Explaining the evolution of technologies, firms, and industries: how complexity and constraints can be simulated

Janne M. Korhonen
janne.m.korhonen@aalto.fi / www.slideshare.net/jmkorhonen
Research seminar 29.10.2010, Aalto School of Economics
Explaining the evolution of
technologies, firms, and industries:
How complexity and constraints can be simulated

Background
PhD research topic: The effects of resource
scarcity on innovation (prof. Liisa Välikangas)
Research question(s): why scarcities sometimes
result to significant innovations, but not always?
Framework: evolutionary theory and complex,
non-linear behavior (complexity); information
theory; theory of computation; simulation models
Unit of analysis: technology/system as artefact
Why I’m here: to outline some ideas!

Agenda
How the world works - information and evolution
Systems as strings of information
Problem-solving as search process
Why interdependency matters
Fitness landscapes
Development of flash smelting
Effects of scarcity, and measuring innovation
Some implications

Why...
...some technologies get adopted, and others are
discarded?

Copper smelters using particular technologies,
1900-2000

To find out why,
let’s take a look how the world works

Saccharomyces cerevisiae (baker’s yeast)
TCP1-beta gene from www.ncbi.nlm.nih.gov
...describing a system
1 GATCCTCCAT ATACAACGGT ATCTCCACCT CAGGTTTAGA TCTCAACAAC
GGAACCATTG
61 CCGACATGAG ACAGTTAGGT ATCGTCGAGA GTTACAAGCT AAAACGAGCA GTAGTCAGCT
121 CTGCATCTGA AGCCGCTGAA GTTCTACTAA GGGTGGATAA CATCATCCGT GCAAGACCAA
181 GAACCGCCAA TAGACAACAT ATGTAACATA TTTAGGATAT ACCTCGAAAA TAATAAACCG
241 CCACACTGTC ATTATTATAA TTAGAAACAG AACGCAAAAA TTATCCACTA TATAATTCAA
301 AGACGCGAAA AAAAAAGAAC AACGCGTCAT AGAACTTTTG GCAATTCGCG TCACAAATAA
361 ATTTTGGCAA CTTATGTTTC CTCTTCGAGC AGTACTCGAG CCCTGTCTCA AGAATGTAAT
421 AATACCCATC GTAGGTATGG TTAAAGATAG CATCTCCACA ACCTCAAAGC TCCTTGCCGA
481 GAGTCGCCCT CCTTTGTCGA GTAATTTTCA CTTTTCATAT GAGAACTTAT TTTCTTATTC
541 TTTACTCTCA CATCCTGTAG TGATTGACAC TGCAACAGCC ACCATCACTA GAAGAACAGA
601 ACAATTACTT AATAGAAAAA TTATATCTTC CTCGAAACGA TTTCCTGCTT CCAACATCTA
661 CGTATATCAA GAAGCATTCA CTTACCATGA CACAGCTTCA GATTTCATTA TTGCTGACAG
721 CTACTATATC ACTACTCCAT CTAGTAGTGG CCACGCCCTA TGAGGCATAT CCTATCGGAA
781 AACAATACCC CCCAGTGGCA AGAGTCAATG AATCGTTTAC ATTTCAAATT TCCAATGATA
841 CCTATAAATC GTCTGTAGAC AAGACAGCTC AAATAACATA CAATTGCTTC GACTTACCGA
901 GCTGGCTTTC GTTTGACTCT AGTTCTAGAA CGTTCTCAGG TGAACCTTCT TCTGACTTAC
961 TATCTGATGC GAACACCACG TTGTATTTCA ATGTAATACT CGAGGGTACG GACTCTGCCG
1021 ACAGCACGTC TTTGAACAAT ACATACCAAT TTGTTGTTAC AAACCGTCCA TCCATCTCGC
1081 TATCGTCAGA TTTCAATCTA TTGGCGTTGT TAAAAAACTA TGGTTATACT AACGGCAAAA
1141 ACGCTCTGAA ACTAGATCCT AATGAAGTCT TCAACGTGAC TTTTGACCGT TCAATGTTCA

