Hi Dorian++ I am grappling with the final sections of the paper... So it's topical.
Making a case for the formal recognition of S-AGI & H-AGI as real alternatives, neglected at the loss of something empirically impossible otherwise.... Can stand on its own feet as argument. But my slant on your suggested issue ... The existing analytic-AGI's grip on the AGI goal is empirically well-exemplified by the very obvious generalized and recognized slippery slope to narrow-AI outcomes down the decades. These domain-bound outcomes may be agreed to be particular instances of an analytic-algorithmic 'cracking' of some aspect of cognition solved by brain tissue. The extent to which the exact function exists in the brain is never actually argued. At least I have never seen one. Indeed it may be that the idea that any particular functional 'part' being surgically excised from the brain's 'whole' ... Is a deeply flawed idea only resolved at the level of the 'whole'. ... And again enter synthetic AGI as a route to dealing with that problem. That is, the very nature of deciding when a certain function has been 'cracked' is, in itself, the very thing synthetic AGI approaches can tackle empirically. Analytic-AGI presupposes that function and its separability. In practice the closest this 'cracking' gets to being argued is in neuromorphic engineering where you can find some deeply neuro-inspired hardware-algorithmic results. In particular I can cite a visual/retinal example. But in the end even these fail to 'crack', say, vision as a analytic equivalent that is empirically proved to solve human vision. Humans always get reinserted as a judge of 'vision' having occurred. š if only the guys at the Dartmouth conference could have known in 2015 we'd still be saying this! They came away from the conference thinking they'd 'crack' human vision "over the summer". Yet here we are.... The deep epistemic issue here is that in 'cracking' something in the manner of analytic-AGI one is always left with a piece of work who's understanding and limits were defined by the humans involved. However useful it may be this is not empirical neuroscience. So my slant on this is to capture this kind of discussion as an indicator that Analytic-AGI being joined by synthetic-AGI approaches is a shift in the nature of the science itself ... It's empirical options and in critique. The introduction of S-AGI changes the landscape of the discourse itself such that the specific 'cracking' contrasts the approaches less than thought because it's never 'apples vs. apples'. Comments? I could package up the specific neuromorphic eng. Example to illustrate the point. Perhaps you have examples of your own? Regards Colin Hales -----Original Message----- From: "Dorian Aur" <[email protected]> Sent: ā3/ā06/ā2015 2:34 AM To: "AGI" <[email protected]> Subject: Re: [agi] H-AGI towards S-AGI Colin et al, We need to ask another set of questions, answering them may provide the required perspective. What kind of human like tasks have been algorithmically cracked so far by AI? How are they solved by our brain? Are these tasks considered āintelligentā by our standards? Dorian On Tue, May 26, 2015 at 4:16 PM, Colin Hales <[email protected]> wrote: Thanks Dorian! Will integrate the edits. Meanwhile: here is the next (short) section: ======================================= 3 Synthetic AGI and embodiment (robotics) It may be a surprise to learn that there is a sense in which synthetic forms have been present since the birth of AI. It occurs in the form of a robotic body. When an analytic A(G)I is clothed in any kind of body there is a tacit acceptance that synthesis is the only thing that physically places the AI in the world with us. Arms. Legs. Torso. Hands. Cranium. In a robot what we are doing is making an inorganic synthetic version. We may place a natural human brain in a completely inorganic body. We may synthesize an organic human brain and put it in a completely inorganic body. We may synthesize an inorganic brain and install it in an otherwise completely organic human body. In all of these cases this choice reflects the knowledge of the deep necessity for arm-leg-torso-hand-cranium physics. It is so obvious to all of us we donāt even think about it. In the embodiment of a robot, without that physics there is no robot. The physics, as a specialized sensory motor apparatus, is far from its natural organic counterpart and is obviously not directly involved in the intelligence of the intelligence of the brain running it all. Yet, from our perspective as makers, we implicitly agree, and have done all along, that synthetic peripheral sensory/motor system physics is essential or there is no robot in the world with us. Is there an analytic counterpart to this? Yes: the virtual world. This is where the analytic brain is placed in analytic sensory/motor āclothesā and then placed in an analytic virtual world. This has been used to great effect in analytic AGI developments. Off-shoots of this technology are now also routinely used to place humans inside an analytic virtual world (immersive virtual realities) or to overlay the virtual world onto our own natural world (āhead-upā display such as Google glass). In many ways analytic āroboticā clothing is nearly as old as the physical synthetic āclothingā that is robotics. This view now presents us a way to contextualize the introduction of synthetic approaches as merely an expansion from peripheral/sensory/motor systems onwards, deeper into brain tissue. Is there specialized brain physics that is literally as essential to robot intelligence as arms and legs are to its embodiment? The answer to that question involves the use of synthetic AGI approaches and their contrast with the analytic equivalent. Be it organic or inorganic, the actual boundary of essential physics deep within brain tissue is actually unknown. It has been assumed all along to be the peripheral/sensory/motor boundary. The introduction of synthetic AGI facilitates a scientific evaluation of that assumption, filling a gap in knowledge that has been there all along. It can now be seen that any claim to deep novelty in synthetic AGI is actually unjustified. The synthetic approach has, in a sense, been there all along. All that the new program of works is doing is moving its boundary deeper into the brain and then joining analytic approaches in scientifically evaluating whether the new boundary is essential in some way in the scope and kind of intelligence expressed by each. ========================================= The next section has this form: 4 AGI development approaches- a expanded spectrum H-AGI can include all forms of computations, algorithmic / non-algorithmic, analog, digital, quantum and classical since biological structure is incorporated in the system. 