatever. It's treated fairly well in List's
>>>>>> paper:
>>>>>>
>>>>>> Levels of Description and Levels of Reality: A General Framework by
>>>>>> Christian List http://philsci-archive.pitt.edu/21103/
>>>>>>
>>>>>> in section "6.3 Indexical versus non-indexical and first-personal
>>>>>> versus third-personal descriptions". We tend to think of the 3rd
>>>>>> person graph of possible worlds/states as if it's more universal ... a
>>>>>> complete representation of the world. But there's something captured
>>>>>> by the index/control-pointer *walking* some graph, with or without a
>>>>>> scoping on how many hops away the index/subjective-locus can "see".
>>>>>>
>>>>>> I liken this to Dave's (and Frank's to some extent) consistent
>>>>>> insistence that one's inner life is a valid thing in the world, Dave
>>>>>> w.r.t. psychedelics and meditation and Frank's defense of things like
>>>>>> psychodynamics. Wolpert seems to be suggesting a "deserialization" of
>>>>>> the graph when he focuses on "finite sequences of elements from a
>>>>>> finite set of symbols". I.e. walking the graph with the index at a
>>>>>> given node. With the 3rd person ... whole graph of graphs, the
>>>>>> serialization of that bushy thing can only produce an infinitely long
>>>>>> sequence of elements from a (perhaps) infinte set. Is the bushiness
>>>>>> *dense* (greater than countable, as Wolpert asks)? Or sparse?
>>>>>>
>>>>>> I'm sure I'm not wording all this well. But that's why I'm glad y'all
>>>>>> are participating, to help clarify these things.
>>>>>>
>>>>>> On 9/12/22 06:13, glen∉ℂ wrote:
>>>>>>> While math can represent circular definitions (what Robert Rosen complained about), there are deep problems in the foundations of math ... things like the iterative conception of sets ... that are attempts to do what Wolpert asks for in the later questions. And it's unclear to me that commutative categories reduce to "finite sequences of elements from a finite set", prolly 'cause I'm just ignorant. But diagrammatic loops in graphs don't look to me like finite sequences.