(c) Hartwall/Kumpula, Jack Daniels
Everything can be coded

Describing a system...
0 1
Material Glass Plastic
Shape Round Square
Contents No pressure Under pressure
Cap Pull Screw
Leftmost column represents elements, 0 and 1
columns represent alleles of an element

Describing a system...
0 1
Material
Shape
Contents
Cap

...system as strings
Material 0 1
Shape 1 0
Contents 0 1
Cap 0 1
Bottles are thus coded in terms of their
elements and alleles (e.g. Frenken 2001, 2006)

The coding can have multiple levels - single 0/1
alleles can represent entire subsystems!
What can be coded?
In principle, everything that is definable by
information ≈ everything! Applications include
• Organization theory (Kauffman and Macready 1995; Westhoff et al
1996; Levinthal 1997; Marengo 1998; Baum 1999; Levinthal and Warglien 1999;
McKelvey 1999a, 1999b; Gavetti and Levinthal 2000; Ghemawat and Levinthal
2000; Marengo et al. 2000; Rivkin 2000; Dosi et al. 2001; Frenken 2001, 2006;
Morel and Ramanujam 1999)
• Political science (Schrodt 1994; Post and Johnson 2000)
• Scientometrics (Scharnhorst 1998)

Bottles coded with four elements of two alleles;
e.g. Frenken (2001, 2006), Kauffman (1993)
Strings for two bottles:
Jdaniels 0 1 0 0
ED 1 0 1 1

4-element, 2-allele design space (S) has 24 = 16
possibilities
1 0 0 0 0
2 0 0 0 1
3 0 0 1 0
4 0 0 1 1
5 0 1 0 0
6 0 1 0 1
7 0 1 1 0
8 0 1 1 1
9 1 0 0 0
10 1 0 0 1
11 1 0 1 0
12 1 0 1 1
13 1 1 0 0
14 1 1 0 1
15 1 1 1 0
16 1 1 1 1
All possible strings:

Simon (1969), Rosenberg (1969), Kauffman
(1993, 2000), Frenken et al. (1997, 1999)
Problem solving task:
Following Simon (1969), can be conceptualized
as searching for the right solution in a solution
(“design”) space.
New variations can be considered as “new
combinations” (cf. Schumpeter 1934).
Due to practical limitations - the size of design
space - this search is boundedly rational.
It thus requires heuristics.

Jdaniels is variation 5, ED is variation 12
1 0 0 0 0
2 0 0 0 1
3 0 0 1 0
4 0 0 1 1
5 0 1 0 0
6 0 1 0 1
7 0 1 1 0
8 0 1 1 1
9 1 0 0 0
10 1 0 0 1
11 1 0 1 0
12 1 0 1 1
13 1 1 0 0
14 1 1 0 1
15 1 1 1 0
16 1 1 1 1

e.g. Simon (1969), Wagner and Altenberg
(1995), Frenken (2001)
Interdependency matters
The system may have interdependent elements;
“mutating” one may have an effect on another’s
“fitness” for purpose.
This defines the system’s architecture.

Example:
(c) Hartwall/Kumpula
Pressurized contents

Example:
This is known as epistasy. Pressure has an
epistatic relationship to the shape of bottle.
Pressurized contents
Round shape

Kauffman (1993), Frenken (2001, 2006)
No epistasy
Material Shape Contents Cap
Material x
Shape x
Contents x
Cap x

Changing the contents of the bottle may mean
that the shape is no longer satisfactory
Epistatic relationship - 1
Material x
Shape x x
Contents x x
Cap x

Similarities to e.g. Quality Function
Deployment Matrix are evident.
Epistatic relationship - 2
Material x x x
Shape x x x
Contents x x x x
Cap x x

This creates difficulties
As epistasy increases, finding the optimal solution
becomes more and more difficult.
This can be represented using a fitness
landscape:
1) assign a fitness to each solution
2) order the solutions based on their differences
Theoretically, three different landscapes can be
identified:

No epistasy: single solution is the best one for
the task. Optimization is very easy.
!"
!#$"
!#%"
!#&"
!#'"
!#("
!#)"
!#*"
!#+"
!#,"
$"
!#,
!#+
!#*
!#)
!#(
!#'
!#&
!#%
!#$
!-!

