Tuesday, March 28, 2017

LaceNet : A set of suggested implementations of Neural Lace technology

Well, he's done it.  Elon has finally launched the venture he's calling "Neuralink".


Prerequisites:  If you have not heard of Neural Lace, you have homework:


Basically, there exists a technology such that a tiny mesh can be injected into a human brain.  Neurons are attracted to, and grow on to the mesh.  Neurons that have grown to the mesh can then be individually addressed by a wire coming out of the Brain.  As of last July, it was announced that a research group was able to do the injection without causing any harm to the test subject, and without the test subject rejecting the mesh by forming scar tissue around it, which has been a huge problem with previous iterations of Deep Brain Stimulation setups.

Disclaimer: I am a Computer Scientist who likes to get lost in Wikipedia articles.  I have no formal training in human brain anatomy, nor do I have any inside information on how Neural Lace technology functions beyond the articles mentioned above.

The LaceNet Module:

I'd like to share my vision for what I'm calling the LaceNet Module.  The LaceNet module is a small Bluetooth micro-controller, approximately the size of a dime, with an on board battery that should last a week or longer.  It will be capable of communicating with other LaceNet Modules, storing configuration data, and pairing to a cellphone, where it can be configured with a custom LaceNet App.

The LaceNet Module should be attachable via neodymium magnets to the LaceNet Base, which is a dumb analog multiplexer attached to the back of the skull.  Any number of Neural Lace meshes can be attached to the LaceNet Base.  Probably just a few at first, and then more as new utility is explored.

The theory is that LaceNet Modules can be easily detached, and the user will simply revert back to Human Brain 1.0.  The modules will also be easy to upgrade, and so long as there are standards for how the modules communicate with the base, you can even have competing companies building lighter weight, longer lasting, or more capable modules without needing any sort of intrusive surgical procedures.

The easiest way to program early versions of LaceNet modules will be with a regular smartphone, and the app.  It shouldn't necessarily need to always be connected with the phone, but if you want to schedule something like recurring stimulation montages at certain times of day, you would be able to easily control timing from the app, which would then load the schedule into the module.

The Shared Neuron Architecture:

The obvious first application of Nerual Lace technology is Deep Brain Stimulation.  The important parts of the brain are well mapped, and with this new technology, stimulation can be cleaner and more controlled than it has ever been in the past.  What I'd like to talk about instead is something I haven't heard anyone else talking about, and what I'm calling the Shared Neuron Architecture.

Imagine having a mesh implanted in your prefrontal cortex in an area commonly used for problem solving.  Now imagine that a colleague has done the same.  Now imagine that any neural stimulation recorded from the neurons on your mesh get sent to the mesh of your colleague, and vice versa.  It might take your two brains some time to adapt to the confusion, maybe even years, but once it does, problems that you are thinking about could inspire solutions from your colleague, and problems your colleague is thinking about could inspire solutions from you.  In time, this would allow the two of you to directly draw off each others experiences.

Access control would be critical for a system like this, as well as training.  These things could be managed in the LaceNet App, but ideally a recording of per user neural affinity would be stored in the module itself.  I'd imagine when you're at work you'd be more interested in sharing neurons with coworkers, and maybe at different times of the day you would prefer to share more with friends or family.

Human brain latency is relatively high.  High enough that, should you feel like it, you could share neurons across the internet, though early use cases would probably be sharing neurons module to module.  Sharing the same neurons with multiple people should also be completely possible, so long as you adjust for relative amounts of influence.

Now imagine a 5-8 graph, that is, a large, connected graph, where everyone is connected to somewhere between 5 to 8 people.  I don't think it would be feasible to share neurons with more than that many people, as it would get extremely noisy.  However, if your connections were arranged in a graph to filter out less interesting problems and relay more interesting problems, you could, in a sense, be connected to thousands, or millions, or billions of people simultaneously.  At that point the LaceNet becomes a globally distributed problem solving machine able to take advantage of the entirety of human consciousness to solve any problems it comes up with.

Training Audio Channels:

Maybe I should save this for the next blog post, but I have a number of ideas around how to solve a much more difficult problem, direct communication.  The Shared Neural Architecture is great for expressing bursts of thought in the form of analog pulses, but it's useless for any sort of detailed expression.

I feel like the most straightforward way to deal with this problem is to re-purpose the parts of the brain used for transmitting and receiving audio.  This is going to require a lot of training at first, but will enable us to communicate at high speeds with the people and machines around us, and it can all go over the same LaceNet architecture described above.

No comments:

Post a Comment