Date post: | 17-Dec-2015 |
Category: |
Documents |
Upload: | evangeline-elliott |
View: | 212 times |
Download: | 0 times |
Agents, artifacts and Agents, artifacts and innovationinnovation
David Lane, Univ Modena and Reggio Emilia David Lane, Univ Modena and Reggio Emilia and Santa Fe Instituteand Santa Fe Institute
Roberto Serra and Marco Villani, UMRERoberto Serra and Marco Villani, UMRE
ISCOM project (FET-IST)ISCOM project (FET-IST)
Premise: a theory of artifact Premise: a theory of artifact innovationinnovation
All artifacts have a history -- as do the modes of All artifacts have a history -- as do the modes of interaction among people in which artifacts figureinteraction among people in which artifacts figure
Propose to develop a theory about the processes Propose to develop a theory about the processes through which artifact histories are realizedthrough which artifact histories are realized How new artifacts come into beingHow new artifacts come into being How their tokens proliferate and become incorporated into How their tokens proliferate and become incorporated into
patterns of human interactionpatterns of human interaction How new patterns of interaction among human beings and the How new patterns of interaction among human beings and the
artifacts they create are generatedartifacts they create are generated
Primary questionsPrimary questions What kinds of social structures support the processes of artifact What kinds of social structures support the processes of artifact
innovation?innovation? How do these structures modulate the conflicting functionalities that How do these structures modulate the conflicting functionalities that
underlie proliferation of artifact tokens and generation of new artifact underlie proliferation of artifact tokens and generation of new artifact types?types?
How do these structures transform themselves as they incorporate new How do these structures transform themselves as they incorporate new kinds of functionality around new kinds of artifacts?kinds of functionality around new kinds of artifacts?
Positive feedback dynamic,Positive feedback dynamic, linking proliferation of new artifacts, new linking proliferation of new artifacts, new patterns of human activity that organize around these artifacts, patterns of human activity that organize around these artifacts, generation of new types of artifacts…generation of new types of artifacts…
Theory as ontologyTheory as ontology
““Theory of innovation” an oxymoron?Theory of innovation” an oxymoron? Aim to present an ontology for phenomena associated Aim to present an ontology for phenomena associated
with artifact innovationwith artifact innovation Kinds of entitiesKinds of entities Interaction modalitiesInteraction modalities How entity properties change as a result of interactionsHow entity properties change as a result of interactions
Value of theory established and demonstrated through Value of theory established and demonstrated through dialoguesdialogues With historical narratives With historical narratives books Lonworksbooks Lonworks With mathematical modelsWith mathematical models
Reciprocality principleReciprocality principle
Agents: human beings, or organizations “in the name of Agents: human beings, or organizations “in the name of which” human beings actwhich” human beings act
Artifacts: entities constructed by human beings, for the Artifacts: entities constructed by human beings, for the use of human beingsuse of human beings
Reciprocality principle: Reciprocality principle: the generation of new artifact types is mediated by the transformation of relationships among agents; and the generation of new artifact types mediates the transformation of relationships among agents.
Agent-artifact spaceAgent-artifact space
Horizontal and vertical relationsHorizontal and vertical relations Network tiesNetwork ties
Agent-agent: recurring patterns of interaction, directed towards Agent-agent: recurring patterns of interaction, directed towards transformations of artifacts, artifact relations, or agent-artifact transformations of artifacts, artifact relations, or agent-artifact relationsrelations
Agent-artifacts: relations of production, ownership, useAgent-artifacts: relations of production, ownership, use Artifact-artifact: functional substitutibility, complementarity (co-Artifact-artifact: functional substitutibility, complementarity (co-
use)use)
Recursive structuresRecursive structures
Agent propertiesAgent properties
Resources Resources artifactsartifacts Permissions Permissions potential field of relationshipspotential field of relationships Cognitive structures Cognitive structures shared, distributedshared, distributed
Attributions and Attributions and directednessdirectedness NarrativesNarratives
Two kinds of innovationTwo kinds of innovation
Better-faster-cheaperBetter-faster-cheaper New attributions of functionalityNew attributions of functionality
Organization of agent-artifact Organization of agent-artifact spacespace
Locus of new attributions: Locus of new attributions: generative generative relationshipsrelationships
Market systemsMarket systems Competence networksCompetence networks Scaffolding structuresScaffolding structures
Model and theoryModel and theory
The theory provides The theory provides an ontology that determines the model entities and processes an ontology that determines the model entities and processes a language to describe the different “histories”a language to describe the different “histories” some specific assumptions and consequences that can be some specific assumptions and consequences that can be
tested, in a mathematical model or by confrontation with datatested, in a mathematical model or by confrontation with data The model to be described here allows a precise statement of the The model to be described here allows a precise statement of the
theory, theory, in a limited contextin a limited context Developing the model and analyzing its results can give rise to a Developing the model and analyzing its results can give rise to a
process leading to modifications of the theory itselfprocess leading to modifications of the theory itself
Constraints from the theoryConstraints from the theory
ReciprocalityReciprocality=> Agents, artifacts and relations among them must be => Agents, artifacts and relations among them must be representedrepresented
