Glen - What a great (continued) riff on the (general) topic, in spite of the thread wandering more than a little (no kinks but far from straight and smooth).
I would like to contrast "learning" with "problem solving" as I think the latter is the key point of what might allow "general intelligence" (if such exists, as you say) to distinguish itself. Many may disagree, but i find the essence of "problem solving" at it's best to be the art and science of "asking the right question". Once that has been achieved, the "answer" becomes self-evident and as Feynman liked to say "QED!" Contemporary machine learning seems to confront the definition of "self-evident" and "quite easily" The 1970's computer-proof of the 4-color problem is a good (on that boundary) example. Perhaps we could say that the program written to do the search of the axiom/logic space is a prime (if obscure) example of "asking the right question" and (though it is a bit of a stretch) the halting/solution of the program represents the "self-evident" answer (nQED?). At the very least, this is how I take the idea of "elegant" solutions to be (though the complexity of the 4-color problem-solution would seem to be a far stretch for what one would call "elegant"). Contemporary machine (deep?) learning techniques (even those emerging in the late 80s such as evolutionary algorithms) seem to demonstrate that a suitably "framed" question is as good as a well "stated" question with the right amount/type of computation. EA, GA, Nueral Nets, etc. are all "meta-heuristics" . I am not sure I can call applications of these techniques, even in their best form, "general intelligence" but I think I would be tempted to call them "more general" intelligence. I would *also* characterize a LOT of human problem-solving as NO MORE general, and the problem of "the expert" seems to frame that even more strongly... it often appears that an "expert" is distinguished from others with familiarity with a topic by *at least* the very same kind of "supervised learning" that advanced algorithms are capable of. Some experts seem to be very narrow, and ultimately not more than a very well populated/trained associative memory, whole others seem very general and are *also* capable of reframing complex and/or poorly formed questions into an appropriate and well-formed enough question for the answer to emerge (with or without significant computation in between) as "self-evident". There are plenty of folks with more practical and theoretical knowledge of these techniques than I have here, but I felt it was worth trying to characterize the question this way. Another important way of looking at the question of what can be automated might be an extension/parallel to the point of "if you have a hammer, then everything looks like a nail". It seems that our socio-political-economic milieu is evolving to "meet the problem of being human" halfway, by providing a sufficiently complex set (spectrum?) of choices of "how to live" to satisfy (most) everyone. This does not mean that our system entirely meets the needs of humanity, but rather that it does at a granularity/structure that it many (if not most) people can fit themselves into one of it's many compartments/slots in a matrix of solutions. Social Justice and Welfare systems exist to try to help people fit into these slots as well as presumably influencing the cultural and legal norms that establish and maintain those slots. The emergence of ideas such as Neurodiversity and this-n-that-spectrum diagnoses seem to help deal with the outliers and those falling between the cracks but this is once again, an example (I think) of force-fitting the real phenomenon (individuals in their arbitrary complexity) to the model (socio-political-economic-??? models). Mumble, - Steve On 3/5/19 2:40 PM, uǝlƃ ☣ wrote: > I can't help but tie these maunderings to the modern epithets of "snowflake" > and "privilege" (shared by opposite but similar ideologues). I have to > wonder what it means to "learn" something. The question of whether a robot > will take one's job cuts nicely to the chase, I think. How much of what any > of us do/know is uniquely (or best) doable by a general intelligence (if such > exists) versus specific intelligence? While I'm slightly fluent in a handful > of programming languages, I cannot (anymore) just sit down and write a > program in any one of them. I was pretty embarrassed at a recent interview > where they asked me to code my solution to their interview question on the > whiteboard. After I was done I noticed sugar from 3 different languages in > the code I "wrote" ... all mixed together for convenience. They said they > didn't mind. But who knows? Which is better? Being able to coherently code > in one language, with nearly compilable code off the bat? Or the > [dis]ability of changing languages on a regular basis in order to express a > relatively portable algorithm? Which one would be easier for a robot? I > honestly have no idea. > > But the idea that the arbitrary persnickety sugar I learned yesterday > *should* be useful today seems like a bit of a snowflake/privileged way to > think (even ignoring the "problem of induction" we often talk about on this > list). Is what it means to "learn" something fundamentally different from > one era to the next? Do the practical elements of "learning" evolve over > time? Does it really ... really? ... help to know how a motor works in order > to drive a car? ... to reliably drive a car so that one's future is more > predictable? ... to reduce the total cost of ownership of one's car? Or is > there a logical layer of abstraction below which the Eloi really don't need > to go? > > On 3/5/19 11:04 AM, Steven A Smith wrote: >> Interesting to see the "new bar" set so low as age 30. Reminds me of my >> own youth when the "Hippie generation" was saying "don't trust anyone >> over 30!". Later I got to know a lot of folks from the "Beat" >> generation who were probably in their 30's by that time and rather put >> out that they couldn't keep their "hip" going amongst the new youth culture. >> >> ... >> My mules are named Fortran/Prolog/APL/C/PERL and VMS/BSD/Solaris/NeXT >> and IBM/CDC/CRAY/DEC and GL/OpenGL/VRPN/VRML. I barely know the names >> of the new >> tractors/combines/cropdusters/satellite-imaging/laser-leveling/??? >> technology. >> >> Always to be counted on for nostalgic maunderings, ============================================================ FRIAM Applied Complexity Group listserv Meets Fridays 9a-11:30 at cafe at St. John's College to unsubscribe http://redfish.com/mailman/listinfo/friam_redfish.com archives back to 2003: http://friam.471366.n2.nabble.com/ FRIAM-COMIC http://friam-comic.blogspot.com/ by Dr. Strangelove
