Chat Archive
<Nefastor> Good evening, everyone. Thanks for having me here tonight
<BJKlein> 9 pm Eastern = Alcor Chat
<BJKlein> It's not being opposed
<BJKlein> <FutureQ> Couldn't they have an appeal and good chance of it for how the gov tried slipping it by under the radar?
<BJKlein> <thefirstimmortal> nope
<BJKlein> <TimFreeman> thefirstimmortal: What's your source of information?
<BJKlein> sorry..
<hkhenson> evening nefastor
<FutureQ> hello and welcome
*** Joins: John_McCluskey (~John@modemcable135.5-202-24.mc.videotron.ca)
<hkhenson> <<--Keith Henson
* BJKlein Chat Topic: Robotics, AI, and Immortality
<BJKlein> French engineer
<hkhenson> in and out though
<BJKlein> French engineer, Jean Roch (nefastor) joins ImmInst to discuss his work on artificial neurons and full-body prosthetics as a pathway to physical immortality
<BJKlein>
http://imminst.org/f...=ST&f=63&t=3070<cyborg01> Hi Nefastor
<Nefastor> I'm convinced cybernetics can be the fastest way to immortality, and I'll try to convince you it is
<BJKlein> excellent.. you wont' have much trouble from most of us.. i'd guess..
<Ocsrazor> I agree with you wholeheartedly Jean!
<TimFreeman> So you keep a biological brain, or upload?
<Nefastor> Both are possible
<cyborg01> Me too Nefastor
<Nefastor> Keeping the human biological brain, however, means higher maintenance
<cyborg01> Uploading is even less expensive
<cyborg01> BCI will take some time to develop
<Nefastor> I use the term "inloading" for the replacement of neurons inside the skull
<Ocsrazor> Jean I would say that in the long term you are correct, but I have doubts about the short term
<Nefastor> I too thought that cybernetic was still far in future, but recent developments are proving me wrong
<gustavo> which ones?
<Nefastor> The development of polymer artificial muscles in an important one, of course
<Nefastor> But not the most important
<FutureQ> inllading is more to my liking then discontinuitous destructive uploding
<Nefastor> So far, the biggest limitation had been the energy source
<Nefastor> And it being solved as we speak
<Ocsrazor> I'm curious as to why you think inloading will be necessary Jean, why replace when you can interface?
<cyborg01> Energy source for wat>
<Nefastor> Energy source for a cybernectic body, of course
<Nefastor> Inloading isn't necessary, it's only possible
<cyborg01> The body is not the more difficult part
<Nefastor> Inloading has its advantages, such as simplifying maintenance of the brain
<Nefastor> The body, indeed, isn't the most difficult part
<TimFreeman> Is the energy source a problem for inloading too?
<Nefastor> But so far, powering it was
<cyborg01> I have a web site on inloading:
http://www.geocities.com/softuploading<Nefastor> The problem with cybernetics is the reliance of technology upon electrical energy
<Nefastor> The body doesn't use electricity for power
<cyborg01> I doubt if that's really a problem
<cyborg01> Just plug in batteries or watever
<Nefastor> So that use to make machine and living tissue not too compatible
<Nefastor> Batteries offer very limited autonomy
<Nefastor> Machines like "Terminator", are very unrealistic in terms of power
<kzzchX> supercaps and H fuel cells maybe?
<cyborg01> You're right Nefastor, this is a robotics issue
<Nefastor> I had thought of using an hydrogen PEM cell on one of my robots, but a better solution has presented itself
<Nefastor> I will direct you to this simple bit of news :
<Nefastor>
http://www.matr.net/article-7673.html<Nefastor> Which is about the recent discovery of an enzyme that turns sugar into electricity
<FutureQ> Nybe miniianture gas turbines and biodegrade produed methne?
<Nefastor> This enzyme bridges the (energy) gap between life and machine
<cyborg01> Wow... posthumans may still be eating food... what an anticlimax
<Nefastor> I also had thought of MEMS gaz turbines. They remain a good (but expensive) option
<FutureQ> So robotots will be suar fiends
<TimFreeman> I don't see the unrealism in the power requirements for a Terminator type. Very small nuclear power sources are covered in Nanomedicine, and a purely robotic system wouldn't even need it to be very small.
