Knowledge Vault 3/11 - GTEC BCI & Neurotechnology Spring School 2024 - Day 1
Making better BCIs: BCIs create synthetic heksors
Jon Wolpaw, National Center for Adaptive Neurotechnologies (USA)
<Resume Image >

Concept Graph & Resume using Claude 3 Opus | Chat GPT4 | Llama 3:

graph LR classDef bci fill:#f9d4d4, font-weight:bold, font-size:14px; classDef cns fill:#d4f9d4, font-weight:bold, font-size:14px; classDef hexer fill:#d4d4f9, font-weight:bold, font-size:14px; classDef skills fill:#f9f9d4, font-weight:bold, font-size:14px; classDef improvements fill:#f9d4f9, font-weight:bold, font-size:14px; A[Jon Wolpaw] --> B[BCIs are slow,
unreliable, untrusted 1] A --> C[BCIs decode CNS
to 'read mind' 2] A --> D[Brain acquires skills
via muscles 3] A --> E[BCI skills learned
through practice 4] A --> F[BCIs create non-muscular
CNS skills 5] D --> G[Skill performance varies,
result stable 6] D --> H[CNS changes continually,
skills maintained 7] D --> I[Skill substrate: distributed
neuron network 8] D --> J[Skill acquisition involves
brain plasticity 9] I --> K[Simple skills involve
brain plasticity 10] H --> L[How are skills
maintained, negotiated? 11] I --> M[Skill networks: changing,
negotiating properties 12] H --> N[CNS maintains skills
via negotiation 13] I --> O[Skill substrate termed
'hexer' network 14] O --> P[Hexers maintain negotiated
skill equilibrium 15] O --> Q[Hexers: 'essentially perfect'
skill maintainers 16] O --> R[Examples: finger flexion,
reflex networks 17] O --> S[Hexers negotiate asymmetric
skill effects 18] S --> T[Motor neurons negotiated
between hexers 19] S --> U[Rat reflex conditioning
shows negotiation 20] O --> V[Hexer produces, maintains
muscle skill 21] F --> W[BCI creates synthetic
hexer network 22] W --> X[Synthetic hexers lack
natural advantages 23] X --> Y[Potential improvements: feedback,
redundancy, negotiation 24] F --> Z[BCIs integrate synthetic
hexers naturally 25] A --> AA[Collaborators acknowledged, especially
Dennis McFarlane 26] A --> AB[BCIs create skills,
hexers maintain 27] AB --> AC[Hexers maintain negotiated
CNS equilibrium 28] AB --> AD[Synthetic hexer integration
improves BCIs 29] A --> AE[Better BCIs primarily
a neuroscience problem 30] class A,B,C,E,F,W,X,Z,AB,AD,AE bci; class D,G,H,J,K,L,N cns; class I,M,O,P,Q,R,S,T,U,V,AC hexer; class AA improvements;


1.-Existing BCIs are slow and unreliable, presenting the "Grand Canyon problem" - they are not trusted for critical tasks.

2.-The prevailing notion is that BCIs decode CNS signals to "read the mind", but this view has limitations.

3.-The brain acquires adaptive behaviors/skills which are normally produced by muscles. BCIs convert CNS activity into new non-muscular skills.

4.-BCI skills, like natural skills, are learned and acquired through practice, as evidenced by animal and human studies.

5.-BCIs aim to create new non-muscular skills, so the predominant approach should be neuroscience to integrate them into the CNS.

6.-A skill is not performed the same way each time - trajectories vary but the result is stable and predictable.

7.-The CNS changes continually throughout life via neurogenesis, synaptic changes, glial changes, etc. Skills are somehow maintained despite this.

8.-The CNS substrate of a skill is a distributed network of neurons and synapses, not a specific synaptic location.

9.-fMRI studies show skill acquisition involves overlapping plasticity in cortex, subcortical areas, cerebellum and spinal cord.

10.-Simple skills like knee-jerk reflex conditioning also involve a distributed network of brain and spinal cord plasticity.

11.-Key questions: How are skills maintained in a changing CNS? How do skill networks retain key attributes and negotiate properties?

12.-New paradigm: Skill networks have two special properties - 1) Changing continually to maintain key skill attributes, 2) Concurrently negotiating shared neurons/synapses.

13.-The aggregate process is a negotiation, keeping the CNS in a negotiated equilibrium that maintains skills, similar to a Nash equilibrium.

14.-The CNS substrate of a skill is termed a "hexer" - a network that produces and maintains a skill.

15.-Together, hexers maintain a negotiated equilibrium that enables skill preservation and acquisition - a property noted by Nikolai Bernstein in 1967.

16.-Hexers are named after the Greek word meaning "essentially perfect" to imply their special skill-maintaining properties.

17.-Examples of hexers include the finger flexion sequence learning network and the spinal reflex conditioning network.

18.-Hexers must negotiate as the nervous system is affected by acquisition of new asymmetric skills like discus throwing.

19.-Axon initial segment of spinal motor neurons, the final gatekeeper of behavior, is negotiated between hexers to maintain skills.

20.-Soleus H-reflex conditioning in rats shows negotiation between the new reflex hexer and old locomotor hexer to maintain symmetrical gait.

21.-A hexer is a network that produces a muscle-based skill and changes as needed to maintain the skill's key attributes.

22.-A BCI creates a synthetic hexer - a network of neurons, synapses and software that produces and maintains a BCI skill.

23.-Synthetic hexers currently lack advantages of natural hexers, limiting BCI reliability. Relevant neuroscience questions about hexers remain unanswered.

24.-Potential improvements: Appropriate starting point, better sensory feedback, signals from multiple areas, redundancy, ability to negotiate, ongoing updating.

25.-Creating better BCIs is foremost a neuroscience problem of integrating synthetic hexers into the natural hexer negotiated equilibrium.

26.-The talk covered work by many collaborators and was supported by several funding organizations. The late Dennis McFarlane was especially acknowledged.

27.-Key conclusions: BCIs create non-muscular skills. The CNS substrate of a skill is a plastic network called a hexer.

28.-Hexers change continually to maintain key skill attributes and keep the CNS in a negotiated equilibrium. BCIs create synthetic hexers.

29.-Synthetic hexers are currently slow and unreliable. Integrating them into the expanded negotiated equilibrium is key to improvement.

30.-While bioengineering and signal analysis are important, creating better BCIs is primarily a neuroscience problem. Difficult scientific issues remain unresolved.

Knowledge Vault built byDavid Vivancos 2024