On Sat, Apr 6, 2013 at 9:47 PM, Piaget Modeler <piagetmode...@hotmail.com>wrote:
> ... just talk about my definition of semiosis: > computationally linking signs to meanings, and vice versa. > Getting a bit desperate, PM? You seem to have asked both how it might be happening in humans AND how it might happen in some artificial mind design, but since this is the AGI list I will assume you stayed on topic and meant the latter. Anyway, the various representation-less models can never be rejected in either natural or artificial brains: from input to understanding in one distributed automagic step, or better still an automagic feedback loop. In this kind of model the "explanations" we are able to generate on demand have as much validity as the explanations patients provide in all the seminal agnosia experiments, where they ask "why can't you see the apple now that you moved it to your left" and they get back "because you distracted me" except for the more accurate "half my brain is missing after the accident". Let me rephrase once again, we just do it. Whatever happens at the semiosis or any other stages is (falsely?) manufactured. Perhaps. Now in the machine intelligence case, if it is symbolic enough to withstand scrutiny, I see no mystery. The a-brain works exclusively with meanings and models and scenarios and other things that are de facto "representing unambiguously themselves" and "hopes" the signing system will work, whether with another entity or with oneself. I don't know about you but it has happened to me that I kept meticulous notes for personal use of one thing or another but ended up mystified because I chose wrong and ambiguous or too few signs (words and sketches). So sign is something that you hope encodes the meaning you want to convey. This system of hope works anyway you can make it work, you may want to subject the other entity, the counterparty, to years of repetitive education so that they stand a chance to understand your more complicated meanings. During education the meanings of signs will have to be rediscovered/reinvented by the trainees through trial and error and any other mechanism we know of, none of which is guaranteed to work. For whatever reason the "5 senses" are more like 1000, (I am including wide ranges of feelings and emotions that are universal enough to have communication and survival value), and they are pre-verbal and non-verbal to a large extent, which again brings us to a non-representational model, if some tribal Amazon leader disapproves of your first time visit and first contact you don't have to connect his frown or threatening gesture or scream to a meaning, whatever you see or hear is as good as the meaning. I will briefly remind the reader that language-creation experiments with embodied agents/robots are a tool of the highest significance in our search for AGI. In those the sign is the observable state of the agent from the outside (eg the position of its bodyparts and any sounds and blinking lights and TCP/IP traffic it produces), while the meaning is everything else. I'd risk asserting that ANY genetic programming experiment involving a multitude of agents and either a) assigning survival value to increased signal-mediated cooperation or b) driven by the ever-increasing utility value of a single agent that tries to maximize his control of all the others, either a) or b) would eventually produce AGI, albeit not particularly friendly. I would also assert that if fear and pride did not "force" humans to participate in education we would never have evolved our signaling and internal states, we'd be more basic, more visceral, than the aforementioned uncontacted tribe. AT ------------------------------------------- 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
