@Mike and @Alan > "Simple" problems are precisely the problems that computers have the > most trouble with. =P > > If my example is so simple, then go ahead and implement it and > demonstrate how simple it is. ;)
Alan, he may not do this, but when I start working on and I have an appropriate simulator or a robot, I may do, and it's really elementary, IMHO even in emergent terms (and won't be that inefficient), as I explained what magical parameters will "emerge" as interesting for a baby in similar settings. The "simple problems" are not hard for the computers, it's hard for humans to understand those "simple problems" - however people keep calling problems that they cannot understand or define "simple". "Visual intelligence", "Spatial intelligence" --- everything is spatial (even time), every coordinate transformation is application of "spatial and visual intelligence", and there are coordinates in sound, touch, in proprioception -- everywhere, even at the lowest level. The essence of vision are coordinates and proportions (relative coordinates). *Why one might **be unable to understand so "simple" problems, and yet call them "simple"?* Things which are simple for fully-functional and grown 20-30 year-old mind, that grew-up in a human civilization built for millenia, is not simple in "raw" terms, in terms of the "machine language of Universe" and not that simple for a self-improiving general intelligence. If something is built on a 834983429839583495898349 previous steps, involving 8743598349583495843 different human agents and machines during the time, and their interactions. Some researchers try to explain the behavior built on those 834983429839583495898349 steps in one direct step and say "whoa, it's impossible". Their ass is... possible. This is about NL, though, stacking bricks or pushing them through holes is simple for a baby, and will be simpler for a computer with sensori-motor generalizing cognition, see the email to Mike. In fact programming it is "harder", requires more education, focus, to understand explicitly in the highest level of your mind trigonometry, triangulation/stereo-vision, reverse-kinematics and so on. A baby doesn't need this. Those explicit mathematical definitions are optional for a self-improving system, reverse kinematics is not needed, just alter the coordinates of the actuator gradually until matching coordinates and matching visual match with tactile and proprioceptive matches. It's so elementary and obvious. Alan, you claim you're a hard worker and you do for years, you should have implemented it about 1000 times already... ... People are stuck in their high levels of abstractions, that's why complex things are easier "for the computers" - i.e. for humans to understand. Abstract things are labeled "complex", children don't manage them, "that's hard". More specific things, based on low generality data, which children can manage, are labeled as "simple" for the adults, but the adults still cannot explain and understand them, they even blame computers for their own stupidity. The higher the level of abstraction, the lower is the raw computing capacity, and the higher the level-difference, the worst the reflection. The highest level have difficulties understanding how levels which are a lot lower work, while that's not true for the levels, which are close, and involve small amounts of data to define at that level. That's how logic or mathematics (the elementary one) seem obvious (it's graspable in mind), while activities involving more details are not, they involve too much data for an ordinary human to realize in her conscious, and conscious is too clumsy and crippled in this domain. @Mike: >What's amazing about this - and Sergio is typical here - Ben and Boris are >just as much in the same boat - is that it's an account of intelligence/mind >that more or less leaves out **movement** - or, at any rate, regards it as >extremely secondary, more or less an "afterthought" after thinking and >perception. >In truth, movement is **primary** - the brain only comes into being with >movement. Mike, how can you be unable to understand, that "coordinate adjustment" in the lowest level of the system, in physical terms in this reality means "movement". Movement is just a change of the coordinates of a part of the body, which is the only way for humans to intensionally output data to the environment. (BCI are not serious yet.) >Can we have a simple example of emergent inference, Sergio? >Here's a simple example of what I would consider emergent intelligence - if a program that deals only in a,b,c,d .. the A-Z alphabet, can suddenly start producing something like - though not the same as - Greek or >Hebraic characters, or cuneiform (without of course knowing beforehand of other alphabets). That's true emergence - and the essence of AGI. This is extremely pathethic. Alphabets in their graphical form (apparently the only POV you see here) are just a sequence of strokes (or a set of vectors), the letters are a set of short sets of strokes, most of them are randomly chosen (among the possible easy/strokes), as long as they are short enough (the complex ones would include the simple ones). "Emergence" here is choosing a sequence of strokes that is different from the ones already in the set, this is randomization. And the letters are derived not from an alphabet, but from the STROKES, which is a set of sequences of coordinate adjustments or transformations. "Motion" is the same thing -- a sequence of changes of the coordinates. You apparently don't see that letters are mapped to sounds (which in fact are mapped to other sequences of coordinate adjustments in the vocal tract). Regarding the other alphabet, the graphical aspect of the Greek alphabet is derived from Shumerian? or whatever, Latin and Cyrillic are derived from the Greek's graphics. There was another cyrillic alphabet, the Glagolitc, which was not derived from the Greek, it was just a set of strokes which are enough to distinguish between the elements of the set. It was phonetically precise, but technically - redundant, the complexity was not needed so it was replaced by the Cyrillic. And that's essential the function of the letters: to DISTINGUISH between the elements in the set, and also - to be short/easy enough. In a digital system it can all be stored in mere 101010100, or in more efficient graphical form. >Or if a program can suddenly move on from building houses with bricks to constructing houses of rocks or logs or cardboard boxes. That's true emergence. >Emergence = the generation of 1) new kinds of elements that cannot be formulaically derived from the existing kinds - and/or 2) new forms that do not fit the same pattern as - cannot be inferred