The Steampunk Forum at Brass Goggles
November 22, 2019, 07:11:56 pm *
Welcome, Guest. Please login or register.
Did you miss your activation email?

Login with username, password and session length
News: Support BrassGoggles! Donate once or $3/mo.
 See details here.
   Home   Blog Help Rules Login Register  
Pages: [1]   Go Down
Author Topic: Frankenstein and R.U.R. Have Come True  (Read 441 times)
Zeppelin Captain

« on: May 25, 2019, 05:22:20 pm »

Once laboratory grown tissue became feasible it was inevitable. Robotic devices are being built from living tissue. The constructs will become more complex over time. Could a conscious artificial organism be possible?
J. Wilhelm
╬ Admiral und Luftschiffengel ╬
Board Moderator
United States United States

Sentisne fortunatum punkus? Veni. Diem meum comple

« Reply #1 on: May 27, 2019, 09:39:23 am »

The answer is yes, but we're dragging our feet on it.

Back in the 1990s, when Neural Networks were the new kid on the block, Sci fi authors like Masamune Shirow (Ghost in the Shell Manga )  were trying to foresee all the different ways one could hybridize digital and living neural circuits. MIT and Caltech were developing small silicon squares with arrays of living neurons, like 16 x 16 matrices, where neurons were pigeonholed in a silicon grid and disturbed with an electric current to encourage them to make connections between each other.

The technique MIT and Caltech used on living neurons back then was very similar to the one you showed above. Typically the number of connections one human neuron will make with surrounding neurons in the brain is in the 1000s. The MIT chip neurons could potentially make 15 connections each for a 4x4 array. The neurons were fed with nutrient and oxygen rich solution. The MIT biological brain chips were being made in 1994-7. That's over 20 years ago.

I didn't follow that progress much longer after that, as I was busy with school. Also, it was a subject of statistical mathematics that is way too esoteric for an undergraduate engineer. That kind of math was something I learned by graduate school, but by then I was too busy with other subjects so I briefly read some books on the subject, and just left it at that. I did try to incorporate it into my research and I briefly bother my research supervisor with the idea of developing a neural network-like cellular automata based fluid mechanics simulator. He flatly told me that he would forbid me from developing that idea, and that instead I should work on a different method he created. Oh well.

The problem is that "machine intelligence" (the way IEEE defines neural computing) is vastly different to "artificial intelligence" (the way IEEE defines logical algorithm based computing, usually of the digital computation variety). In terms of academic understanding, Artificial Intelligence is far more advanced than Machine Intelligence. Even the way we design neural networks is to *emulate* them as part of an Artificial Intelligence computer program, in other words, a virtual neural network - typically implemented in speech, facial, image recognition devices. Very rarely do you ever see an actual neural network hardwired into a chip. In fact I hazard a guess that commercially that does not exist at all.

For some reason, direct Machine Intelligence neural network hardware has not been used too much. We prefer digital computation (smaller, more practical, computers are as fast as we want them to be). We need to accelerate this kind of cybernetic machine intelligence interface with living tissue so we can really start talking to computers and back, and presumably build a hybrid Digital - Neural brain as predicted by Masamune Shirow.

One example of research on that direction of Machine Intelligence, is a chip made up of transistors which reproduces the behavior of a single synapse. Presumably this could be an interface between a living neuron and a silicon based neuron:

It literally takes hundreds of transistors just to try to reproduce anything that looks like a synaptic firing. I guess this chip would be considered "analog computing" although a synaptic signal is really a hybrid between a digital signal and an analog signal (it's basically analog with a sharp "cutoff" that does behave in a digital way, like instead of 0s and 1s, it's more like 0 and 0.8775, 0.5, or 0.338, etc. depending on all the input signals to the neuron)
« Last Edit: May 27, 2019, 10:22:45 am by J. Wilhelm » Logged

Pages: [1]   Go Up
Jump to:  

Powered by MySQL Powered by PHP Powered by SMF 1.1.20 | SMF © 2013, Simple Machines Valid XHTML 1.0! Valid CSS!
Page created in 0.192 seconds with 16 queries.