<Nefastor> Indeed cyborgs will still eat food
<Ocsrazor> Something else you might be interested in Jean is the use of NK ATPases on artificial membranes
<Nefastor> Very small nuclear reactors are impossible
<cyborg01> But electricity is pretty cheat isn't it?
<cyborg01> Centralized power plants will do the job
<Ocsrazor> but difficult to store cyborg
<Nefastor> The problem with fission is that it only occurs when there's enough fissile material in a small space.
<cyborg01> No need for each of us to think about food all the time
<cyborg01> I mean, you got a rechargeable battery, and any where you go there will be power outlets
<Nefastor> The thing about using glucose-electricity conversion is that a single source of energy (food) power both the bio and mechanical parts of the cyborg
<FutureQ> I kind of like the idea of culinary bots
<gustavo> I was thinking of keeping my natural brain but getting an artificial, robotic body. Now you tell me it's the other way around.
<FutureQ> robot gormets, hehe
<kzzchX> There's an interesting thought, is the process reversible? Battery powered humans no longer needing to eat? Solar powered even?
<Nefastor> There aren't AC outlets everywhere you go, on this planet
<BJKlein> Nefastor, how would you define the technological singularity?
<cyborg01> Nefastor: the neuromatrix will be quite large in size... an early upload will not be able to move around easily, anyways
<FutureQ> or off planet?
<Nefastor> I'm not sure of your use of the word singularity in this context (I'm french remember
)
<TimFreeman> Sorry to mislead. I agree about very small fission being impossible. The nuclear power source in nanomedicine was just a speck of radioactively hot mineral and a heat engine; no ongoing fission. Lifetime was about 10 ...
<TimFreeman> years.
<BJKlein> heh.. when computers become smarter than human.. basically.. and they have access to their own source code = self improvement at the speed of light
<Nefastor> I will shortly talk about small fission power
<John_McCluskey> Jean, I've got a question about uploading... Have you thought of any method to attempt to make a backup for a human brain? Anything short of cryonics?
<TimFreeman> I agree that using glucose is more elegant, assuming you have a lot of biological tissue to support too.
<cyborg01> John_McCluskey: I think that's technically possible
<John_McCluskey> Using 2004 technology? Tell me more...
<Nefastor> In order : small fission power is possible but inefficent, run a search on SNAP-9A and NASA. The complexity of the brain is less great than it appears, in any case it is a finite system and can therefore be replaced by another finite system.
<cyborg01> Not 2004 tech... but within a few decades
<Nefastor> I do not intend to backup the brain : I'm a bit afraid of seeing disembodied copies of me haunting my networks
<FutureQ> Ummm, not so finite if you conider plasicity
<John_McCluskey> Ah, then we can do nothing except watch our brains slowly decay until then... I'm looking for ways to begin this process in this decade.
<cyborg01> Nefastor is your research mainly in ANNs?
<Nefastor> Plasticity is mainly in synaptic coefficients (speaking of neural networks simulation). I'm in the process of creating a supercomputer designed for that kind of work, with AMD
<cyborg01> Cryonics will do if you arnt averse to it
*** Joins: sjvan (~S.J.@adsl-68-74-193-31.dsl.milwwi.ameritech.net)
<Nefastor> Yes, at the moment my work is on very large NN's
*** Joins: sellinios (~sellinios@195.167.90.98)
<cyborg01> You mean VLSI ANN implementation?
* John_McCluskey thinks we can wait unti 9 pm to address the Alcor situation.
<Nefastor> Not a VLSI
<cyborg01> So... software
<Ocsrazor> Jean there is a great deal more to plasticity in the biological substrate than just synaptic coefficients
<Nefastor> I have designed a supercomputer architecture optimised for NN simulation, and which can scale high enough to someday emulate the entire human brain (within the next five years)
<sellinios> hi immortals !!!!!!!!!!
<Ocsrazor> this is one of the major problems with ann's compared to real nn's
<MichaelA> John McCluskey, are you related to Peter McCluskey?
<cyborg01> Nefastor: you think cluster architectures are insufficient for that job?
<kzzchX> Nefastor: In real time?