Innovation leads to modifications of the Innovation leads to modifications of the role of agents as of agents as well as of the well as of the meaning of artifacts of artifacts=> both must be endogeneously generated=> both must be endogeneously generated
External fitness functions make no senseExternal fitness functions make no sense DirectednessDirectedness (towards transformations of artifact space) (towards transformations of artifact space)
=> agents have intentionality=> agents have intentionality
Model: homo faberModel: homo faber
agents use artifacts, produced by other agents, agents use artifacts, produced by other agents, to build artifacts, which can be used by yet other to build artifacts, which can be used by yet other agentsagents
the meaning of artifacts is defined by which the meaning of artifacts is defined by which agents use them, for whatagents use them, for what
the role of agents is defined by what they make, the role of agents is defined by what they make, and by the agents with which they interactand by the agents with which they interact
gift economygift economy
ArtifactsArtifacts
Artifacts are (currently) represented by Artifacts are (currently) represented by numbersnumbers
Agents produce (numbers) by applying Agents produce (numbers) by applying functions (to numbers)functions (to numbers)
Agent propertiesAgent properties
production recipes production recipes (+ - 2 10 7) produces 5(+ - 2 10 7) produces 5
goalsgoals i.e. (roughly) what new i.e. (roughly) what new
artifacts it wants to produce artifacts it wants to produce (more precisely: from (more precisely: from which existing artifacts it which existing artifacts it wants to exapt)wants to exapt)
stocksstocks
list of artifacts known to list of artifacts known to the agent the agent
list of other agents known list of other agents known to the agentto the agent
score that it attributes to score that it attributes to its relationships with other its relationships with other agentsagents
““style” style” A set of parameters which A set of parameters which
determines the propensity determines the propensity of the agent to innovate, of the agent to innovate, etc.etc.
Standard dynamics, without Standard dynamics, without innovationinnovation
at each time step an agent at each time step an agent at random is selected for at random is selected for updatingupdating
for each recipe, it searches for each recipe, it searches (among the stocks of its (among the stocks of its suppliers) for the required suppliers) for the required inputsinputs
if the inputs are found, the if the inputs are found, the product is produced and product is produced and inserted immediately into inserted immediately into the stockthe stock
the stocks of the suppliers the stocks of the suppliers are reducedare reduced
If one of the inputs is not If one of the inputs is not found, the agent searches found, the agent searches for another artifact of the for another artifact of the same type from its same type from its acquaintancesacquaintances
If found, same as beforeIf found, same as before Otherwise, the agent Otherwise, the agent
passes to the following passes to the following reciperecipe
and the counter and the counter associated to the unused associated to the unused recipe is updatedrecipe is updated
External market
A
B
C
1
D
E
2 F
G
3H
I
J
4
K
5 L
M
6
NO
Raw
mat
eria
l
Initial conditions: external market
InnovationInnovation
an agent can modify its an agent can modify its productsproducts (by creating a (by creating a new new reciperecipe), OR), OR
it can enhance its it can enhance its knowledge of other agentsknowledge of other agents
Creating new recipes: goalsCreating new recipes: goals
the agent the agent looks at the portion of artifact space which it knowslooks at the portion of artifact space which it knows chooses a specific goal to pursue, based upon an chooses a specific goal to pursue, based upon an
artifact which it does not produceartifact which it does not produce tries to “come close” to that goal by using the tries to “come close” to that goal by using the
available operators on the available inputsavailable operators on the available inputs if the agent succeeds (within a given range), it if the agent succeeds (within a given range), it
puts the new recipe in its set of active recipesputs the new recipe in its set of active recipes
Goal setting by imitationGoal setting by imitation
A
B
C
D
E F
G
H
N
ML
J
P
O
I
K
Imitation (of a randomly selected artifact) is a simple way to sample artifact space
the clusters of artifacts are likely to be high reward zonesThe agent tries to build an artifact similar (within a given threshold) to the selected one (which represents the “goal” of the imitation)
Q
Generating a new recipeGenerating a new recipe
new recipes can be generated bynew recipes can be generated by Changing the order of the inputs (leaving the support Changing the order of the inputs (leaving the support
unaltered)unaltered) changing some inputschanging some inputs Changing the order of the operatorsChanging the order of the operators Changing the operators (eg by crossing different Changing the operators (eg by crossing different
recipes)recipes)
Imitation worldImitation world
Imitation can produce self-sustaining Imitation can produce self-sustaining loops, and the system is able to surviveloops, and the system is able to survive
Introduction of successful novelties ends Introduction of successful novelties ends after a transient period of timeafter a transient period of time
Goal setting by “jumping”Goal setting by “jumping”
A
B
C
R
Q
OF
D
I
E M
L
G
N
PSH
The jump allows The jump allows the exploration of the exploration of new regions of new regions of artifact spaceartifact space
Social innovationSocial innovation
In general, an agent has a limited knowledge of In general, an agent has a limited knowledge of artifacts and other agentsartifacts and other agents
An agent can also try to innovate together with An agent can also try to innovate together with another agent, chosen from among those it another agent, chosen from among those it knowsknows
Agents that cooperate share recipes and cross Agents that cooperate share recipes and cross them to create new onesthem to create new ones
Agents can acquire new ties, with whom they do Agents can acquire new ties, with whom they do not have a customer-supplier relationshipnot have a customer-supplier relationship
Question 1: IntentionalityQuestion 1: IntentionalityWhat difference do goals make?What difference do goals make?
