John, The proper formulation of the problem is:
**what generative form (such as a formula or algo) will generate *ever new diverse forms* [or objects] within a given class so as ultimately to embrace more or less the whole ever expanding range of diverse forms within a class ** notes: 1) it is crucial to take a step back and realise that there are many “generative forms” (or would anyone prefer “generators”?) - many scientists and AI-ers think that formulae/algos are the only generative forms – that nothing can proceed without one. But this, as I will come to another time, is nonsense. 2) the proper focus is on generating *ever new diverse forms* – the reality for any real world AGI of getting to know classes, is that you begin with one or a few examples, and continue to encounter ever more diverse examples - you start by seeing one dog – a terrier say - and then go on to meet ever new diverse forms – a chihuahua, a dachsund etc ; ditto you start by seeing one kind of chair and then encounter ever new diverse forms; you start by meeting one crazy AGI-er, or individual within any human group/class whatsoever - one who thinks it’s all about complexity – and then go on to meet new diverse AGI-ers/individuals – s.o. who thinks it’s all about patterns, or some form of logic etc. in terms of fonts, you start by encountering one kind of font, and then encounter one new diverse kind after another – ad infinitum for the rest of your life – as Hofstadter demonstrates well. 2a) for real AGI’s like humans, there is no limit to this business of generating new forms – we can indeed go on to recognize more or less the whole of any class; there may be the odd controversial, difficult-to-classify examples – in fact this is virtually guaranteed – but in general, we have no major problems embracing more or less the whole class 2b) this is indeed a simple crystallisation of the problem of AGI – in every form of the problem, incl. conceptualisation, object recognition, creativity (object formation) - you must deal with and actively or reactively generate new diverse forms/objects within a class – and there is no set range of forms for any class – each class is open and everexpanding. I think I’ve solved this problem now pretty damn definitively at a philosophical level – why the generative form can’t be mathematical, and can only be iconic/graphic* – and why the former is and can only be narrow AI, and only the latter is AGI – but I’m still inwardly digesting. Later. P.S. *Another way of putting that is ..... why language/concepts can’t be mathematical and can only be iconic/graphic.Written. language did indeed begin with icons/graphics – and you won’t find any form of language in our ordinary culture expressed in mathematical form – check it out. From: John G. Rose Sent: Wednesday, January 16, 2013 4:54 PM To: AGI Subject: RE: [agi] The Vast Field of Cultural Icons Mike, You don't need to generate ALL, only some. Also, general object generation can use "seed" data. For example, picture a mathematical process as a movie projector and the filmstrip is seed data. The images projected are the objects generated. That's one way of looking at it... John -------------------------------------------------------------------------------- From: Mike Tintner [[email protected]] John: So what are the formulaic processes for general object generation and recognition? John, We’re making further progress IMO because we’re both looking directly at the problem. Ben (and Opencog) and the vast majority of AGI-ers won’t do that. I’m going to say more on this in a while, (including reframing the problem still more precisely), but in the meantime, you & others might care to look at a specifically algorithmic consideration of the problem (in a form well-set by Hofstadter): Is there a Universal Image Generator? https://researchspace.auckland.ac.nz/bitstream/handle/2292/3851/344cris.pdf?sequence=1 The conclusion: “though this is an abstract result, it may have import for several areas in graphics that deal with compressible signals. In essence, new representations and pattern generation algorithms will continue to be developed; there is no feasible “super algorithm” that is capable of all things.” Comments? AGI | Archives | Modify Your Subscription ------------------------------------------- AGI Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424 Modify Your Subscription: https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657 Powered by Listbox: http://www.listbox.com