<Nefastor> Indeed
<Nefastor> Clusters have too limited interconnect capacity
<Nefastor> The brain is a processing memory, so interconnect capacity is much more important than processing power
<cyborg01> Right... the von neumann bottleneck
<Nefastor> I have therefore designed a machine that has a huge capacity, fine-grain interconnect
<FutureQ> That's the fundamntal diffference IMHO between brain and computers
* John_McCluskey is not related to Peter McCluskey (at least 2 or 3 generations back)
<Nefastor> This machine will soon be developped into hardware, with help from the Brookhaven National Laboratory
<Ocsrazor> its that and the speed of the dynamics FutureQ
<cyborg01> What kind of architecture is it?
<Nefastor> My new architecture tries to bridge the gap between the brain and the computer : it's not a known architecture
<MichaelA> Gotcha
<TimFreeman> Hmm, are there brain chemicals that are important to cognition that diffuse a ways? Then you have an event in one neuron influencing nearby neurons that aren't connected. But the bandwidth of this diffusion couldn't be ...
<TimFreeman> high, so maybe it would be easy to simulate.
<kzzchX> Is it clocked or asynchronus, or a little of both?
*** Joins: davidbe (~davidbe@pcp08866359pcs.sabrna01.az.comcast.net)
<John_McCluskey> Nefastor: Is this supercomputer at all similiar to the Octigabay System? (
http://www.octigabay...ons/product.htm )
<Nefastor> NDA's have been signed so I can't get into much detailed, but I'll just say that the machine has been designed with one goal : being powerful enough to emulate a human brain in real time
<Nefastor> OctigaBay's machine is complitely different from mine
*** Joins: paul (~paul@24-205-216-227.cs-cres.charterpipeline.net)
<Nefastor> So are the following : QCDOC, IBM BlueGene, Cray RedStorm
<John_McCluskey> Using any FPGA's? (I work for Xilinx )
<Nefastor> We are using FPGA's
<Nefastor> But probably not Xilinx, sorry
<John_McCluskey> Please don't say the "A" word.
<Nefastor> I won't
<Nefastor> But we'll probably use A's products
<TimFreeman> Do you use floating point to simulate the neurons, or some other arithmetic?
<Ocsrazor> excellent, I was just about to ask about the dynamics - biological neural networks rewire on the tens of millisecond time scale - will you be able simulate this?
<Nefastor> The neurons are simulated using a specific model I crafted for low-performance computers. I call it the "1-bit neuron"
<FutureQ> At lestit's not the "M" word
<John_McCluskey> Hmm... Stratix2 multipliers top out at 370 Mhz... our analysis is that it will be hard to get much above 250 MHz in real designs.
<TimFreeman> But if they're 1-bit neurons, I'm not sure you need to multiply.
<Nefastor> Don't worry, I'm not using the FPGA's embedded DSP blocks
<John_McCluskey> Ah... lower resolution... 18 bits is overkill anyway.
<cyborg01> Simulation can be somewat independent of architecture
<Nefastor> The usual neuron models aren't useful for what I have in mind. The 1-bit model operates in the time-domain
*** Joins: sellinios (~sellinios@195.167.90.98)
<Nefastor> The 1-bit neuron's accuracy depends only on temporal accuracy in the system, and then again, plasiticy helps us here
*** Joins: planetp (~planetp@24-205-216-227.cs-cres.charterpipeline.net)
<Nefastor> Actually, my model is very close to how real neurons work
<cyborg01> Can your machine do floating point at all?
<Nefastor> Data is represented as a PWM signal
<planetp> Nefastor - are you saying you account for all the complex neurotransmitter activity?
<Nefastor> Yes my machine is a super-scalar supercomputer
<cyborg01> We don't really know what real neurons do at the moment
<planetp> I agree
<John_McCluskey> I take it you have looked at Hugo deGaris's designs and moved beyond them?
<Nefastor> We have a good idea of it, though
<gustavo> Nefastor, since you are emphasizing so much the similarity between your model and a real brain: do you think your machine will be conscious? Will it have intentionality?
<cyborg01> As long as the architecture is generic, it can simulate the brain, given enough teraflops
<Ocsrazor> actually cyborg, we have a damn good idea of how single neurons operate
<kzzchX> I'm thinking i need to read this theoretical neuroscience text i've got
<Nefastor> Nope, I started from the ground, although I recognise Kohonen's influence in my work
<cyborg01> Ocsrazor: which model is that?
<Nefastor> The Kohonen model includes forgetting
<cyborg01> You mean SOMs?
