Topic: Neuro-Electronic Interfacing
Immortality Institute Live Discussion: Feb 16, 2003
Special Moderated Interview with Peter Passaro(Ocsrazor)
[>] Peter Passaro
[>] Neuro-Electronic Interface Resources
[>] Peter Passaro Background Info
-----
OmniDo: Your study in what might be dubbed "Nero-cybernetics" is an admirable persuit.
Ocsrazor While i was assembling that background info, I came across a quote by Hawking:
'We must develop as quickly as possible technologies that make
possible a direct connection between brain and computer, so that artificial
brains contribute to human intelligence rather than opposing it.'-- Stephen Hawking
Ocsrazor: I had never heard that expressed by anyone excenking about this myself, it's great to see others thinking about this in the same way.
OmniDo: Such works would benefit the people as a whole, as well as individuals like Hawking himself.
MichaelA: "How many neurons do you typically work with at once, and why?"
Ocsrazor: about 10,000. That is the # we can comfortable get on top of our chips and still have them lay flat.. there is about 25 K total on the dish, but only 10K on the electrodes. Just in case anyone has't read the background, my system is a 60 electrode multi-electrode array with a disassociated culture of embryonic rat or mouse neurons
OmniDo: May I inquire as to how the disassociated cells manage to acquire their "link" with the electrodes?
Ocsrazor: They simply grow and form a network on top of them. Any neuron near an electrode can be stimulated or be recorded from. We estimate that we are getting about 1% of the dish
OmniDo: The cells have a natural tendency to connect to the electrodes? Or is there some special catalyst which causes their direct connection?
Ocsrazor: the electrodes are flush with the glass on the bottom of the dish so the neurons just grow over them. There are ways to pattern neurons to grow in certain places but we haven't had a need for that... Yet
tmoenk: What are some of the near term goals for this field, and what are the challenges that are currently being encountered?
Ocsrazor: Near term for in vitro systems are to understand the signalling in a small network... to get the encoding/decoding mechanisms worked out most of the major challenges are in understanding the data
OmniDo: Im visualizing a system of interconnected action potentials, much the same way as mapping a motherboard's connections...is that a close approximation?
Ocsrazor: a lot more complicated because you have both a time and a space component and every cell in the dish could possibly be connected to any other cell. The culture is only about a mm square so very high connectivity.
OmniDo: Ah. A computer chips lay-lines are static and unchanging, with constant data-type. These I presume are not only changing their lines, but also changing data over time...correct?
Ocsrazor: they can change their connections, but that is slow... what changes fast is the pattern of activity...
Alex: "I'm curious about the work done to decode the messages used by th neural cells in communication, their "language" so to speak"
Ocsrazor: We are just beginning to understand these... the best metaphor is probably mathematical transformations... and perform all sorts of mathematical functions on the patterns
OmniDo: According to what has been discovered with human neural development, many neurons connect and send various signals based upon reception and response. Those links that receive responses are inclined to continue while those that do not, degenerate. Do these patterns follow similar rules within your experiment(s) ?
Ocsrazor: Absolutely, the saying is "neurons that fire together, wire together". It is also true that neurons that are slightly off synch will inhibit each other and not form connections.
OmniDo: If we presume that these neurons are not only interconnected with the transistors but also with themselves, has there been any testing involving patterned stimuli from the transistors; as so to facilitate similiar pattern development amongst the neighboring neurons?
Ocsrazor: We are beginning to expose the dish to more complicated patterns of stimuli now. Early on it was just one or two electrodes.. now we can stimulate all 60 in varying patterns
Ocsrazor: and just to be clear, these are just simple electrodes not transistors...
OmniDo: Ah, One way communication?
Ocsrazor: No, we can stimulate and record from the electrodes but there is no solid connection
Alex Asks: "So when cells communication, what they're doing is transmit and carry out transformations on information as well as being changed by the flow of info through the network?"
Ocsrazor: They most likely perform transformations as a group, and the individual neurons will change cnxns in response to different stimuli
Alex Asks: Also "what's exactly the incentive for the neural cells to start communicating? I mean what gets them going, random permutations in addition to possible electrical stimulations?
Ocsrazor: If we don't stimulate at all they generate a fairly periodic bursting. There seems to be some inherent pattern generator.
OmniDo: Could you surmise these peroid burstings as random or natural attempts at saying "hello? Is anyone there?" to their fellow neruons?
Ocsrazor: They are seen in a lot of different systems and are probably a developmental mechanism for forming initial connections.
Omnido: Essentially a "Reach out and touch someone" system?
Ocsrazor: Yep, and then slight variations in the system cause more developed patterns to start forming
Eliezer Asks: What's the most interesting discovery you've made so far?