!"#$%&
!"'$!"#&
!"($!"'&
!")$!"(&
!$!")&
Fully correlated
landscape: one
fitness peak.
“Hill-climbing”
optimization
strategy works
every time.

e.g. Kauffman (1993, 2000); Frenken (2001,
2006)
!"
!#$"
!#%"
!#&"
!#'"
!#("
!#)"
!#*"
!#+"
!#,"
$"
!#,
!#+
!#*
!#)
!#(
!#'
!#&
!#%
!#$
!-

Chaotic landscape: maximum epistasy. No
correlation between difference and fitness.
!"
!#$"
!#%"
!#&"
!#'"
!#("
!#)"
!#*"
!#+"
!#,"
$" !#,-$"
!#+-!#,
!#*-!#+
!#)-!#*
!#(-!#)
!#'-!#(
!#&-!#'
!#%-!#&
!#$-!#%
!-!#$"

2006)
Chaotic landscape:
no correlation
between difference
and fitness.
Optimization
impossible; random
search only viable
search strategy.
!"#$%&
!"'$!"#&
!"($!"'&
!")$!"(&
!$!")&

!"
!#$"
!#%"
!#&"
!#'"
!#("
!#)"
!#*"
!#+"
!#,"
$"

Intermediate epistasy: multiple options, local
and global optima.
!"
!#$"
!#%"
!#&"
!#'"
!#("
!#)"
!#*"
!#+"
!#,"

!"#$%&
!"'$!"#&
!"($!"'&
!")$!"(&
!$!")&
Roughly correlated
landscape: fitness
peaks and valleys.
Simple hill-climbing
may get stuck to
local optima;
random search too
inefficient. Random
mutation hill
climbing works!
2006); Dennett (1995)

!"
!#$"
!#%"
!#&"
!#'"
!#("
!#)"
!#*"
!#+"
!#,"
!#+-
!#*-
!#)-
!#(-
!#'-
!#&-
!#%-
!#$-
!-!#
Hello, real-life problem! This is what they tend
to look like.

For simulating social systems, see e.g. Gilbert &
Troitzsch (2005), Gilbert (2008)
The neat thing:
We can simulate various problem-solving
situations without having to know the details of
the situation;
We just need to estimate the epistasy and
solution space.

Arbitrarily large solution spaces
1 000000
2 000001
3 000010
4 000011
5 000100
6 000101
7 000110
8 000111
9 001000
10 001001
11 001010
12 001011
13 001100
14 001101
15 001110
16 001111
17 010000
18 010001
19 010010
20 010011
21 010100
22 010101
23 010110
24 010111
25 011000
26 011001
27 011010
28 011011
29 011100
30 011101
31 011110
32 011111
33 100000
34 100001
35 100010
36 100011
37 100100
38 100101
39 100110
40 100111
41 101000
42 101001
43 101010
44 101011
45 101100
46 101101
47 101110
48 101111
49 110000
50 110001
51 110010
52 110011
53 110100
54 110101
55 110110
56 110111
57 111000
58 111001
59 111010
60 111011
61 111100
62 111101
63 111110
64 111111

Copper smelting furnaces 1900-2000.
Design space size S = 3 981 312.
Back to real life!
R 0000 0000 0000 0000
R1930 0000 0001 0001 0110
Rdry 0000 1100 0000 0100
ROX 0001 0000 0000 0100
E1930 0010 0110 1101 0000
E1950 0010 0110 1100 0000
F1950 1120 1121 2121 0111
F1970 1121 1121 2121 0111
F1990 1131 1121 2121 0101
IF1950 0022 1120 2110 0101
IF1990 0022 1121 2111 3001
T 0121 0010 1010 0200
Tdry 0131 1110 2110 2200
C 0021 1121 2101 1101
IS 1121 0020 2101 0300
V 0022 0020 2110 0200
M 1121 1121 2110 0300
BF 1100 2000 0010 0200
PY 1120 2000 0010 0200

In 1900, blast (BF) and reverberatory (R)
furnaces are widespread.
BF 1100 2000 0010 0200 R 0000 0000 0000 0000

R 0000 0000 0000 0000BF 1100 2000 0010 0200
PY 1120 2000 0010 0200
In 1902, pyritic process (PY) is tested. It’s
autogenous, but limited by lack of suitable ores.

