product evolution: computer-aided recombinant design by customer-driven natural selection kamal...
TRANSCRIPT
Product Evolution:Computer-aided Recombinant Design by Customer-driven Natural Selection
Kamal Malek Noubar Afeyan
MIT Media Lab / The Center For Bits and Atoms
Meeting on Emergent Engineering
October 16, 2002
Confidential – Do not Duplicate
Evolving New Products
Step 1Sequence the
Product Genome
“Featurize “/ code the product as a set of genes
Step 2Create
Recombinant Designs and Expose
them to “Natural Selection”
Allow customers to interact and choose over the web
Step 3 Evolve the “Fittest”
Customers/ developers select, refine to produce optimal designs
Confidential – Do not Duplicate
Confidential – Do not Duplicate
Confidential – Do not Duplicate
So What’s the IDEA?
• IDEA: Interactive Design by Evolutionary Algorithms (patents pending)
• Featurize design into genes and define alleles• Consumer votes through web on relative appeal of
recombinant designs • Preferences extracted through multiple generations• Collective action – segmentation – preference
extraction• Results: Designs, Insights, Affinity, Innovation
Confidential – Do not Duplicate
IDEA: Interactive Design by Evolutionary Algorithms
A set of design A set of design direction direction candidates and a candidates and a description of the description of the underlying design underlying design intentintent
Featurize Featurize candidates candidates extracting significant extracting significant design features and design features and attributes, attributes, establishing an establishing an allowable range of allowable range of variationvariation, and , and identifying identifying design design constraintsconstraints
This “featurization” is This “featurization” is encoded into the encoded into the design design genotypegenotype in a way that in a way that enables new enables new design design candidatescandidates to be to be generated automatically generated automatically within the vast design within the vast design universe defined around universe defined around the stated design intentthe stated design intent
Recombinant Design
Confidential – Do not Duplicate
Evolutionary Algorithms Evolve the “Fittest” Designs
Initial Design PopulationInitial Design Population
Evolutionary AlgorithmEvolutionary Algorithm
Fitness AssessmentFitness Assessment
Fitness WeightedFitness WeightedBreedingBreeding
MutationsMutations
More Fit PopulationMore Fit Population
Confidential – Do not Duplicate
10 Features of 10 Options Potential Design Population: 10 Billion Designs
Discovery of preferred product designs and market segments
Consumer Population5,000 Heterogeneous Users
Confidential – Do not Duplicate
Examples
PACKAGING
CONCEPTTESTING
PROMOTIONALDESIGN
PRINT MEDIA
PRODUCTS
Confidential – Do not Duplicate
Model
Logo
Background
Perfume Bottle
Ad Layout
Confidential – Do not Duplicate
Crossover
Confidential – Do not Duplicate
Mutation
Confidential – Do not Duplicate
1,000+ Participants Explored A Design Space Rendered In Real-Time
Potential Design Universe:Over 340,000+
Possible Advertisements!
Confidential – Do not Duplicate
1
2
5
4
3
1,000+ Participants Converged on 5 Major Design Themes from a Design Universe of 340,000+
100%
Confidential – Do not Duplicate
Summary
• First steps toward evolving products through computation, customer selection and web-based collectivism
• Insights into segmentation• Enabling to designers, market
researchers as well as product management