4.1 An example H-AGI biological/organic synthesis + Analytic (HYBRID) DORIAN 4.2 An example H-AGI: partial inorganic + Analytic (HYBRID) COLIN 4.3 An example S-AGI: totally inorganic synthetic (SYNTHETIC) COLIN ============================= here's a lot of referencing to be done too. I am so buried in unviewed emails.... but I gotta go. I am painting and shovelling and .... I wish I had a robot. :-) regards Colin Hales On Wed, May 27, 2015 at 4:46 AM, Dorian Aur <[email protected]> wrote: Colin et al, That's a good introduction to consciousness, we need a more direct/ practical approach to AGI - the hybrid system can be the fastest and less expensive approach to AGI and anyone from computer science, electronics, nanotechnology to neuroscience can contribute. 4 The hybrid approach to AGI The origins of the entire problem started a few decades ago when by mistake action potentials were approximated by stereotyped digital events. As a result many scientists were encouraged to imagine that brain computations can be thoroughly simulated and mapped on digital computers using connectionist models. It became a mob opinion and in spite of recent refutation, this flawed view continued to be sustained and all brain initiatives followed this vision. "Donāt be trapped by dogma, which is living the results of other peopleās thinking for six decades." Understanding the brain language and the development of AI techniques are highly co-dependent.To understand the main problem we can start with two relevant examples of algorithmic development. a. The simulation on digital computers can faithfully reproduce the characteristics of the flight b. āRealisticā models of neurons (e.g. Hodgkin-Huxley) simulated on a digital computer do not succeed to display or generate intelligent behavior This gap between (a) and (b) can be easily explained. In the first case the simulation on a digital computer is successful since the model is able to realistically include the physics of flight. In the second case biological structure uses molecular/quantum computations to integrate meaningful information . Such biophysics responsible for intelligent behavior is not included in current models ( e.g. . Hodgkin-Huxley) neither in any AGI attempts. Since molecular/quantum computations can be hardly reproduced on digital computers replicating these computation using any algorithmic approach is far more difficult.We already know that wiring together a set of non AGI systems may never generate AGI. What is the solution? We know that the loss of natural biophysics leads to issues in case of the second model . Clearly, to solve the problem one needs to find a way to include the full model of computation generated within biological structure . Having built a system that evolves in a similar way our brains do will solve the problem and guarantee that the resulting ācomputing machineā will be able to integrate meaningful information.At least two phases are needed to construct a mind using biological building blocks A.The first phase will require growing a biological structure either from natural stem cells or from induced pluripotent cells. Providing nutrients, oxygen and environmental interaction is needed to shape the structure and control spatial organization of cells . B. The second phase will create a virtual world in which the evolving biostructure can be trained to learn and experience live scenes following a specific gradual program. It is likely that after training the hybrid system will be able to mimic human behavior in the ārealā world. The first phase will require developing a system and technology to grow a biological structure. The entire development will be regulated using a computer interface equipped with microcontrollers and different nanosensors. The digital computer will obtain real-time information regarding the state of the evolving structure and detect the need of neurotrophic factors, nutrients and oxygen. This phase will allow biological building blocks to self-assemble and organize into discrete, interdependent domains. Different ways to deliver nutrients, oxygen, and achieve spatial and temporal control of living tissue by manipulating molecular and genetic technology can be explored (Delcea et al., 2011; Lewandowski, et al., 2013; Takebe et al., 2013; Deisseroth and Schnitzer, 2013; Wickner and Schekman, 2005). Dielectrophoretic actuation will be used for cell manipulation to shape the evolving 3D structure (Pethig et al., 2010; Reyes, 2013; Velugotla et al., 2012). In addition, carbon nanotubes will provide the physical support for development. They can be used to create conductive structures to perform bidirectional communication between the evolving biostructure and computers. This will allow monitoring the evolution of neurons, glial cells, ... delivering neurotrophic factors and engineering all structures. The second phase will require to build bidirectional communication between the evolving brain and the computer to create a virtual world and enhance learning. One can read and interpret the information processed in the evolving structure by using data recorded from different nanosensors. Using computer technology a virtual world will be able to provide accelerated training. Substitutional reality will enhance learning, the evolving brain will be able to mimic human behavior in the real world. The entire model can be schematically conceptualized as an interactive training syste [The entire original message is not included.] ------------------------------------------- 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