<Nefastor> It works a bit like the Hebb's rule in reverse
<Ocsrazor> the statistical nonlinear models can closely represent the operation of single neurons
<cyborg01> I'm working on figuring out that model now
<Nefastor> It's not a recent model so I think all NN books talk about it
<cyborg01> I don't think we have such a model yet.. except the compartmental model
<Nefastor> From the Kohonen model, I moved to a model working in the time-domain and suitable for implementation on 1-bit processors
<John_McCluskey> I'm beginning to realize that it takes many clock cycles for your neurons to decide on an output.
<Nefastor> There is a bunch of real-life prototypes in my lab, giving life to robots
<Nefastor> John, thinking in terms of cycles is what I try to avoid in my model
<cyborg01> What's a one-bit processor?
<Ocsrazor> if he can get it down to a millisecond he is in the ballpark John
<Nefastor> Because it implies a clock, and synchronisation, which don't exist in our brain
<John_McCluskey> self clocked or completely asynchronous logic? tricky in VHDL.
<kzzchX> heh, real life neurons don't have a clock signal
<Nefastor> Everything is tricky in VHDL
<cyborg01> Unless you're talking about analog?
<Nefastor> I tried analog neurons
<Ocsrazor> but they have to respond to their neighbors within a given time window kzzch
<planetp> Unless you account for the complex neurotransmitter activity you are not representing anything close to human cognition, other than the associate/pattern recognizing/thinking portions.
<FutureQ> Don't tell the watche makers wwhat's afooote
<Nefastor> But I couldn't easily implement learning in fully analog hardware
<TimFreeman> You bother building robots? Are simulated environments inadequate?
<Ocsrazor> mammalian cortical neurons have a critical period of roughly 100 ms...
<Ocsrazor> they are extremely sensitive to temporal correlation wtih their neighbors
<kzzchX> but a clock implies a driving signal coming from a central location, so i stand by my claim.
<FutureQ> sim'd environment is inadequate for me, that's for sure. Who does the tech support?
<kzzchX> critical response window != clock
<cyborg01> Maybe a central clock is not necessary
<Nefastor> Vector machines aren't appropriate either, you know, Kz ?
*** Joins: Guest (~Guest@ip68-2-190-124.ph.ph.cox.net)
<Ocsrazor> no kzzch, its a bottom up clock not a top down one
<Nefastor> A central clock isn't necessary, actually the firing of neurons is highly desynchronised
<cyborg01> I'm only aware of Kohonen's work on network models, not single neuron models...
*** Joins: Guest1 (~Guest@ip68-2-190-124.ph.ph.cox.net)
<kzzchX> well, if you want to keep redefining the word clock until your right, be my guest.
<TimFreeman> Either you do your own tech support on the mechanical robot, or your own tech support on a simulated environment, unless there are good enough commodity mechanical robots out there.
<Nefastor> Technically, you don't work on single neurons, Cyborg, always on networks
<John_McCluskey> A central clock is extremely handy if you are doing time division multiplexing to implement virtual neurons.
<Ocsrazor> single neuron models are very robust, the network models are very poor right now
<Nefastor> Who would implement TMD in any kind of modern supercomputer, John ?
<cyborg01> Ocsrazor: I don't know of a single neuron model that is as good as the compartmental model
<Nefastor> TDM doesn't scale to buses of 100's of MHz
<cyborg01> TMD=?
<FutureQ> Tim, I mean if I'm nside a box, not walking around the so called real env, then who minds the supercomputer? Do I trust them to not shut it and me and my friends off, take a nap ,amnd oops!
<John_McCluskey> It seems to me that given the Gigabytes of DRAM possible in a small place, you can store a lot of neuron data in a stick of DRAM.
<Nefastor> TDM : Time-Division Multiplexing
<Nefastor> Indeed
<Ocsrazor> cyborg -> see Spiking Neuron Models by Gerstner as a good reference
<TimFreeman> FutureQ: I was talking about the machines Nefastor mentioned having in his office. Putting you in the box isn't a current plan. :-).
<Nefastor> The machines I have in my office include robots
<Nefastor> They aren't "boxes"
<John_McCluskey> Do the robots have full duplex IO? Input sensors feeding the network, which drives the motors?