In the 1930s, electric furnaces (E1930) spread
(incl. Imatra), and reverberatories are improved.
BF 1100 2000 0010 0200
PY 1120 2000 0010 0200 R1930 0000 0001 0001 0110
R 0000 0000 0000 0000
E1930 0010 0110 1101 0000

Electricity shortage after WW2 focuses search
for alternative sources of smelting energy.
BF 1100 2000 0010 0200
PY 1120 2000 0010 0200 R1930 0000 0001 0001 0110
R 0000 0000 0000 0000
E1930 0010 0110 1101 0000

First flash furnace in 1949; integrates roasting,
uses concepts proven in fluid bed reactors.
BF 1100 2000 0010 0200
PY 1120 2000 0010 0200 R1930 0000 0001 0001 0110
R 0000 0000 0000 0000
F1950 1120 1121 2121 0111
E1930 0010 0110 1101 0000

Search & constraints:
The effect of constraints: focuses search on
improving attributes that are affected by scarcity.
Constraints act as a focusing device (Rosenberg
1969); focusing search to certain attributes.
This is known as function space search after
Bradshaw (1992) and Frenken (2001).

Measuring differences
When artefacts (or systems) are coded as strings,
measuring their differences is easy!
Hamming distance measures the differences
between two solutions:
Hamming (1950)

Hamming distance measures the number of
different elements in a “design.”
BF 1100 2000 0010 0200
PY 1120 2000 0010 0200 R1930 0000 0001 0001 0110
R 0000 0000 0000 0000
Hamming distance 1 Hamming distance 4
Measuring differences

Hamming distance can be used e.g. to measure
radicalness of innovation, creativity, novelty...
BF 1100 2000 0010 0200
PY 1120 2000 0010 0200
F1950 1120 1121 2121 0111
Hamming distance 11
Measuring innovation

Architectures can be decomposable or non-
decomposable (Frenken et al. 1999; Simon 1969)
Epistasy, revisited
n1 n2 n3 n4 n5 n6 n7 n8 n9
w1
w2
w3
w4
w5
w6
w7
w8
w9

Non-decomposable architecture: exhaustive
search requires 2N trials (512)
n1 n2 n3 n4 n5 n6 n7 n8 n9
w1 x x
w2 x x
w3 x x
w4 x x
w5 x x
w6 x x
w7 x x
w8 x x
w9 x x
Epistasy, revisited

Decomposable architecture requires 23 + 23 +
23 = 24 trials, which can be done in parallel!
Why e.g. distributed teams work
n1 n2 n3 n4 n5 n6 n7 n8 n9
w1 x x x
w2 x x x
w3 x x x
w4 x x x
w5 x x x
w6 x x x
w7 x x x
w8 x x x
w9 x x x

Most real life problems are nearly
decomposable - they can be solved in parallel
Near-decomposable:
n1 n2 n3 n4 n5 n6 n7 n8 n9
w1 x x x
w2 x x x
w3 x x x
w4 x x x x
w5 x x x
w6 x x x
w7 x x x x
w8 x x x
w9 x x x

Sukuraman et al. (2008)
Low epistasy technology
Figure 2: Revised Design7
Marketing
At the beginning of the spring semester, the Rowan University Business Department showed interest in
working with the project through the Entrepreneurial course. The engineering team met with a group o
business students to show them the drawings and explain the concept of the project after which they
decided to work on the marketing side of the project. At present, the business team is doing research on
areas the finished project can be marketed to. They are also presenting to a group of possible
Technologies developed for
use in the developing world
tend to be highly modular to
accommodate variety in
components, i.e. they are
near/fully decomposable,
such as this bicycle-powered
grain crusher

Scarcity causes search activity; focuses search
on certain dimension (energy, in this case)
The effects of scarcity (?)

Alternative energy solutions are found.

