Matt, *** Quantum computing may be useful for a lot of things, but AGI isn't one of them. All the evidence says that neural networks are the way to solve vision, language, and robotics. ***
Even supposing that formal neural nets are the best way to emulate human-like AGI (which I don't necessarily agree with), why would this oppose the possibility of creating superhuman AGI using quantum computing infrastructure in manners differing from how the human brain works? It seems you're suffering from a severe limitation of the imagination here my friend ;) On Fri, Apr 2, 2021 at 9:48 PM Matt Mahoney <[email protected]> wrote: > > Neural computing is not reversible. In particular, updating a synapse loses > its previous state. The output clipping function is also not invertible > across its range. > > Quantum computing may be useful for a lot of things, but AGI isn't one of > them. All the evidence says that neural networks are the way to solve vision, > language, and robotics. > > On Fri, Apr 2, 2021, 11:14 PM Steve Richfield <[email protected]> > wrote: >> >> Two quick comments, >> >> 1. Remember my past comments regarding bidirectional computing - that >> neurons are obviously doing this, and conventional computers find it >> onerously slow doing such computations. >> 2. I was once involved in a project to solve time-dependent Schrodinger >> equations - basically, pretending to be a quantum computer. It worked - sort >> of, but there were many computational anomalies, equivalent to multiplying >> things by what amounts to be 0/0 or needing to solve non-converging infinite >> series’, so despite being able to choose between equivalent formulas, >> sometimes we got solutions, but other times not. Besides, aside from being >> super fast, it isn’t obvious that quantum computing is desirable for ANY >> real-world computations. >> >> Steve Richfield >> On Fri, Apr 2, 2021 at 11:03 AM Nanograte Knowledge Technologies >> <[email protected]> wrote: >>> >>> Ben >>> >>> Thanks for your response. My apologies for not being clear enough. I wasn't >>> suggesting running a quantum simulation on a classical machine, but >>> referring more to including the objective as a theoretical consideration, >>> design wise. Specifically, given the advancement of quantum science, >>> suggesting to not exclude the development of the quantum aspect due to the >>> lack of a quantum computational platform. In other words, suggesting a >>> holistic approach to a generalized version of AGI. >>> >>> Rob >>> >>> >>> >>> ________________________________ >>> From: Ben Goertzel <[email protected]> >>> Sent: Friday, 02 April 2021 17:05 >>> >>> To: AGI <[email protected]> >>> Subject: Re: [agi] Toward a Useful General Theory of General Intelligence >>> >>> Hmm, it seems to me that while dealing essentially with classical >>> computing hardware, it makes most sense to focus on real-number >>> probabilistic reasoning and real-number-probabilistic-types, >>> real-number-probabilistic programs, etc. >>> >>> When dealing w/ quantum computing hardware then it makes sense to >>> focus on complex-number probabilistic reasoning / types / programs, >>> etc. >>> >>> I'm not convinced that simulating quantum AI algorithms on classical >>> computing is a useful practical AGI strategy... >>> >>> However I do want to give Hyperon a sufficiently flexible type system >>> and compute framework that one can run it on a computer with a >>> Quantum Processing Unit and it can manage the interface w/ the quantum >>> AI software running on the QPU via elegantly and efficiently >>> interleaving complex-number-probabilistic-types w/ >>> real-number-probabilistic types. Basically Hyperon won't in its >>> initial classical form run on QPU but it should be able to treat QPU >>> processes as monads and reason about their states using proper >>> treatment of amplitudes in its type system... >>> >>> This is def. not the main focus now though, it's just something that >>> will come along for the ride if we design things right... >>> >>> ben >>> >>> On Fri, Apr 2, 2021 at 2:57 AM Nanograte Knowledge Technologies >>> <[email protected]> wrote: >>> > >>> > Thanks Ben >>> > >>> > Well put and deeply appreciated. However, I do not think one has to wait >>> > before including quantum wave functions into the architectural design for >>> > generalized AGI. The knowledge is involving so rapidly, and research into >>> > pinpointing the connection between partial collapse and the environment >>> > are advancing rapidly. >>> > >>> > Granted, it would depend on the approach taken to developing a >>> > generalized version of AGI. However, the key to unlocking true quantum >>> > computing, including AGI as a quantum-enabled mind, probably resides in >>> > the nuclear processes inherent in the wave function, in all states and >>> > single state specifically. Understood, the need for reversibility being >>> > the control required to prevent the Halting problem from occurring in the >>> > environment. >>> > >>> > For example, I think Musk's AGI-related efforts in particular seem to >>> > favour this approach, to enable the transformational mechanism within a >>> > quantum computer and connect that directly to the human brain (as >>> > environment), thereby facilitating the quantum flux required for >>> > informational flow to become persistent and pervasive. >>> > >>> > As you beautifully stated in response to recent criticism about your >>> > decision to now adapt your approach to AGI development, as pure >>> > relevance, which you based on years of past learning. So should these >>> > emerging developments (even in relative infancy as new science) have >>> > relevance. Perhaps even offering the potential for making a quantum leap >>> > in development? >>> > >>> > Question: If you focussed on the core mechanism for enabling a >>> > generalized AGI, would that not make sense to now pursue that to the >>> > fullest extent? >>> > >>> > A core approach would still remain constructivist, enabling control >>> > hierarchies to be operational, but in a different sense though, rather >>> > constructing from the core outward, not via a classical approach of >>> > inverse reductionism, in the sense of a fractal approach, instead >>> > primarily relying on what emerges for the inherent nature of >>> > superpatterns. >>> > >>> > Viewing this similarly to the Euro-tunnel project, drilling from both >>> > directions. If the science-in-engineering proved correct, the tunnel >>> > would be encouraged to meet up exactly and integrate as a whole. >>> > >>> > Rob >>> > >>> > ________________________________ >>> > From: Ben Goertzel <[email protected]> >>> > Sent: Friday, 02 April 2021 07:17 >>> > To: AGI <[email protected]> >>> > Subject: Re: [agi] Toward a Useful General Theory of General Intelligence >>> > >>> > > >>> > > Your thoughts on quantum entanglement and AGI? >>> > > >>> > > Rob >>> > >>> > I suspect that quantum computing will enable even smarter AGIs than >>> > digital computing, but that digital computing can serve as the >>> > substrate for AGIs with greater than human GI ... >>> > >>> > I have argued recently that quantum probabilistic logic is a close >>> > approximation to (an isomorphically transformed version of) uncertain >>> > paraconsistent logic, >>> > >>> > https://arxiv.org/abs/2101.07498 >>> > >>> > -- this direction of thinking may ultimately help w/ the design of >>> > quantum reasoning engines. For quantum reinforcement learning of >>> > quantum neural nets and similar, I suspect evolutionary learning may >>> > be a good approach, with mutation and crossover operations that can >>> > occur w/in uncollapsed systems w/o causing collapse. I also think >>> > fully exploiting the power of QM for AI may require understanding weak >>> > measurement and "partially collapsed" quantum systems like one sees in >>> > quantum biology. There are a lot of exciting directions here but I >>> > currently doubt they are going to be necessary for the first >>> > breakthrough to human-level AGI. They may end up being breakthroughs >>> > made *by* the initial digitally-implemented human-level or moderately >>> > transhuman AGIs. >>> > >>> > -- Ben >>> > Artificial General Intelligence List / AGI / see discussions + >>> > participants + delivery options Permalink >>> >>> >>> -- >>> Ben Goertzel, PhD >>> http://goertzel.org >>> >>> “He not busy being born is busy dying" -- Bob Dylan >> >> -- >> Full employment can be had with the stoke of a pen. Simply institute a six >> hour workday. That will easily create enough new jobs to bring back full >> employment. >> > Artificial General Intelligence List / AGI / see discussions + participants + > delivery options Permalink -- Ben Goertzel, PhD http://goertzel.org “He not busy being born is busy dying" -- Bob Dylan ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T9a211170d976967d-M948561899de46c9a2ce4c2cf Delivery options: https://agi.topicbox.com/groups/agi/subscription
