7 ms·

I'd love for a neuroscientist to weigh in on this - what is the possibility that something like this could be used to generate an artificial neural network whic

I'd love for a neuroscientist to weigh in on this - what is the possibility that something like this could be used to generate an artificial neural network which imitates, even crudely, the functions of the human brain?

networked 13y ago
>what is the possibility that something like this could be used to generate an artificial neural network which imitates, even crudely, the functions of the human brain? If possible, this would open a huge ethical can of worms: how can we tell such a simulation is not conscious? Would deleting its runtime data after however many simulation cycles be tantamount to murder? Edit: I am not saying that the above concerns should stop us from working towards developing human brain simulations since the potential benefits of those are just too great. Rather, it is something we have to have in mind as they get more complex and closer to the real thing.
- alexjeffrey 13y ago
  I'm not necessarily saying we _should_ create a simulation of the human brain - just asking whether it would be possible. There are other much more concrete concerns with creating strong AI too - avoiding accidentally creating a paperclip maximiser comes to mind: http://wiki.lesswrong.com/wiki/Paperclip_maximizer http://wiki.lesswrong.com/wiki/Paperclip_maximizer
- comex 13y ago
  Heh... as serious as they may be, the idea of one of those hypothetical ethical issues science fiction authors have written a million almost-but-not-quite-plausible takes on finally becoming real makes me giddy. I suppose it has already happened for certain issues, but nothing so dramatic as considering a simulation conscious! Then again, it will likely be many years still before that actually comes up.
robbiep 13y ago
Not a neuroscientist but final year med with a strong interest in neuro- This will certainly be useful for attempts to recreate the functionality of a brain in silico - big research efforts such as the blue brain project [1] have probably been using even early forms of this data. Having said that, in some ways it won't provide the useful information that will enable the functionality to be restored to the connections. heres why. Current efforts involving simulated brains rely both on older, less precise versions of what this nature article talks about - let's call that the roadmap, as it is a map of where all the neurons sit and where they connect to. This project drastically increases the resolution of this roadmap. However the additional information needed is what traffic is carried by each road. The brain as we understand it today consists not just of the connections between the neurons but also that each neuron, and each different network of the brain, secretes it's own neurotransmitter. Because some of these neurotransmitters are depolarising and some are inhibitory, and they connect widely across the brain and are much more closely related to the actual information processing, we need to have this information before we will be able to better approximate a brain. In all likelihood this is information gathering is probably underway right now. From what I understand the 'BigBrain' project aims to do for the human brain what the Allen Brain atlas did for mouse brains, that is not only provide a map of the connections but also stain for gene expression and (from memory) neurotransmitter presence. With this added information we will be closer to our goal of getting skynet sentient. [1] http://bluebrain.epfl.ch/ http://bluebrain.epfl.ch/
ton2lavega 13y ago
Depends. First, a biological network is not just connections of neurons, it's weighted connections of neurons. And even saying that is a huge approximation. It's not just about mapping the topology of the brain, just having a neural network of 1e10 vertices and 1e12 links will not give you an artificial intelligence. It's just not that simple. The connectome paradigm is not sufficient. It is hip because of computer scientists using artificial neuron networks to achieve awesome stuff. But that doesn't mean that the complexity of the human brain only resides in the number of neurons and interactions. It's not guaranteed that we are getting closer of the goal just by adding billions of neurons and connections. I find this wannabe holistic approach a bit confusing as it is horizontal whereas I am convinced that what we need is a vertical approach : up from graph topology down to an accurate biophysical understanding of synapses and information transmission, all across multiple scales, from the angström up to microns. That's the key : multiscale modeling, not single-huge-scale modeling.
- webXL 13y ago
  I totally agree. The brain is drastically more complex than a CPU, which could be reverse engineered in such a fashion. That's because we have a solid understanding of what happens when electrons flow through a series of specialized circuits, and we can control (and model) the input and output of each one. We cannot yet control or predict that flow with a human neuron, I don't think. At least this map doesn't get us closer in that regard, as much as I want it to.
stinos 13y ago
Not a neuroscientist, but used to work in the field, and my first idea was "well by cutting it up, they also cut all connections". And those connections are far more important for understanding how the brain works than seeing where an individual neuron is physically located. Unless off course they can somehow get the connections out of it.
etrautmann 13y ago
Systems Neuroscientist here. I've only skimmed this paper so far, but it seems as though this dataset is limited to studying cytoarchitecture (what types of neurons are in what regions of the brain) as well as some tracing of major fiber pathways. This does not represent the successful conclusion to connectomics efforts to map every neuron and every connection in the brain (see Sebastian Seung's work and the eyewire project [1]). It's hard to tell exactly how useful this dataset and model are. There are other exciting developments that map not just neuroanatomical structure but also the distribution of proteins and gene expression [2],[3] [1] http://eyewire.org/ http://eyewire.org/ [2] http://med.stanford.edu/ism/2013/april/clarity.html http://med.stanford.edu/ism/2013/april/clarity.html [3] http://smithlab.stanford.edu/Smithlab/Array_Tomography.html http://smithlab.stanford.edu/Smithlab/Array_Tomography.html
adrianhoward 13y ago
Another "not a neuroscientist" - but I do have a slightly-more-than-layperson knowledge and interest in the area, and a degree in AI if that matters. "I'd love for a neuroscientist to weigh in on this - what is the possibility that something like this could be used to generate an artificial neural network which imitates, even crudely, the functions of the human brain?" Pretty much zero. Because it really just focussed on the organisation of the neurons, and ignores the complexity of the neurons themselves and the chemical/electrical/etc. communication between 'em. Neurons themselves are pretty complex beasts, and we're still learning things about them and the level of "processing" that is happening within neurons, as opposed to between 'em. For example as recently as ten years ago everybody knew that most of the computation was implicit in the neural connectivity of the synapses. We now know that there is significant computation within individual neurons - in the dendrites of all things (previously thought to be pretty much passive carriers of output from other cells - just wires basically). (See http://www.annualreviews.org/doi/abs/10.1146/annurev.neuro.28.061604.135703 http://www.annualreviews.org/doi/abs/10.1146/annurev.neuro.2... ). Don't get me wrong - it's useful work. But nowhere close to the level of detail needed to run people-sims.

