[HS]

My first idea was to create a state of focused multimodal attention. The focus would be variable (by concentration and location) and backed by passive cognition/processing to generate property/function/value approximations. More abstract cognition could contemplate the results of the passive layer, employ reasoning for significant elements, and modify attention to solve unknowns.

[infurl]

That sounds a lot like a large language model so I think you must be on the right track. There are still a lot of problems left to solve in that area, not the least of which is efficiency. There must be a better way to do it than vast numbers of tensor calculations.

[frankinstien]

That sounds a lot like a large language model so I think you must be on the right track. There are still a lot of problems left to solve in that area, not the least of which is efficiency. There must be a better way to do it than vast numbers of tensor calculations.

Perhaps, but despite analog being very efficient with audio media, rather than improving bandwidth and encoding techniques such as phase modulation to accommodate analog processes digital approaches prove to be better at storing, distributing, manipulating, and compressing. Everything that has gone digital improves as to what is possible to manage information, and consciousness will likely benefit from the same process. True-to-form optimizations are improving; locally run models do pretty well and have a smaller footprint. Making models smaller in the inference mode allows for utilizing a GPU's memory management capabilities where multiple models, given sufficient memory, can be cached, and switching between the two or more avoids the loading and reloading delays.  But, loading from CPU ram is what Pytorch can do and provides for queuing up models in CPU ram, again improving loading performance.

[WriterOfMinds]

... and modify attention to solve unknowns.

I particularly like this part. Learning and sensing are active, not just passive; we interact with our environment in intentional ways to increase the useful information we get.

[DaltonG]

"Requirements for functional equivalence to conscious processing?"

First off, I recommend you step back and reconsider where processing occurs. Processing doesn't occur in consciousness. As Damasio states, Consciousness is a Convergence Zone. Consciousness receives the results of preprocessing, and preprocessing is different for each modality, some of which occurs at the level of a modality's sensory field. The outputs from a sensory field may then pass through a variety of preprocessing stages related too the particular modality in question. Common to all is gradient detection which serves to convert analog sensory signals to digital. This makes it possible to perceive Qualia used to discriminate between perceptions of the past with the present (color intensities, prosodic content, etc.). From early stage mapping comes the ability for focus to isolate that which is of interest from peripheral elements in the sensory field while still be aware of the presence of peripheral elements.

Eventually, the results arrive at the level of awareness, a circular buffer, each leg conforming and preserving the modalities output map. The same set of feeds are also fed to the cortical LTM (bottom up priming) triggering related LTM outputs (top down) to the comparison buffer where the inputs from the sensorium are juxtaposed with the contents of the input buffer in awareness. The results of comparison are then passed on to Consciousness as a discrimination between sensory pattern input and memories from the remembered LTM past. One of the primary purposes of Consciousness is the implementation of volitional inhibition which provides for a Ready-Set-Go response execution - timing and control over what's an appropriate response.

Any conscious processing takes the form of Thinking which usually occurs due to a mismatch between bottom up inputs and top down memories. Common in cases like mismatches in context and its effects on meaning).

"My first idea was to create a state of focused multimodal attention."

Normal Attention and Focus can only be trained on one modality at a time. Attention takes 2 forms; as a reflex at the sensory organ level and/or early preprocessing stage (distraction) and as a volitional act. Focus is a physical orientation of the modalities sensory field as the result of the reflex or volitional activity.

I'm going to assume that your reference to a Passive Layer is equivalent to the Unconscious or Subconscious awareness level.