André’s Reviews > How the Body Shapes the Way We Think: A New View of Intelligence > Status Update

Examples, Summaries
・understand by building
・unseen intel
・insects=ubicomp=companies
・infostructuring thru sensorymotor coord: phys.motion=selfstimulate diff sensory channels→correlat.; proper morphology=rich data w lil effort; coord reduces complxty
・offload neuralwork into morphology+env
・intrinsic dynamics→achieve tasks for free: Passive-dyn.Walker; self-stabiliz.
・brain-body mutual control
・art. evol in realworld

Where is memory
・"storehouse" (mem as stored structs/comp.analogy)=problems of cognitive paradigm→symbol grounding?
・mem never directly observed→inferred from behavior
・bellshape of water fontain not stored as struct inside fontain but emergent→brain=dyn sys
・Bovet "Delayed Reward Learning w/o Mem": art. simple neuralnet mouse offloads mem into env→exploits interaction w env
・sensory-motor for recogn, mem, categories

Company as Embodied Intellig Sys:
・adapt to & exploit unpredict. environm/ecol niches
・top-down mgt vs initiatives of lower employees: "emergent nature of strategy" (Mintzberg)
・managers mimetic behavior but seeing≠understand (frame of ref)
・constituents: env, behvrs/tasks, agent
・design no part in isolation (complete agent)
・uncertainty→dont wait for info→provoke fdback (bodily interact w env→sensory stimul→learn)
…

UbiquitousComp
・Sakamura TRON house 1988; MarkWeiser
・structure the environm (3-constituents principle)
・agent design: smartphones are parasitic on humans, they let themselves be carried around =exploit ecol. niche
・ubi comp nets≠Internet, bc embodied: sensory&motor, auton. env-interact

Cybernetic+Organic/Cyborg
・offload mem to env (writing)
・prostheses
・Mussa-Ivaldi lamprey
・Potter: rat neurons multielectrode array

Collective Intel:
・design global behavior emergent from simple rules of local interaction: Racial segregation in society [ThomasCSchelling, Epstein&Axtell], ant algo, flocks (birds); self-org=adaptive, robust, cheap, scalable, req less ctrl
・agent design principles applicable to groups of agents
・general purpose vs specialized modules (no cell div in tech); simulate specialization by turn off features vs cheap design

Evolution:
・populations=evol's most valuable resource; non-diverse agents vs env. change→all perfom eq badly→evol leads nowhere
・forces: cumulative selection, variation, self-org
・evolve brain & body together, vs neural net ctrlr for prefab. body
・ScalableComplxty: genome encodes process not structure; process modeled as GeneticRegulatoryNetw
・fluid: agent=large num of units (cells)→small mods or disruptive (fitness)

Adrian Thompson used FPGAs to test evolved circuits. Because he was doing evolution in the real world (no simulator), the evolved circuits started to exploit the material properties of the FPGA itself! Thompson tried to figure out which components on the FPGA were connected by removing the non-wired ones→The circuit no longer worked due to weak el.magnetic interactions betw the disconnected components and the circuit

DevelopmentalRobotics:
・in order to learn ctrl of own brain-body, organisms start simple→new skills from old; start full complex=no learn
・body image: self-explore body→discover attractor states
・social interact→sensory-motor
・motiv: dev or happy?

Evol.Algos:
・way we are made shapes way we conceptualize world & design art. intelligence→artificial evol. no DesignerBias→different forms of intel
・break complxty barrier

Design principles of (art) intel bio-systems:
・{ecol. niche/environm, tasks, agent}
・CheapDesign: exploit niche→gravity etc vs compute
・ParallelProcsss, loosely coupled thru interact with env
・CompleteAgent vs parts
・SensoryMotorCoord→self-gen snsry stimultn
・EcoBalance in complxty of sensory/motor/neural; Task distrib btw morphlgy/materials/ctrl/env-inter
・Value: driving forces
・Redndcy
・FrameOfReference
・TimeScales

・synthetic method: understand by building
・"embodied intel" 1985s reaction against intel as abstract symbol processing (algo. approach)
・turn from human to insect-like intel
・common sense ≠ set of rules→req. real world interaction
・disembod. ANN [prepared data]
・Brooks "Intel without Representation"→avoid complex model of outside world, just look at it: "World is its own best model"→system-env. interaction→body/robot