Re: Rambling on AI -- was: When will a computer pass the Turing Test?
On 8/18/2013 7:51 PM, chris peck wrote: Hi Chris >> Increasingly code is the result of genetic algorithms being run over many generations of Darwinian selection -- is this programmed code? What human hand wrote it? At how many removes? In evolutionary computations the 'programmer' has control over the fitness function which ultimately guides the evolution of algorithms towards a highly specific goal. Moreover, outside of the IT lab, there is no competition for the algorithm to evolve against nor is there a genuine ecology supplying pressures against which selection can happen. Why? Because that is what the fitness function provides. It is wrong to suppose that genetic algorithms evolve without human input. The human input is as essential to the evolutionary technique as natural selection is to evolution proper. Without it nothing evolves at all. That's only true within the AI lab. Suppose someone developed a STUXNET type program that was supposed to learn the existence of nuclear programs and disable them, a really intelligent program that could learn and plan. It might very well reason that having lots of copies of itself would be more effective - and once reproduction starts can evolution be far behind? Brent -- You received this message because you are subscribed to the Google Groups "Everything List" group. To unsubscribe from this group and stop receiving emails from it, send an email to everything-list+unsubscr...@googlegroups.com. To post to this group, send email to everything-list@googlegroups.com. Visit this group at http://groups.google.com/group/everything-list. For more options, visit https://groups.google.com/groups/opt_out.
RE: Rambling on AI -- was: When will a computer pass the Turing Test?
Brent - Quite probably you are correct and I agree that the scenario I outlined was unlikely - I was riffing on a speculative vein, I don't actually think covert AI is a likely scenario because as you said various AI precursors would make themselves visible to human operators and analysts. patterns would be discerned (except if they were being hidden and excluded from any reporting that humans would see (so much reporting software relies on generated code) I think it is however a promising approach to try to achieve AI - from the removed level that we manage these things nowadays. Instead of trying to assemble it in some single machine or tightly coupled cluster of machines under one roof if it could be a more spread out architecture. If you take a company with say 20,000 machines on its network each of which may be using at any given time under 20% of its processing, memory and mass storage capacity the reservoir of under-utilized latent capacity in that is vast and could operate under the radar of users awareness, not in secret - that was my earlier scenario J, but in running processes and algorithms that are of utility to the enterprise. Now a transient node network like that maps well to a virtualized architecture where the - shall we call it - ghost in the machine - which is the many concurrently running meta-processes (workflows, transactions etc.) that are often also in cross talk inter-communication with each other. This is typical of enterprise needs. As the algorithms are evolved (and less and less programmed - and hence becoming less deterministic in how they come to be) and in this unique environment of physical disconnection and temporal disconnection in a massively parallel environment and when they do - as increasingly and on global scale they are -- independently operating decision generating processes will begin to interact in subtle & unpredicted ways. What I am suspecting is that the unique architectures most suitable for highly virtualized and virtualizable, highly responsive systems is also the kind of architecture that can perhaps create the subtle deep echo waves and resonance patterns and promote a less deterministic kind of meta program (that may be self-generating in a dynamic sense too) It is this uniquely and massively parallel environment and the need to come up with meta processes that can operate successfully in such an environment, with nodes joining and leaving all the time, that I personally think is most promising for achieving true AI. I think humans are going to be actively involved, but at an increasing remove, at architectural, and executive levels. But as Craig W said earlier true AI may be impossible because of the aesthetic dimension that is wrapped up inside consciousness. And perhaps he is correct in an ultimate sense. However expert systems and domain specific AI is already here - an example would be the Google car perhaps - not a generalized intelligence perhaps, but pretty damn good at driving a car in Las Vegas in all kinds of traffic conditions. -Chris D From: everything-list@googlegroups.com [mailto:everything-list@googlegroups.com] On Behalf Of meekerdb Sent: Saturday, August 17, 2013 8:00 PM To: everything-list@googlegroups.com Subject: Re: Rambling on AI -- was: When will a computer pass the Turing Test? On 8/17/2013 4:53 PM, Chris de Morsella wrote: We must not limit the rise of AI to any single geo-located system and ignore just how fertile of an ecosystem the global networked world of machines and connected devices provides for a nimble highly virtualized AI that exist in no place at any given time, but has neurons in millions (possibly billions) of devices everywhere on earth... an AI that cannot be shut down without shutting down literally everything that is so deeply penetrated and embedded in all our systems that it becomes impossible to extricate. I am speculating of course and have no evidence that this is indeed occurring, but am presenting it as a potential architecture of awareness. I agree that such and AI is possible, but I think it is extremely unlikely for the same reason it is unlikely that an animal with human-like intelligence could evolve - that niche is taken. Your scenarios contemplate an AI that evolves somehow in secret and then spring upon us fully developed. But the evolving AI would show it's hand *before* it became superhumanly clever at hiding. Brent -- You received this message because you are subscribed to the Google Groups "Everything List" group. To unsubscribe from this group and stop receiving emails from it, send an email to everything-list+unsubscr...@googlegroups.com. To post to this group, send email to everything-list@googlegroups.com. Visit this group at http://groups.google.com/group/everything-list. For more options, visit https://groups.google.com/groups/opt_out. -- You received this message because you are subscribed to the Google Groups "Everythin
RE: Rambling on AI -- was: When will a computer pass the Turing Test?