<FutureQ> I have a more mundane question. I see cyborgism rising out of replacements for disabled people's limbs susch as my own useless legsw. How long befreI can lo[p these off and get walking on bionics and how will the attch?
<Nefastor> Indeed
<Nefastor> My robots use reflex-based control logic to achieve high-agility motion
<John_McCluskey> Damn! you should sell them on ThinkGeek!
<BJKlein> as this may shed more light on your work/thinking about BCI, can you tell us why you feel cryonics is not good and/or possible?
<cyborg01> Interfacing is a serious problem.. especially if you take uploading into consideration
<Nefastor> Neural-networks are implemented on 8-bit microcontrollers and are still fast enough
<Nefastor> I didn't say Cryonics wasn't good nor possible... I'm simple not too sure about them
*** Joins: John_Ventureville (~John_Vent@24-117-207-248.cpe.cableone.net)
<BJKlein> ahh.. sorry
<John_Ventureville> hello everyone
<cyborg01> Will your design be available to the public, or just for industrial use?
<FutureQ> Tim, I see, it sounded like but wasnt, gothca.
<Nefastor> It seems cells die when you freeze them
<BJKlein> die?
<Nefastor> So I don't feel like killing my brain cells
<BJKlein> can you define 'die' in your words?
<kzzchX> Penetrated with ice shards?
<Nefastor> Because of the formation of ice in the cells, which rips the cellular membrane
<John_McCluskey> Yours is still good and fresh, obviously... just wait another 30 years :-)
<John_Ventureville> what about advanced nanotech putting those displaced atoms and molecules back into working order?
<TimFreeman> Um, Alcor has been doing vitrification lately, they claim. No ice in the brain.
<BJKlein> thus, you're really saying it's impossible to put it back together, no?
<Nefastor> In 30 years I'll replace my dying neurons with nanomechanical ones
<John_Ventureville> Nefastor, how old are you?
<Nefastor> I THINK it's impossible, I haven't followed all the developments in Cryonics
<Ocsrazor> sorry Tim as far as I know with vitrification you still get crystals right now - not perfected yet :(
<Nefastor> I'm 27
<cyborg01> Nefastor: BCI is even harder than cryonics
<Nefastor> I'm not familiar with the acronim, what's BCI ?
<cyborg01> Brain computer interface
<John_Ventureville> so you are young enough that you have time to see your technology really mature and help you
<Ocsrazor> I would disagree very strongly cyborg
<Nefastor> Ah OK
<Nefastor> Yes I'm actually thinking that before I'm too old technology will have matured and be ready for keeping me alive
<kzzchX> I'll agree with one of you when either technology is actually working as planned, until then, why bother saying which is harder?
<cyborg01> I take cryonics to mean destructive scanning and uploading
<Nefastor> But I'm taking things in my own hands
<FutureQ> Note: on a popular sci-fi show "MutantX" they protrayed "vitrificaton" using buzz words like ischemia, suspended animation, and free raicals. It will play again next sunday on the WB ch.
<Nefastor> Too bad I don't have TV
<Ocsrazor> cyborg - what do you do with the scan assuming you can get it?
<John_Ventureville> Ocs, as I understand it you get "microfractures" with vitrification which when compared to the old style of freezing are nothing to worry about
<cyborg01> Ocsrazor: why, do the simulation of course
<Nefastor> Still, the neurons' membrane holds all our memory, damaging it is damaging your very self
<Nefastor> The neuron's membrane isn't just a container
<Ocsrazor> agreed nefastor
<cyborg01> Some distortion is unavoidable
<Nefastor> That is why I don't like cryonics too much
<TimFreeman> John_Ventureville: Right, the current vitrification does fracture because of the low temperature storage, mostly. Comparison with the old technique isn't pertinent to Nefastor because he's rejecting both.
<John_McCluskey> Jean, assuming someone (with CIA or Bill Gates level of resources) builds a human level machine with your serial neuron architecture... How the hell are you going to program it? let alone download a particular human brain?
<cyborg01> I'm skeptical about it too... but BCI is *much* harder
<Nefastor> But I can't tell if in 500 years we won't be able to reconstruct lost memory from a frozen brain
<Ocsrazor> Dendritic Spines and their dynamics ARE the substrate of memory itself
<Jonesey> futureq:mutant x is a neat show though low budget
<Nefastor> John, I don't know how a brain model will be programmed in my machine : right now I'm only making the hardware
<cyborg01> Ocsrazor: those are only short-term memory
<TimFreeman> How do you test the hardware before you can program it?