from - existing forms. This is even more pathetic. The essence of building is not the material, building with toy bricks and big construction bricks is the same in terms of coordinate transformations. Rocks, cardboard boxes - that's just about size, weight, strenght, you see inessential features. The essential functions are such as stability, separation from the environment, cover (a plane over the ground plane) and coordinate transformation in order to achieve it. The coordinate transformations are all the same: change the coordinates of the arms/hands/actuators, and respectively - the coordinates of the objects in those actuators, and stack the objects in order to achieve the essential features of the buildings - stability, separation from the environment, strenght, coverage ... >Could you explain how "emergent **inference**" isn't a contradiction in terms? Isn't something that emerges precisely something that by definition CANNOT be "inferred" from what has gone on before? Every stage of >evolution, natural and technological, is non-inferrable from the previous one. That's a huge bullshit. Study the history of technology (and biological) and you'll see every invention comes in its time for a reason and because there were prerequisite technologies, and that was some last step and integration of them. But yes, you should understand the domain in its low levels which drive it forward, not see only inessential features such as appearance (embryon example). Have you studied embryology? What's not inferrable from the previous stages? Cells start to divide and specialize, tissues are built of cells, if there are so and so cells of particular type in particular coordinates/area, the spatial appearance of the body will be so and so. If the level of the hormones in particular part of the body in particular time is so and so, particular cells will grow or divide or specialize, or migrate here or there. What's "uninferrable" - you see the fetus and everything as hollow graphical objects without internals, that's one of your cognitive problems. >For example, the brain is not inferrable from a distributed neural net, "Distributed neural net" - what's this? Yes, the brain is inferrable by the physical/biological whatever laws are behind it. >ditto simple amphibian legs from fins, It's exactly how it probably has happened, but not in terms of legs or fins, but in terms of alteration of division/inhibitory cycles/intensity, which in essence is an alteration of molecules in the DNA, which is inferrable, if you have that data in full detail. >the jet is not inferrable from the propellor etc. Stop bullshitting - yes, visually it's not, but essentially it is. In fact the function of the propellor is the same like the function of the jet engine -- to push air/gas at high speed through the wings. When the propellor hits its limits, something more efficient should be found. Jet engine inherits the ROCKET engines (look into the history of aircrafts, before the jet aircrafts there were conventional planes with additional rocket engines used for short burtst). Do you know for how many centuries before the jet there were rockets and China (fireworks)? Another source for inferring the jet engines is swimming, breathing, baloons exhausting the air inside etc. The FUNCTION (moving gas quickly) is the same for all of them, and the function of the propellor is the same as well. Other factors for shaping the jet-engine exactly as it is are the quality of the materials - weight, strenght, availability, etc. Aerodynamics - higher speeds change the requirements for the wing geometry etc. >The tablet is not inferrable from the p.c. Wrong, "PC" is a computer and the tablet is a computer, with all of the aspects of this. The only "essential" difference is in the size. The smaller size in electronics is "inferrable" by the advent of the appropriate technologies - physics. The tablet is inferrable from the following: -- People want to carry computers with themselves and then to have them in their pockets (that's one of the factors for "inferring" miniaturization) -- There are cheap enough and light enough touch screens -- There are cheap enough CPUs and other hardware -- The CPUs, screens and software are fancy enough, and the price is low enough to attract customers and make this economically profittable (There were commercial tablet computers since late 80-ies, they were just called "Palmtops" or "Handhelds" and the masses didn't know or didn't care about them, except in SF movies) -- etc. And volia, you have a tablet. It's the same principle with all inventions - no invention comes from the nowhere, it's a result of combination of prerequisites. Right, most of the people are not smart enough to know about, understand and combine the prerequisites, that's why the inventors are "special", even though theinvention is in fact systematic work and focus on the problems, and it is inferrable, as long as you know what the inventor knows or more. >I've realised that "emergent inference" is a v. useful concept - but useful; I suspect, precisely because it is a contradiction in terms. Emergence does not occur by inference but by totally different processes. LOL. Yes, emergence that you think about is some kind of a magical random change in a successful direction, which you can't understand/explain. The random changes, whose origin the system doesn't understand , don't help it to progress intentionally, it's a mystery thing. We're interested in the progress which is understandable and can be modelled. > ONe way of highlighting this is : can you give one example of any new kind of logic that has been "inferred" from previous kinds? Or any one new kind of geometry, that has been inferred from previous kinds? .How >could topology have been inferred? Or freeform geometry? Or Riemannian geometry? How many times should we explain about sensori-motor generalization. Study the history of human knowledge and mathematics and the lifes of the mathematicians and you'll see how they inferred everything. Not from the math itself, from their needs and the limits. Geometry as a field is inferred from sensory inputs as shortcuts for calculating and dividing areas on the ground. Wherever there's a demand created by a problem that can't be solved using existing systems, something is updated or invented in order to solve it, and this succeeds if the prerequisites are available. -- Todor http://research.twenkid.com http://artificial-mind.blogspot.com ------------------------------------------- AGI Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-c97d2393 Modify Your Subscription: https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-2484a968 Powered by Listbox: http://www.listbox.com