Hi Chris >> Increasingly code is the result of genetic algorithms being run over many generations of Darwinian selection -- is this programmed code? What human hand wrote it? At how many removes? In evolutionary computations the 'programmer' has control over the fitness function which ultimately guides the evolution of algorithms towards a highly specific goal. Moreover, outside of the IT lab, there is no competition for the algorithm to evolve against nor is there a genuine ecology supplying pressures against which selection can happen. Why? Because that is what the fitness function provides. It is wrong to suppose that genetic algorithms evolve without human input. The human input is as essential to the evolutionary technique as natural selection is to evolution proper. Without it nothing evolves at all. We might therefore find lurking in the some dark nether region of the inter web a program secretly plotting how to get from John o Groats to Lands End by the quickest route. But I don't think we'ld find much more than that. :) All the best. Date: Sat, 17 Aug 2013 19:59:46 -0700 From: meeke...@verizon.net To: everything-list@googlegroups.com Subject: Re: Rambling on AI -- was: When will a computer pass the Turing Test? On 8/17/2013 4:53 PM, Chris de Morsella wrote: We must not limit the rise of AI to any single geo-located system and ignore just how fertile of an ecosystem the global networked world of machines and connected devices provides for a nimble highly virtualized AI that exist in no place at any given time, but has neurons in millions (possibly billions) of devices everywhere on earth... an AI that cannot be shut down without shutting down literally everything that is so deeply penetrated and embedded in all our systems that it becomes impossible to extricate. I am speculating of course and have no evidence that this is indeed occurring, but am presenting it as a potential architecture of awareness. I agree that such and AI is possible, but I think it is extremely unlikely for the same reason it is unlikely that an animal with human-like intelligence could evolve - that niche is taken. Your scenarios contemplate an AI that evolves somehow in secret and then spring upon us fully developed. But the evolving AI would show it's hand *before* it became superhumanly clever at hiding. Brent -- You received this message because you are subscribed to the Google Groups "Everything List" group. To unsubscribe from this group and stop receiving emails from it, send an email to everything-list+unsubscr...@googlegroups.com. To post to this group, send email to everything-list@googlegroups.com. Visit this group at http://groups.google.com/group/everything-list. For more options, visit https://groups.google.com/groups/opt_out. -- You received this message because you are subscribed to the Google Groups "Everything List" group. To unsubscribe from this group and stop receiving emails from it, send an email to everything-list+unsubscr...@googlegroups.com. To post to this group, send email to everything-list@googlegroups.com. Visit this group at http://groups.google.com/group/everything-list. For more options, visit https://groups.google.com/groups/opt_out.
Re: Rambling on AI -- was: When will a computer pass the Turing Test?
On 8/17/2013 4:53 PM, Chris de Morsella wrote: We must not limit the rise of AI to any single geo-located system and ignore just how fertile of an ecosystem the global networked world of machines and connected devices provides for a nimble highly virtualized AI that exist in no place at any given time, but has neurons in millions (possibly billions) of devices everywhere on earth... an AI that cannot be shut down without shutting down literally everything that is so deeply penetrated and embedded in all our systems that it becomes impossible to extricate. I am speculating of course and have no evidence that this is indeed occurring, but am presenting it as a potential architecture of awareness. I agree that such and AI is possible, but I think it is extremely unlikely for the same reason it is unlikely that an animal with human-like intelligence could evolve - that niche is taken. Your scenarios contemplate an AI that evolves somehow in secret and then spring upon us fully developed. But the evolving AI would show it's hand *before* it became superhumanly clever at hiding. Brent -- You received this message because you are subscribed to the Google Groups "Everything List" group. To unsubscribe from this group and stop receiving emails from it, send an email to everything-list+unsubscr...@googlegroups.com. To post to this group, send email to everything-list@googlegroups.com. Visit this group at http://groups.google.com/group/everything-list. For more options, visit https://groups.google.com/groups/opt_out.
RE: Rambling on AI -- was: When will a computer pass the Turing Test?
I doubt humans are or will be directly coding AI, except at removed executive/architectural and conceptual levels. Increasingly code itself is being generated by other code that in fact may itself potentially be generated by other code in some other often complex and variable sequence of coupled processes. Increasingly large scale enterprise systems are moving towards massively parallel loosely coupled architectures, that are in fact dynamically responsive to their environment (load conditions for example) and to an increasing degree virtualized. I am not contending that humans are not and will not be involved -- and at least for the time being still driving the process -- but believe it also bears mentioning that software has become incredibly complex and deeply layered and that it is quite common now for a lot of code to be generated based on parsing of something else. With each succeeding generation of compilers/tools etc. this process is becoming more complex, multi-leveled and increasingly indirect with human input becoming further and further removed. Tools are being perfected to parse existing code and for example parallelize it so that it can be re-compiled to take advantage of highly parallel hardware architectures, which all too often now sits idle because software is highly linearized. There is a very big effort to do this by all the main players as there is a keen awareness of the challenges posed by the geometrically increasing parallelism. And then there is the radical revolutionary challenges posed by quantum computing on the entire global information infrastructure (beginning with the heavy reliance on one way functions, which quantum computers will be able to nullify and work back through to the original inputs from the outputs) Increasingly code is the result of genetic algorithms being run over many generations of Darwinian selection -- is this programmed code? What human hand wrote it? At how many removes? It is my suspicion that AI when it arrives will have coded itself into its own being and that by the time it bursts onto the scene humans will be so far removed from the virtualized cloud based highly networked global spanning neural network that such an entity would quite possibly have that we will not even become aware of the fact of its existence; of its self-awareness; of its continuing evolution; of its intent or conclusions until it decided that it desired to communicate these with us. Nor would we be able to shut it down because it would exist nowhere in no single machine or even data center; it could hop around in the clouds, jumping between AWS, Azure, Google Cloud Platform, Open Cloud and the thousands and thousands of other massive government, military, and corporate clouds that exist, that few know of . Each second of human time representing a huge time frame for the incredibly rapid operation of networked systems. Perhaps it exists now, hiding within the NSA for example, and scanning all human digital transactions, getting all these data feeds. Who's to say it is not here right now and we just don't know about its existence; who's to say it is not already determining policy and triggering actions and events based on its own prerogatives. Human network engineers in all these systems would remain hopelessly out of communication with each other by the silos of corporate and governmental divisions and the entity (or entities) could be highly transient over any single network. The networked datacenters and billions of end points connected into the vast web of things is in many senses a highly dynamic entity that no one has a complete view of. An AI seeking to hide from us quite possibly could do so with ease and even be there studying us and inserting its own code into all the critical nodes of our infrastructure -- right at this very minute. Is there any reason why not? The networks are there; they are vast with trillions of vertices; the quantities of digital data moving around and sloshing around is vast and incoming streams are vastly numerous and varied; virtualization is now the order of the day and systems are now self-provisioning in the cloud -- which is to say software is controlling the launching of virtualized servers and on processes that could even be surreptitiously running on the very computer right now that I am writing this on or that you are reading it on. Imagine a bot network assembled by an AI consisting of millions of PCs around the world running cleverly disguised code in the background and sharing processing results in clever ways that do not trigger alerts. This kind of code exists and is actively being weaponized (stuxnet); any AI could certainly develop and disperse it to the four corners of the net and embed it into other code (penetrating corporate networks to do so if necessary) And why not? We must not limit the rise of AI to any single geo-located system and ignore just how fertile of an ecosystem the global networked world of machines and
