Computational Neuroanatomy

Posted 21 Nov 2007 at 11:58 UTC by steve Share This

A Technology Review article describes the emerging field of Computational Neuroanatomy, where researchers look for automated methods of creating "wiring diagrams" of the neural networks that make up the brain. The human brain is estimated to have 100 billion neurons with 100 trillion synapses connecting them. Creating a schematic of a network that complex is not something that can be done manually, so scientists are looking for ways to automate the process with advanced imaging techniques. In the article Sebastian Seung talks about work being done at the MIT's Department of Brain and Cognitive Sciences and the Seung Lab.

10 petabyte thumb drive = grandpa, posted 22 Nov 2007 at 02:58 UTC by Rog-a-matic » (Master)

Mapping the physical connections would be an important step, but the weightings of the connections contain much of the data, so there would be a lot more work beyond this to record the actual contents of a brain. I think it will eventually be done, and probably much quicker than estimated, just like the human genome project.

Hopefully disk drive technology will increase to accommodate the project! Let's see 100B neurons, 100T synapses, we might be talking 10+ petabyte. Steve, have you heard estimates?

The next logical question is, if all the information from a human brain could be transferred to an artificial network, computer, or even another biological brain, would that be the person? The memories would be there, habits, etc, but are we just a biological machine?

Reverse engineering the brain, posted 22 Nov 2007 at 13:11 UTC by motters » (Master)

I also think that the first structural brain upload might occur sooner than most people believe. This will provide a great deal of useful information about how the brain is organized at a detailed level although there will remain further work to be done before an accurate whole brain simulation can be run.

For microscopy based reverse engineering of the brain's structure it may also be possible to do spectroscopic analysis to reveal chemical concentrations and perhaps infer things like the activity of neurotransmitters.

Storage space, posted 22 Nov 2007 at 18:12 UTC by steve » (Master)

I haven't seen any estimates but it would be very large. They're just talking about storing the raw data but if you want to run a simulation, you also need to store a lot of state information about each neuron - their complicated little things in their own right.

The MIT guys estimate that it would take about three billion person-years of work to manually map out a schematic of a human brain. If I understand correctly, the current automated technique they're using reduces that by something between 100 and 1000 times. So, optimistically, they've already it got it down to three million person-years. Their looking for a mapping technique that could reduce the time needed by another 10^6 fold.

Anyway we've got plenty of time to come up with the storage space. :)

This isn't Norman, this is his evil simulation calling!, posted 2 Dec 2007 at 11:28 UTC by The Swirling Brain » (Master)

I'm assuming that from wiring diagram they are trying to get to a point where they can understand how a brain works and perhaps simulate it. From a software engineering point of view, I don't see how this will be done either. The brain isn't just a memories but also has thought processes. So, you'd have to map all the connections, you'd have to somehow figure out all the neuron's weights. Even then, would you be able to really reproduce the software? So, if you could put a person to sleep, download all his brain's data and put it on a mammoth hard drive, to simulate all of the neurons real time would be a lot of cycles. On top of that, if you got any of the data even slightly wrong, like one neuron was set at 1.92349234 and should be 1.92342432, would that one neuron cause the whole brain not to function correctly? Even if you could transfer all the info to another unused brain like [insert public figure personality here], you still might loose even a bit of information in the transfer that might make a mess of things. Like what if all the neurons were approximated fuzzy values plus or minus 100 thousanths. Would that be close enough to keep all the though processes and brain algorithms functioning and memories still intact? When you have 100B neurons and 100T syn and if only perhaps even one or a handful of those could make the difference of a working simulation and a non-working one, and just a little bit off on all of them that could be a real bummer to debug!!! I'm not sure we really know for sure what all neurons do yet and if they may even have some weird biological functioning or some other besides just neural net weights. Some of them may have "special" functions (I'm not talking about esp by the way, just normal special functions) or some of them may have weird if-then type functions. Some may have chemical biases or chemical sensor type functioning that do something when the hormones change that we don't know about. I have a hard time believing they are just simple numerical weights like our software neural nets have. Like there's chemical processes that push short term memory to long term. Pitutary, hippocampus stuff, dual lobes, area of the brain wired just to do certain functions like sight and smell, motor activies. It's way more than just a bunch of numbers and wires. The spinal cord and nerve cords that run throughout the body also make up functioning of the brain! Like I'm betting our body is more distributed processing than just a couple of lobes upstairs. Even if there's only a minute amount of distributed processing that happens at the bottom of the spinal cord, is that enough to break the program? It's cool that they are trying and I'm sure they'll learn a lot, but I'm afraid this project is doomed to failure unless they can reverse engineer the manual! :) I can pretty much guarantee that when they do it the first time that it won't work. That's to be expected. But can they ever get a brain transfer and simulation to work would be the question? Like are there just too many huge hurdles to jump for this project not to fail for a total recreation of a working person's brain or even one that wasn't working but where they could even read memories? I'm just very, very doubtful.

Once you could do it though, ethics would interesting. Like when you bought someone back as a simulation, is it really them? Are they really considered alive with all rights and responsibilities of a live human? Would it be cruel and inhumane if you start the simulation and then at any whim stop it again? It's just a robot, right? What if a woman wanted to bring dead-hubby back to life in a robot body! ICK!!! :-) What if you could copy and create a bunch of one person even duplicate yourself? What if someone could duplicate you then mold that personality to have an altered personality of you and then make it call your mother and make her cry? What if they revived some evil nut job like a Palpatine galactic ruler figure so that he could rule the galaxy in his evil way again for all eternity! bwa-ha-ha-ha!!! [excellent]

See more of the latest robot news!

Recent blogs

30 Sep 2017 evilrobots (Observer)
10 Jun 2017 wedesoft (Master)
9 Jun 2017 mwaibel (Master)
25 May 2017 AI4U (Observer)
25 Feb 2017 steve (Master)
16 Aug 2016 Flanneltron (Journeyer)
27 Jun 2016 Petar.Kormushev (Master)
2 May 2016 motters (Master)
10 Sep 2015 svo (Master)
14 Nov 2014 Sergey Popov (Apprentice)
Share this page