networked 13y ago

>what is the possibility that something like this could be used to generate an artificial neural network which imitates, even crudely, the functions of the human brain? If possible, this would open a huge ethical can of worms: how can we tell such a simulation is not conscious? Would deleting its runtime data after however many simulation cycles be tantamount to murder? Edit: I am not saying that the above concerns should stop us from working towards developing human brain simulations since the potential benefits of those are just too great. Rather, it is something we have to have in mind as they get more complex and closer to the real thing.

alexjeffrey 13y ago
I'm not necessarily saying we _should_ create a simulation of the human brain - just asking whether it would be possible. There are other much more concrete concerns with creating strong AI too - avoiding accidentally creating a paperclip maximiser comes to mind: http://wiki.lesswrong.com/wiki/Paperclip_maximizer http://wiki.lesswrong.com/wiki/Paperclip_maximizer
comex 13y ago
Heh... as serious as they may be, the idea of one of those hypothetical ethical issues science fiction authors have written a million almost-but-not-quite-plausible takes on finally becoming real makes me giddy. I suppose it has already happened for certain issues, but nothing so dramatic as considering a simulation conscious! Then again, it will likely be many years still before that actually comes up.

robbiep 13y ago

Not a neuroscientist but final year med with a strong interest in neuro- This will certainly be useful for attempts to recreate the functionality of a brain in silico - big research efforts such as the blue brain project [1] have probably been using even early forms of this data. Having said that, in some ways it won't provide the useful information that will enable the functionality to be restored to the connections. heres why. Current efforts involving simulated brains rely both on older, less precise versions of what this nature article talks about - let's call that the roadmap, as it is a map of where all the neurons sit and where they connect to. This project drastically increases the resolution of this roadmap. However the additional information needed is what traffic is carried by each road. The brain as we understand it today consists not just of the connections between the neurons but also that each neuron, and each different network of the brain, secretes it's own neurotransmitter. Because some of these neurotransmitters are depolarising and some are inhibitory, and they connect widely across the brain and are much more closely related to the actual information processing, we need to have this information before we will be able to better approximate a brain. In all likelihood this is information gathering is probably underway right now. From what I understand the 'BigBrain' project aims to do for the human brain what the Allen Brain atlas did for mouse brains, that is not only provide a map of the connections but also stain for gene expression and (from memory) neurotransmitter presence. With this added information we will be closer to our goal of getting skynet sentient. [1] http://bluebrain.epfl.ch/ http://bluebrain.epfl.ch/

