supporting the complex requirements of a long-term project for whole brain emulation
TRANSCRIPT
Supporting the complex requirements of a Supporting the complex requirements of a long-term project for whole brain emulationlong-term project for whole brain emulation
Dr. Randal A. KoeneDr. Randal A. KoeneCarboncopies.orgCarboncopies.org
MTA2014
OVERVIEW
Brief Basics + Rationale for WBE
Success of Long-Term Project with Complex Requirements
Proposed Method & Example from Experience
A Piece-Wise, Sustainable Path to WBE
BASICS: TERMINOLOGY
http://koene.carboncopies.org
Objective: SIMSubstrate-Independent Mind
Feasible method: WBEWhole Brain Emulation
Bio to WBE transfer:“mind uploading”
BASICS: PURPOSE
In Search of Everything
Personal limitations
To Understand Everything
To Create Anything
Enhance & Expand Capabilities: Need ACCESS
Brain Interfaces & ultimately SIM give maximum access & maximum flexibility
BASICS: ADAPTABILITY FOR OUR SPECIES
Suitable for this environment, this little niche in time and space
Bigger picture perspective: Many challenges we are not at all suited for
Changes are coming, we may even be causing them!
Species Survival & Species Thriving require excellence in adaptabilityResponsibility to prepare our species while we can
BASICS: REVERSE ENGINEERING A MIND
Small Brain Prostheses show it IS POSSIBLE: Cochlea, Retina & HIPPOCAMPUS
Hippocampus: Episodic Memory
Biomimetic chip replaces function
Works in rats (many experiments), works in monkeys
3 years to human trials!
Berger et al, USC
BASICS: SYSTEM IDENTIFICATION
Unknown Sysem (bio / chip) = black box with I/O
Know your goal, discover relevant signals(chip example: 1s & 0s)
Berger hippocampus: assume “spike” times are relevant: evidence in sensory, motor & synaptic change
Learn transfer function for input spike patterns to output spike patterns
NOT the same as a 'simulation'that abstracts function
BASICS: WBE IS MANY CONNECTED PROSTHESES
Whole Brain roadmap: Break black box with 100B neurons into many connected sub-systems as feasible or smaller than Berger's
Connections tell us how sub-systems can interact
Characterize function within each sub-system
Emulate dynamic function representations
Test & Iteratively improve
Structure: ConnectomeProof-of-concept 2012
Function: each neuron?Current Aim!
Emulation: modelplatform
Application &Validation
MISSION: LONG TERM & MANY REQUIREMENTS
Tech. requirements to iterate towards on 4 pillars
Actual R&D activitiy based on short-term interests
Trends may not lead to goals without a special effort
What may happen if other trends dominate, e.g. AI?
MISSION: CARBONCOPIES.ORG
Maintain a roadmap:Know goals in terms of requirements+ actual activity / trends
Top-Down needsBottom-Up feasibility
Encourage meeting requirements and filling gaps by emphasizing / starting specific activities
Big-picture vigilance: apply where most needed
METHOD: EXAMPLE, 2013 FOCUS
High-resolution functional data acquisition 'every neuron'
Boyden (MIT) to Silicon Valley 2012: Molecular Ticker Tape & Brain Activity Map (BAM)
Carmena (UC Berkley) interestin WBE, connect with Boyden: joined BAM, ultrasound + wireless probe at micron scales
PoBAM group: 'every neuron at 1ms' – BRAIN Initiative
METHOD: EXAMPLE, PROTOTYPES EMERGING
'Neural Dust', 126um – 20um, 5um tail, ultrasound power & communication
Animal trials in June 2014!
MIT/Harvard: similar, with infrared 'RFID-like' system
Kording: bio-vector 'pulled' nanowire array
Cross-disciplinary 'hybrid' solutions toward a BrainNet
METHOD: EXAMPLE, BMI STRATEGY
Long-term operational interfaces to very large number of neurons without harm to the brain
Hi-Res BMI: In-Brain-LoopUse data otherwise inaccessible (sort memory)Deliver data otherwise impossible to generate (parameter fitting, super object recognition, etc.)
Needs to understand brain circuit protocol (see Berger)
SUSTAINABLE: PIECE-WISE TO WBE
Interfaces – Prostheses – WBE more feasible than 'clean-room' path
Pieces, e.g. retina, hippocampus & small whole brains (drosophila 2018?)
Sustainably to WBE – if a body attends to big-picture, iteratively deciding nudges
SUSTAINABLE: INCENTIVES FOR 'ARCHITECTING'
Architecting: e.g. PoBAM & Carboncopies.org
Personal experience of fragile support for cross-disciplinary big-picture(shared problem...)
Grant funding encourages specialized silosInvestment encourages short-term focus
Long-term success probability linked to robustness
Effort along multiple possible solutions+Sharing for hybrid solutions
SUSTAINABLE: ROBUST MISSION DESIGN
No single points of failure:
● Technology● Labs & People● Resources for the work
Technology & Labs & People well structured
Frailty of resources continues to hamper predictable architecting (like carboncopies.org)
Robustness: A 2014 goal.
SUMMARIZING
SIM: Improve ourselves while we improve our machines
Trends & exponential curves don't automatically lead to goals
WBE monitored and nurtured: WBE is now lab-acceptedConnectome proof-of-conceptPoBAM – functional access5 years to fruitfly WBE project?
Robustness: Structural improvements needed!
WBE Coffee Table Book