This can be investigated by comparing a goal-This can be investigated by comparing a goal-directed system, with a system without goalsdirected system, with a system without goals
The non-goal-oriented system (NG) is generated The non-goal-oriented system (NG) is generated by allowing an agent to develop a new recipe by by allowing an agent to develop a new recipe by combining some of its inputs and operators combining some of its inputs and operators chosen at randomchosen at random
Diameter of artifact spaceDiameter of artifact space LEFT: WITH GOALS LEFT: WITH GOALS
RIGHT: WITHOUT GOALSRIGHT: WITHOUT GOALS
Average number of recipes per agent, fraction of known Average number of recipes per agent, fraction of known agentsagents
What difference do goals make?What difference do goals make? Preliminary answersPreliminary answers
No-Goal world is more robust (and No-Goal world is more robust (and predictable)predictable)
Goals generate much broader and Goals generate much broader and diversified artifact space, with greater diversified artifact space, with greater diversity in agent fatesdiversity in agent fates
Question 2: Innovative activityQuestion 2: Innovative activityWhat is the relationship between agents’ propensity
to innovate and structure of artifact space?
We can vary parameters that control agents’ We can vary parameters that control agents’ innovation rate, jump frequency, jump sizeinnovation rate, jump frequency, jump size
We can compare systems in which agents all have We can compare systems in which agents all have the the samesame innovation propensities with systems in innovation propensities with systems in which they are which they are heterogeneousheterogeneous
Homogeneous agents:Homogeneous agents: left: diameter (per agent) left: diameter (per agent)
right: number of artifacts currently in systemright: number of artifacts currently in system
Innovators prevail in heterogeneous system:Innovators prevail in heterogeneous system: left: dist of innov prob; left: dist of innov prob;
right: no. of successful projects as function of innov probright: no. of successful projects as function of innov prob
QuickTime™ and aTIFF (LZW) decompressor
are needed to see this picture.
Varying jump frequency and rangeVarying jump frequency and range
High jump probability in homogeneous systems High jump probability in homogeneous systems increases the diameter of artifact space, but increases the diameter of artifact space, but makes the system more fragile (a higher makes the system more fragile (a higher proportion of agents die) -- the more agents, the proportion of agents die) -- the more agents, the more fragilemore fragile
In heterogeneous systems, the frequent jumpers In heterogeneous systems, the frequent jumpers develop fewer successful projects develop fewer successful projects
BUT if jump BUT if jump rangerange also varies, a mixture of short also varies, a mixture of short and long jumpers performs very well -- and the and long jumpers performs very well -- and the latter outperform the former! latter outperform the former!
Question 3: agent relationshipsQuestion 3: agent relationshipsWhat produces stable relationships? What produces stable relationships?
Are these generative?Are these generative?
We compare system effects when agents We compare system effects when agents choose innovating partners randomly with choose innovating partners randomly with those when agents use a criterion to those when agents use a criterion to choose partnerschoose partners
Criteria examined include: past success; Criteria examined include: past success; aligned directedness; role heterogeneityaligned directedness; role heterogeneity
Scoring relationships on the Scoring relationships on the basis of past successbasis of past success
V ij t 1 V ij t tij V ij t
tij
2 if there isa successful jo int project at time t
0 otherwise
Choosing partners accordingChoosing partners according to V to V
This criteria gives rise to reciprocal This criteria gives rise to reciprocal relationships that are very stable in time relationships that are very stable in time (though rarely last “forever”)(though rarely last “forever”)
Partnering on the basis of V gives rise to Partnering on the basis of V gives rise to substantially richer artifact spaces than substantially richer artifact spaces than does random pairingdoes random pairing
Formation of successful partnershipsFormation of successful partnershipsLeft: partnering based on VLeft: partnering based on VRight: random partneringRight: random partnering
0
0,1
0,2
0,3
0,4
0,5
0,6
0,7
0,8
20 40 60 80 100
120
140
160
180
200
220
240
260
280
Number of projects per agent
Fre
qu
ency
00,050,1
0,150,2
0,250,3
0,350,4
0,450,5
20 40 60 80 100
120
140
160
180
200
220
240
260
280
Number of projects per agent
Fre
qu
en
cy
Generative potentialGenerative potential
Aligned directedness Aligned directedness closeness of goals Heterogeneity Heterogeneity difference between
variance of outputs -- specialists pair with generalists
Mutual directedness Mutual directedness satisfying history -- V PermissionsPermissions Action opportunitiesAction opportunities
Partnering on generative Partnering on generative potentialpotential
All three measured criterion alone produce All three measured criterion alone produce results similar to those for Vresults similar to those for V
Some combinations tend to do even better Some combinations tend to do even better than the single criteria…than the single criteria…