ton2lavega 13y ago

Depends. First, a biological network is not just connections of neurons, it's weighted connections of neurons. And even saying that is a huge approximation. It's not just about mapping the topology of the brain, just having a neural network of 1e10 vertices and 1e12 links will not give you an artificial intelligence. It's just not that simple. The connectome paradigm is not sufficient. It is hip because of computer scientists using artificial neuron networks to achieve awesome stuff. But that doesn't mean that the complexity of the human brain only resides in the number of neurons and interactions. It's not guaranteed that we are getting closer of the goal just by adding billions of neurons and connections. I find this wannabe holistic approach a bit confusing as it is horizontal whereas I am convinced that what we need is a vertical approach : up from graph topology down to an accurate biophysical understanding of synapses and information transmission, all across multiple scales, from the angström up to microns. That's the key : multiscale modeling, not single-huge-scale modeling.

webXL 13y ago
I totally agree. The brain is drastically more complex than a CPU, which could be reverse engineered in such a fashion. That's because we have a solid understanding of what happens when electrons flow through a series of specialized circuits, and we can control (and model) the input and output of each one. We cannot yet control or predict that flow with a human neuron, I don't think. At least this map doesn't get us closer in that regard, as much as I want it to.

stinos 13y ago

Not a neuroscientist, but used to work in the field, and my first idea was "well by cutting it up, they also cut all connections". And those connections are far more important for understanding how the brain works than seeing where an individual neuron is physically located. Unless off course they can somehow get the connections out of it.

etrautmann 13y ago

Systems Neuroscientist here. I've only skimmed this paper so far, but it seems as though this dataset is limited to studying cytoarchitecture (what types of neurons are in what regions of the brain) as well as some tracing of major fiber pathways. This does not represent the successful conclusion to connectomics efforts to map every neuron and every connection in the brain (see Sebastian Seung's work and the eyewire project [1]). It's hard to tell exactly how useful this dataset and model are. There are other exciting developments that map not just neuroanatomical structure but also the distribution of proteins and gene expression [2],[3] [1] http://eyewire.org/ http://eyewire.org/ [2] http://med.stanford.edu/ism/2013/april/clarity.html http://med.stanford.edu/ism/2013/april/clarity.html [3] http://smithlab.stanford.edu/Smithlab/Array_Tomography.html http://smithlab.stanford.edu/Smithlab/Array_Tomography.html

adrianhoward 13y ago

Another "not a neuroscientist" - but I do have a slightly-more-than-layperson knowledge and interest in the area, and a degree in AI if that matters. "I'd love for a neuroscientist to weigh in on this - what is the possibility that something like this could be used to generate an artificial neural network which imitates, even crudely, the functions of the human brain?" Pretty much zero. Because it really just focussed on the organisation of the neurons, and ignores the complexity of the neurons themselves and the chemical/electrical/etc. communication between 'em. Neurons themselves are pretty complex beasts, and we're still learning things about them and the level of "processing" that is happening within neurons, as opposed to between 'em. For example as recently as ten years ago everybody knew that most of the computation was implicit in the neural connectivity of the synapses. We now know that there is significant computation within individual neurons - in the dendrites of all things (previously thought to be pretty much passive carriers of output from other cells - just wires basically). (See http://www.annualreviews.org/doi/abs/10.1146/annurev.neuro.28.061604.135703 http://www.annualreviews.org/doi/abs/10.1146/annurev.neuro.2... ). Don't get me wrong - it's useful work. But nowhere close to the level of detail needed to run people-sims.