Ocsrazor: So far it hs been alot of just developing the techology to do this... But recently... we have started to understand some of the encoding going on.. so we are starting to be able to feed in stimulus and know what the response is going to be
Alex Asks: "so when you design a device to interface with the brain, you want to present to the neural network so it sees the device as another network, right?"
Ocsrazor: Yep. You want to mimic the natural pattern of activity. You want to speak to i in a language it understands
OmniDo: What "language" per say has been decoded? I'm sure if the patterns are based upon input and interaction, what sort of "Grammar" will demonstrate useful with the interface?
Ocsrazor: Right now we understand that a certain stimulus will produce a certain "response curve" in our dish... enough so that we might be able to control a robot with it... the reponse is a particular pattern of activity that gets stronger if given the right stimulation
Alex Asks: "what kind of technologies do you envision that will come about to use brain interfacing, such as artifical vision, even artifical language centers or auditory centers, or maybe a direct connection to the net/virtual reality"
Ocsrazor: artificial vision is getting very close... if you look in my resources page there is a link to the Utah lab who are the leaders in artificial vision
OmniDo: Robots, as well as any other electro-mechanical system require very specific input. Based upon neuro-chemical action potentials, how would you decipher exact digital coding with a robotic interface? Do you plan to use an analog system for input/output instead?
Ocsrazor vision has the advantage of alot of preliminary neuroscience has already been done.. for the robots we are using a very simple interface, right now it only has four commands... forward, back, right, left in specific increments. The strongest patterns from the dish are selected to represent these signals and the patterns are grouped into certain areas of activity. The signalling is completely electric right now from what we record through the chip
OmniDo: Is there a limited lifespan to the nerual network? If an established, functional, and integrated network works to perform these organized input/output schemes, how long wil it live/last before parts of the network start dying off?
Ocsrazor: right now our record culture is two years olde
OmniDo: Thats impressive.
Ocsrazor: That is about the lifespan of the animal.
FutureQ: "Are you involved in the work on the area of the brain that is involved in acting out in the brain an action slightly before it is sent to the motor system?"
Ocsrazor: No, that is the Nicolelis lab at Duke. Our system is completely in the dish
Alex Asks: "Do you envision that we'll reach the point where synthetic senses function better than the organic senses?"
Ocsrazor: Absolutely, eventually we should be able to get higher bandwidth and different physical senses expanding our visual range should be pretty easy once you can build a visual implant. Same goes for other senses too
OmniDo: Do you have an estimate for a time frame that this research is expected to produce beneficial and applicable results?
Ocsrazor vision and motor stuff, very soon... even though the tech will be crude at first, it will still make for viable med devices. Our system seeks to get really fine control of the networks... and is more like 10-15 years before maturity... our stuff is actually dependent on increases in computational power and better software for data processing
OmniDo: Assuming that this technology can also be applied to human neurons, do you think that its feasible to expect a longer lived (and possibly faster adapting) network that uses human neurons instead of animal ones?
Ocsrazor: There isn't a significant difference between human and rodent neurons in terms of complexity... longevity I don't know if anyone has speculated. The difference is probably in the wiring, not the individual cells
OmniDo: Current research indicates that some neurons within the human brain can live to be as old as the human itself. This would be promising if your research could be adapted for use with human neural archetecture.
Ocsrazor: Most of them you have from birth til death
Alex Asks: "what about constructing synthetic neurons?"
Ocsrazor: thats the plan ;^) One half of our lab is the "dry" side all electrical engineers. They do neuromorphic engineering... mimicking neurons in silicon and building better recording/stimulating devices for us biologists
OmniDo: There are some researches attempting to conect transistors with neurons; adapting their design to known pre-existing neural archetecture and patterns. Do you forsee a simiar work being implimented within your lab?
Ocsrazor: no, the other way around actually. The people working on transistors have moved to electrodes because neurons have a tendency to crawl... they can't be fixed to a transistor
MichaelA Asks: "Aren't neurons extremely complex? how far away are we from creating indistinguishable-from-original synthetic neurons?"
Ocsrazor: we have a few in our lab ;^) they perform indistinguishably from real neurons when hooked into a biologic netwrok same electrophysiologic properties.. just ALOT larger
Alex Asks: "As technology advances, we'd probably start replacing damaged neurons with artifical neurons or even augmenting the organic neurons...?"
Ocsrazor: We can already replace neurons with stem cell grown ones. I think it will be a while before you see internal replacement by silicon. Interfaces will come much sooner
OmniDo: Unfortunately that is another area of research that is pending; many religious and ethical resistances are attempting to put a halt on that development.
Alex Asks: "i've been also wondering how accurate the current neural models are in approximating the biological neuron?"
Ocsrazor: Very close Alex...the models can't exactly work out some of the more complex integration going on in the dendrites.. but I think this will be solved soon... simple neurons are well completely mathematically modeled
OmniDo: Is the speed of communication between neurons one of the inhibiting factors of interpretation during research? If so, do you forsee that the advances in computer tech and instrumentation may solve these issues?
Ocsrazor: It is not exactly the speed, we can see a majority of the signal... it is the interpretation pf that much signal by a human... so we do need better computational power to record from larger networks and... better software to interpret it, artificial neural networks to read our biological ones
Ocsrazor: ironic huh?
OmniDo: Indeed.
Alex Asks: "How long before you'll start scaling up to very large systems, say on the order of millions+ neurons?"
Ocsrazor: it is a matter of understanding, we could do that now, but we wouldn't understand it any better so...
OmniDo: Better to start small, and work your way up.
Ocsrazor: we are going to stick with a small system and describe it completely
OmniDo: Makes sense. Then you'll have a far better base for understanding larger networks.
Ocsrazor: the in vivo people, vision, motor cortex do look at millions... but they don't have to understand it to get some level of control... because of the statistical nature of the system you can read from a small number in a population...
Alex Asks: "What about other details that are found in biological neural networks, such as neurotransmitters, hormones, all kinds of chemicals traveling around the crevices of the brain, how important are they to creating accurate representatives of the neurons?"
Ocsrazor: This is important and difficult to study..Neurotransmitters are most likely causing large area effects and regulating whole networks
OmniDo: Neurotransmitters are also input/output specific, as well as responsive to bio-feedback. This type of "dynamic chemical programming" could be responsible for many if not all the complexities revolving around neural interactions. Perhaps as tech advances, we will find ways to monitor these interactions as well? It seems more than probable.
Ocsrazor: Yes, we are working on including neurotransmitter(NT) sensors into the system and doing some stimulation with NTs squrited into the dish... NTs are very important in forming connections between neurons as well so we would like to look at the effects on developing the networks
OmniDo: Heh, forgive me, but that sounds a bit primitive from the cells perspective. Usually neurotransmitters are produced in very specific quantities. Squirting them into the dish might have the effect of dumping a human brain into a swimming pool of alcohol. (Just as an example) But I understand your approach.
Ocsrazor: the MEMs people here at GT are developing a fine control system to place a very small amount of NT in a very specific place as well as a 3D culutre system with a circulatory system with an extremely fine level of control
OmniDo: Ah, very promising indeed. What kind of funding is your research group receiving for all of this? OmniDo If you dont mind me inquiring...
Ocsrazor: Ocsrazor NSF mainly... it is probably going to be one of the finest collaborative labs for neuroscience in the wotld
OmniDo: Do you have an estimate of an ammount? Or is that deemed classified information?
Ocsrazor: there are 7 main proffs involved and I don't know all their funding but I would estimate in the 15-20million range... our lab is not hurting ;^)
OmniDo: How many members of your research group are involved for carreer reasons, as opposed to a dire-straight desire for pure research and knowledge.
Ocsrazor: Almost all of us are students Omni, I'm going Academia most likely
OmniDo: Thats reassuring.
Ocsrazor: but a lot of the engineers will go on to industry... we have a very ecclectic mix of people
OmniDo: Peter, you have brought to our attention some wonderful research and developments with which you are partaking. Do you have anything in particular you'd like to share as to what you are planning for upcoming developments?
Ocsrazor: the near future holds a lot of math and programming for me... I have applied to go to the Complex Systems Summer School at Santa FE inst... and I'll be getting deep into data mining and neural network programming
BJKlein: The official chat has now come to a close.. we'll open it up to all.. everyone is free to chat for as long as you'd like
Alex: it was nice talking to you, i learned quite a bit tonight... i've thought about applying to santa fe inst for an internship one summer
Ocsrazor: kuhl, glad you enjoyed it
OmniDo: programming seems to be a very important factor now. It is no longer an issue of hardware so much as it is applicable and useable software.
Ocsrazor: exactly Omni, we have more data than we can handle... too much for a human brain to crunch
OmniDo: Time to start using those 3GHz processors 'eh?
Ocsrazor: we have a couple of mid range SGIs in the lab... that we use for recording
OmniDo: That field of research is taking some new advancements too, as developers begin to use higher frequency lasers for their optical media. Very soon, we will have DVD's that are triple and quadruple layer, as well as increasing the storage size to its molecular limits.
Ocsrazor: very cool, we just expanded our raid system to over a Tbyte to deal with with our dataflow
OmniDo: However for most mainstream media, the current DVD systems should prove adequate. TB hard drives arent far away either.
Ocsrazor 60 channels at 10 Khz goes pretty quick
OmniDo With a slave of just 5 200GB,s you can easily set up a sever with 1TB of data storage.
Ocsrazor they need to be quick though
OmniDo: The average market price for 200GB Hard drives is about $250-350 US dollars.
