The BBK™
The Brain Building Kit™
Synthetic Reflexes for Robots, Processes, Avatars, and Prostheses
Content Version: 0.1.4
Hover over line items for more information.
INTRODUCTION If we have AI, why do we need a Brain Building Kit?
  • AI is the sizzle. Synthetic Reflexes is the steak
    AI expresses human intention through machines. Expert systems benefit from human expertise. AI does not emulate brain structure/function. It captures high-quality high-value solutions including adaptability to problem variations.
    Modern AI began in the late 1950s and early 1960s using ideas of Norbert Wiener and Walter Pitts based on the hypothesis that a neuron collected information and delivered a single pulse when conditions were met.
  • Intelligence is seductive and well-funded
    AI wins chess and go games. AI discovers hidden associations in big data. AI imitates live pilots in autonomous vehicles. AI adds value to products and services.
    From playing chess and go to discovering patterns in big data, countless teams are organized to find methods to extract value using cybernetics and associative connections.
  • but REFLEX is what keeps animals running
    From knee-jerk and gaze vergence to running and throwing, the spinal cord performs sophisticated recruitment of motor system oppositions based on sensory inputs. Autonomic functions and postural maintenance as well as imitating other animals can be done for fun and profit.
  • Both The BBK and real brains exhibit reflexes
    Sir Charles Sherrington (Nobel prize 1932) said "The brain is the crowning achievement of the reflex system". The BBK implements economical, maintainable, smooth, reliable, and injury-resistant reflexes; not intelligence. The BBK maps reflexes into machines.
    This makes products and services cheaper and more reliable.
COMPETITORS Why Are Nervous Systems Hard?
  • No simple models of neurons are published
    Neurons are cells in community acting for common good. Many complex theoretical models are in use. One uses a dendritic antenna and a pulse generator. Another forms association coefficients between "factoids". None resemble actual neurons in structure or function.
  • AI speculations model complex functions
    AI researches abstract solutions to complex problems. Solutions are re-engineered on artificial machines. Implementations may or may not resemble neural solutions. Expert solutions may not embody legacy knowledge.
  • HTM, neural nets, and neuromorphics simplify
    Weakly associated networks can simulate some neural actions. They can connect, strengthen, weaken, and disconnect. Effectively, these are resonators for repeating data. They increase the Signal/Noise ratio over many samples. These can be very effective machines. But the similarities to neurons end quickly on "drill-down".
  • All successfully solve high-value problems
    Scientific and engineering news is very exciting. New ways to use AI and other cybernetics are blooming. It is an exciting time to be in AI. But none are useful for teaching how brains work.
  • But all rely on the legacy model neuron
    The cybernetic model proposes a single neural pulse. The dendrite is an integrator of inputs. When sufficiency is reached, a pulse is generated. The pulse propagates from the axon hillock to boutons. This is provably a minority operation, at best.
IP Novel IP developed over 17 years (1st IP slide).
  • Reflexes are easier to model than intelligence
    Reflexes balance outputs without making complex decisions. It is about guaranteeing position and secretion. No knowledge or expertise is involved. Instead, outputs are to be varied to achieve a goal.
  • The BBK recreates living sensorimotor systems
    Living sensory systems are very information-rich. Living motor systems are capable of refined actions. Both, much more so than is expected in robotics.
  • It emulates materials from nano to macro scale
    Angstrom: Ion/protein interactions for muscle and nerve. Nano: Aggregate sensing/actions by shapes/bundles. Micro: Orgnized opposed aggregates. Macro: Postural/Perceptual consolidation.
  • Operational reflex system need no programming
    Programming is intrinsic and universal. Reflexes are trained, not programmed.
    It's chief value is reducing manufacturing/operating costs
APPLICATIONS Where can The BBK be deployed?
  • Complex life-like robotic reflexes
    Emulated intrinsic structures exhibit emulated actions. Sub-pixel perception and sub-voxel movements. Least-action and necessary/sufficient power-usage.
  • Simple adaptation to new conditions
    A "possible" movement is trained into reflex. Proprioception "selects" adjustments needed. New adaptation remains as "muscle memory".
  • Game avatars respond like living creatures
    Avatars obeying physics can have virtual reflexes. Avatars can learn your strengths and weaknesses. "Knowledge" is an individual property of the avatar.
  • Prostheses behave like natural limbs
    Portions of nervous systems are viable. A prosthesis can have native reflexes. Prosthesis reflexes can learn too.
  • Individual component market value
    The BBK™ can be delivered without specialization. Specialized instances can be trained by expert staff. Synthetic muscle market value is independent of specific use. Synthetic retina likewise.
  • Valuable new insights for clinical neurologists
    Use electrodes to identify/control specific signals. Identify new differential diagnostics for known pathologies.
  • All this without a line of customizing code
    Programming is used to generate a generic BBK™. No programming is used to train/specialize the BBK™
TODO What is left to do?
  • First demonstration robot is being designed
    Some sensors/actuators have not yet been built. Unknown impediments are likely to arise. The scale of innovation suggests there will be many.
  • Bridging concepts still under development
  • Synthetic muscle is being refined
    A muscle analog has been designed, but not built. This design operates very like skeletal muscle. Drill-down analysis shows similarity to twitch units. Signal distributor driver has been designed.
  • Synthetic eyes are being refined
    Mammalian simple lens, complex receptor plane are needed. Insect compound eye is designed, awaiting implementation.
  • Synthetic midbrain and cerebellum are being refined
    Potential FPGA, but sparse SISD to MIMD may be used. Transforms known to the founder will be used.
  • Mack Truck Insurance
    The founder has not taught the principles sufficiently. A team must be taught the principles. This should be a "trusted" core team. Weekly (if not daily) seminars must be included.
COST How much will it cost
  • Salaries, overhead, and other business costs
    To be discussed.
  • Patent fees
    To be discussed.
  • Likely initial launch costs are about $5M
    Startup team size is to be discussed.
  • Ramp up after demonstration projected at $20M
    Sustaining team size is to be discussed.
TIME How long will it take?
  • A gifted staff produces a demo in 6 months
    See OBJECTIVE for gross design elements.
  • A usable BBK for retrofitting robots in 1 year
    Once a demonstration robot is made lessons are learned. Those lessons will feed back into refining The BBK™ Synthetic reflexes should be deliverable in 1 year.
  • The potential has no currently conceived limit
    Meeting our 1st and 2nd goal barely scratches the surface. The potential applications take little imagination. The IOT could be endowed with reflexes.
OBJECTIVE First Demonstration Robot
  • 360° 3D "Fly Eye" camera
  • Six legs with synthetic muscles
  • A BBK with eye input, leg muscle output
  • A switch to choose pursuit/escape
  • It will either chase you, or you can chase it
  • It plays hide and seek
SCIENCE Basics of IP Principles
  • Real neurons do more than cybernetic neurons
  • Neural organs do more than AI clusters
  • Muscles do more than robotic actuators
  • Sensory organs do more than robotic sensors
  • Only principles found in living animals are used
REFLEXES Physiological Information
  • Nature of Physiological Information
  • Sensory to Perceptual Transduction
  • Action to Actuation Transduction
  • Internal Representation and Operations
  • The Reflex Arc
VALIDITY Avoiding Cognitive Bias
  • Types of Cognitive Bias
  • Quality and Absence of Sources
  • Interference from Accepted Theory
  • Speculations, Projections, and Inventions
  • Engineering Viability Test
  • Mathematical Mappability Test
  • Drill Down Observability Test
  • Working Hypotheses
LINKS Early Demonstrations
Copyright©2002-2020 Jonathan D. Lettvin, All Rights Reserved