This notebook expands the creative engineering side of the encyclopedia. Each idea is written as a concept card rather than a procedure.
Many serious research directions start as strange analogies: the eye as a screen, the cortex as a phosphor field, memory as reconstruction instead of storage, or a prosthesis as a learned language instead of a camera replacement.
The useful move is not to ask only “is this real today?” The better sequence is:
What is the wild version?
What is the closest known science?
What is the missing transducer or code?
What harmless model could represent the idea?
What test would make it less vague?
What would make it unsafe or unethical?
That keeps imagination active while still making the idea sharper.
The strongest screenless systems may carry less image and ask more of trained perception.
world
-> selected cue
-> active exploration
-> learned interpretation
-> useful action
The “medium” is not just glass, phosphor, light, current, sound, or touch. It can become the stable relationship between cue and person. The device emits a compressed sign. The user learns how that sign behaves when they move, ask, reach, navigate, or remember.
Design challenge:
The companion roadmap for turning this into prototypes lives in Project Tracks: Perception As The Medium.
The scene skeleton converts a camera feed into a sparse map:
open floor
walls
doors
stairs
people
moving objects
reachable objects
hazards
The output could be visual overlay, spatial audio, tactile feedback, or a future prosthetic cue. It does not try to reproduce surfaces, colors, or textures. It tries to answer “where can I go and what matters?”
Depth-first vision prioritizes distance and collision risk before identity.
Useful channels:
This is less glamorous than restored sight, but it targets immediate mobility value.
The edge lantern extracts object boundaries and projects or encodes them as high-contrast cues. It could be useful for low vision because edges often matter more than texture.
Possible outputs:
Motion-only vision ignores static detail and reports change.
Why it matters:
Any AI-assisted interface should encode uncertainty. A wrong confident cue can be worse than no cue.
Possible confidence cues:
A phosphene font is a learned alphabet of artificial visual sensations. It may include arrows, dots, lines, pulses, sweeps, and simple glyphs.
The concept accepts that phosphenes are not pixels. It asks which percepts are stable enough to become symbols.
An artificial fovea uses gaze or attention to decide where high information density goes. The periphery receives only motion, hazard, and orientation cues.
This mirrors natural vision:
central high acuity + coarse periphery + active eye movement
A retinal interpreter preprocesses sensor data into signals that resemble useful retinal operations:
It does not need to perfectly copy biology. It only needs to give the remaining visual pathway a more usable signal.
Cortical handwriting uses time as a binding mechanism. Instead of presenting a shape all at once, it traces the shape.
letter A
-> left stroke
-> right stroke
-> crossbar
-> perceived form
This borrows from handwriting, gesture, and raster scanning.
A personal percept map records what each stimulation or cue pattern feels like to one person.
Map fields:
The map becomes the display surface.
A cross-modal HUD divides information across channels:
| Feature | Channel |
|---|---|
| Collision risk | Tactile |
| Object identity | Speech or audio icon |
| Direction | Spatial audio |
| Text | Spoken output |
| Motion | Peripheral visual or tactile pulse |
| Confidence | Tone quality or pulse strength |
The goal is not to force every feature through one weak channel.
A low-resolution prosthesis may show a vague object while an AI narrator labels it. The user receives both:
where it is + what it probably is
This could make low-resolution visual cues more useful without pretending they are high-resolution images.
A prosthetic or substitution system could include games that train:
This treats training as part of the interface, not an afterthought.
Different tasks need different encodings.
| Mode | Priorities |
|---|---|
| Walking | path, obstacles, stairs, moving objects |
| Reading | text line, character shape, OCR confidence |
| Faces | face location, identity if known and permitted, expression confidence |
| Kitchen | hot objects, edges, containers, labels |
| Workshop | tool location, sharp edges, hand proximity |
| Outdoors | vehicles, curbs, traffic direction, signs |
RF is weak as a direct “send images into the brain” explanation. It becomes more concrete when treated as a power or communication link to a local transducer.
external controller
-> RF power/data
-> implanted receiver
-> local optical/electrical/acoustic output
-> target tissue
This still has hard implant and safety problems, but the mechanism is clearer.
Ultrasound may be more interesting as a selector or modulator than as a full image carrier.
Concept:
low-bandwidth cue
-> focused ultrasound target
-> temporary modulation
-> paired audio/tactile/visual cue
The idea is to coordinate channels, not create a high-resolution sonic TV in the brain.
Remote magnetic fields are spatially coarse. A local converter would shift the hard part:
magnetic field
-> local magnetoelectric or inductive receiver
-> local signal
This belongs in speculative transducer design, not direct remote memory or image insertion.
A biological phosphor is a conceptual target that lights, signals, or changes state when stimulated. The closest real families are optogenetics, fluorescent markers, and light-sensitive biological tools.
The speculative version asks whether tissue-compatible materials could become safe intermediate converters between external energy and neural activity.
A display shadow mask ensures each channel reaches the intended output region. A neural shadow mask would be any structure or algorithm that prevents stimulation from spreading to the wrong perceptual channel.
Possible forms:
This is a frontier concept. The grounded version is sparse cueing:
small cue set -> user interpretation -> useful behavior
The far-edge version is arbitrary rich image or memory generation without a clear mechanism yet.
These are good for speculative writing or future-facing diagrams, but should be marked clearly:
For any idea, ask: