notes : modular conscious entities

Once conscious systems are virtualized, we will have the ability to do some unusual things, for example : separate two halves of the brain and putting them on different tasks, or turning off inhibition at will. Such events happen already, and are caused by damage to or sugical manipulation of the human brain. For a virtual conscious system ( or a heavily modified biological neural system ), we should be able to recreate these events reliably and reversibly.

Furthermore, it may be possible to swap bits of different individuals. I am not sure if I mean individuals in the human sense or individuals as conscious software entities. A mature human consciousness would not take well to having its frontal lobe swapped out for that of another person, but it may be able to handle, with practice, swapping out sensory-motor modules. If you train modules from the start (from birth, or from first awakening for a software entity) to be able to be swapped, we could create a set of self assembling processing modules that could instantly adapt to new problems. However, If you try to use bits of "people" in this framework they might go insane rather quickly.

What might be more feasible would be extremely high bandwidth communication between normal humans, using implantable electrode arrays. It would be like using a language more expressive than any spoken language. Depending on exactly how you link together separate individuals, you may get radically different results. For instance, wiring one man's motor cortex to another man's frontal lobe would seem like a very bad thing, since having your super-ego controlled by your friend's left toe is never a good idea. A person would probably be able to use a direct sensory-motor link though, since this would more closely resemble normal communication, like speech.

It is also possible that strong, human level, artificial intelligence will first emerge from such a modular framework. I am not extremely familiar with the state of the art in software AI, but I believe that as the field progresses the training of a software solution for a task will become a significant part of the development process. I can imagine the day that someone takes a group of pre-trained software AIs and links them together under a common executive module, and accidentally creates something we might call conscious.

Tue, Nov 18, 2008  Permanent link

Square Hypercolumn Lattice

This model more closely reproduces some of the results in this paper (Bressloff et al, 2000). However, some of the vision people I have talked with have said that there is no compelling evidence for the type of lateral connections assumed by the model in this paper.

and just for fun,

Fri, Aug 29, 2008  Permanent link

Moths

This is another toy (it is unclear if such things belong here). However, I thought the emergent complexity resulting from multiple interacting automata was worth showing.



you will need the latest version of java to view this



( click and drag things )

if you
...click in an empty area a moth appears
......on a moth it turns into a light
......on a light it goes away
...drag a moth it follows
......a light the string appears
...tie the string to a moth the light will follow

( for some reason it stopped working ? )

Sun, Aug 24, 2008  Permanent link

... ( toy ) ...

( you probably need the latest version of Java to run this )

Mon, Aug 18, 2008  Permanent link

This is the same algorithm as in the last post, but I played around with the color scheme and rendered a longer video.



note : I would like to embed applets in these posts, however I don't have another web hosting service. Is it possible to host applets and other content besides text and images directly on spacecollective ?

Sat, Aug 16, 2008  Permanent link

More Hallucinations

This video expands upon the model from my previous post. It now incorporates connections between orientation selective cells in the visual cortex. If I made this right ( and I'm not sure I did ), cobweb like lattices of lines should be stabilized. I couldn't generate a proper map of the orientation selectivity in visual cortex, so I just seeded random orientation selectivity. This is not physically realistic and probably changes the results. The equations are the same as in my previous post, but with a new term Eorient added into du/dt which accounts for input from other orientation selective cells.



The degree of excitation from orientation selective connections at some point r is proportional to the sum over the activation of all cells in the neighborhood of r, where input is weighted as a Gaussian function of distance, and is strongest when both cells are tuned to the same orientation, and cell r lies along the direction of the preferred orientation of cell r' . U and Theta are fields ( represented at floating point matrices in the code ). The bold subscripts represent particular points in the filed. U represents the synaptic activation ( correlated to neuron firing rate ), and Theta represents the preferred stimulus orientation for a cell.

Tue, Aug 12, 2008  Permanent link

Hallucination Simulation

Under certain conditions the cortex of the human brain can act to propagate wavelike activity (these aren't the usual 'brain waves'). Such states can be induced by epilepsy, administration of hallucinogenic drugs, migraines, and even by certain stimuli such as strobe lights of particular frequency. This video represents a cross-section of spontaneous pattern formation in a simple model of primary visual cortex. Over the course of the video, parameters governing a simple model of hallucinations are varied. The logarithmic spiral property of this image comes from projecting back through the retinocortical map, showing approximately how wave activity in the visual cortex might appear to the observer. The color and seven fold symmetry are artificially imposed.



The equations look pretty typeset but require a lot of explanation. I should return and explain more once I understand it myself.

Wed, Jul 30, 2008  Permanent link

Ermentrout-Cowen model of hallucinations

Thu, Jun 5, 2008  Permanent link

Diffusion Limited Aggregation

These images and algorithms might be of interest to the collective



( posted images are from an article by Paul Bourke )

Thu, May 1, 2008  Permanent link

Natural

From the family (phi/z+i*z)*a+c, where phi is the golden ratio. I'm not certain what other effects played in here, these results were unexpected.

Tue, Apr 1, 2008  Permanent link