<Ocsrazor> from someone who works directly in BCI cyborg, I think this is a much less difficult problem than simulating consciousness
<Nefastor> However, I have given a little thought to a neuro-description language, kind of VHDL for neural networks
<Ocsrazor> cyborg they are ALL memory
<FutureQ> yeah, I've keptup with it. I try to watch for how our memes are dissiminated to the public.
<Ocsrazor> both long and short
<cyborg01> Ocsrazor: you have repeated missed the probing density problem... right now our electrode arrays are pathetic
<Ocsrazor> but quickly increasing in resolution
<Ocsrazor> our field is a prerequisite to building accurate enough models to simulate
<cyborg01> How many neurons you need to interface with? close to 100 billion!
<Nefastor> Regarding BCI, I believe in direct connection to neurons, in vivo. It will possible soon if I'm not mistaken
* John_McCluskey thinks that a language for neural network design is *very* interesting.
<Ocsrazor> Agreed Jean
* BJKlein official chat ends in a few minutes.. feel free to stay longer to discuss Jean's work and then Alcor
<gustavo> Nefastor, sorry for insisting in my previous question: do you expect your model to replicate "human" phenomena such as learning, emotions, conscousness, intentionality, envy, personality, etc.
<gustavo> love
<Nefastor> Ideally, the signal for each connected neuron will travel through a carbon nanotube
<cyborg01> Nefastor: I think the BCI at least requires a little nanotech
<FutureQ> Mutantx, the one I told about vcr codes Sat, 28 1:00 PM 9 WGN 205008 Sun, 29 1:00 PM 3 KWBP 25184 title "A Normal Life".
<Ocsrazor> Cyborg from our research here I do not believe you will need to interface to every neuron to get a robust simulation
<gustavo> a sense of identity
<Nefastor> Gustavo, I don't know how an artificial brain will behave. I simply have no idea at all
<Nefastor> I will just make it and see
<cyborg01> Ocsrazor: that way, you lose a lot of memories... acceptable, but not really nice
<gustavo> ok
<gustavo> thanks
<John_McCluskey> Jean, how many neurons do you think you can stuff into a box the size of a PC?
<Nefastor> My belief, however, is that a system that works ALMOST like a brain should exhibit self-awareness
<Ocsrazor> you lose the resolution of the memories, not the memories themselves, how much remains to be seen
<TimFreeman> How would you test for self-awareness?
<cyborg01> Nefastor: I have thought about the BCI problem a lot... see my web site
<Nefastor> John, for now, very few neurons compared to a human brain. Within 20 years, I should be able to fit many billions
<cyborg01> And how big would that computer be?
<John_McCluskey> Heh heh... let me tell you sometime about these next gen FPGA coming out later this year...
<Nefastor> There is no way to test for self-awareness. A program can be made to display "I think therefore I am", but won't be sentient
<FutureQ> The trick will be watching to seewhicjh destructive or incremental will actually show signs of self awareness. My mony's on incremental.
<Jonesey> tell away John_McCluskey
<Nefastor> Self awareness isn't even a demonstrated feature of humans
* John_McCluskey wishes he could blab about the new products. but can't.
<Nefastor> John, we have to talk regarding FPGA's : I'll soon order thousands of them
* Jonesey straps John_McCluskey in and shines light in his eyes, start talkin
<TimFreeman> Hmm. Self-awareness is overrated if even humans can't do it. It would be nice to know more about myself, though.
<cyborg01> I firmly believe machines can be self-aware
<FutureQ> sur it is, watch a human peer at a mirror, just like a chimp but mecaque no.
<Jonesey> heh it might not be so nice TimFreeman
<Nefastor> Do you realise we aren't even aware of how our body works ? that we have had to study it ?
<FutureQ> don't you meqn conciousnesws?
<Nefastor> Possibly, a machine that looks out for itself is self-aware
<Nefastor> That is easy to test
<TimFreeman> Nefastor: My lack of self-awareness is frequently obvious to me.
* John_McCluskey agrees with Jean that humans are barely sentient.
<John_Ventureville> I've always wished humans had a much more more innate control over their bodies.