However, wood gas was not an integrated,
efficient system - flash smelting was.
+
-
Integration
(≈epistasy)

Other implications...
The theories used here are applicable to a large variety of
complex systems, and have interesting implications, e.g.
• Systems with moderate complexity have the highest
fitness (Schilling 2000)
• The more complex a problem is, the more solutions
differ from each other (Kauffman 1993)
• The speed of evolution is inversely related to the
complexity of system’s architecture (Kauffman 1993)
• Imitation strategies are less successful, the more
complex the system being imitated is (Rivkin 2000)
• S-curves are easy to explain: local/global optima!

Computing organizations
“Computational organization theory?” Based on
theory of computation, organizations as
computational systems (e.g. Saraceno and Barr
2002; also Wolfram 2002; Sipser 2006)
e.g. showing the limits of theory: the fastest way
to determine what an organization will do is to
let the organization do it; thus,
“shortcuts” (=theories) are limited, thus,
the predictive power of our theories is
inherently limited. ☐

One more thing...
The question “why we have so much variety” is
much less interesting than the question
why the lack of variety?

Current variety?
How many organizational forms there are in
≤ 50 person firms today?

Possible variety?
Exercise: how many different organizational forms
could a 1-50 person company have, if there are
just three levels in hierarchy?
e.g. possible positions are 1) top management,
2) middle management, 3) worker?

If you started at the Big Bang (≈ 13,8 billion years
ago), you’d have to go through
solutions per second to find all the possible
combinations
2 500 000

References (1)
Baum (1999). Whole-part coevolutionary competition in organizations. pp. 113-136
in Baum, McKelvey (eds.) Variations in Organization Science. In Honor of Donald
T. Campbell.
Bradshaw (1992). The airplane and the logic of invention. pp. 239-250 in Giere, R.
(ed.) Cognitive Models of Science.
Dennett (1995). Darwin's Dangerous Idea.
Dosi et al. (2001). Bridging contested terrain: linkin incentive-based and learning
perspectives on organizational evolution. Nelson-and-Winter Conference, Aalborg.
Frenken et al. (1997). Self-organization and the economics of technical change, in
Actes de 46me Congrés Annuel de l'Association Française de Science
Economique
Frenken et al. (1999a). Interdependecies, near-decomposability and adaptation.
pp. 145-165 in Brenner, T. (ed.) Computational Techniques for Modelling Learning
in Economics.
Frenken (2001). Understanding product innovation using complexity theory. PhD
thesis.
Frenken (2006). Innovation, Evolution, and Complexity Theory.
Gavetti and Levinthal (2000). Looking forward and looking backward: cognitive and
experiental search. Administrative Science Quarterly 45(1) pp. 113-137
Ghemawat and Levinthal (2000). Choice structures, business strategy and
performance: a generalized NK-simulation approach. Working paper 2000-05,
Wharton School
Gilbert (2008). Agent-based models.
Gilbert and Troitzsch (2005). Simulation for Social Scientists.
Hamming (1950).
Kauffman (1993). The Origins of Order.
Kauffman (2000). Investigations.
Kauffman and Macready (1995). Technological evolution and adaptive
organizations. Complexity 1, pp. 26-43
King (2007). The Evolution of Technology for Extractive
Metallurgy over the Last 50 Years—Is the Best Yet to Come? Journal of Metals Vol.
59, No.2, pp. 21-27
Levinthal (1997). Adaptation on rugged landscapes. Management Science 43, pp.
934-950
Levinthal and Warglien (1999). Landscape design: Designing for local action in
complex worlds. Organization Science 10, pp. 342-357
Marengo (1998). Interdependencies and division of labour in problem-solving
technologies. Seventh International Schumpeter Society Conference, Vienna.
Marengo et al. (2000). The structure of problem-solving knowledge and the
structure of organizations. Industrial and Corporate Change 9, pp. 757-788
McKelvey (1999a) Avoiding complexity catastrophe in coevolutionary pockets:
strategies for rugged landscapes. Organization Science 10, pp. 294-321.
McKelvey (1999b). Self-organization, complexity catastrophe, and microstate
models at the edge of chaos, pp. 279-307 in Baum, McKelvey (eds.) Variations in
Organization Science. In Honor of Donald T. Campbell.

References (2)
Morel and Ramanujam (1999). Through the looking glass of complexity: the
dynamics of organizations as adaptive and evolving systems. Organization
Science 10, pp. 278-293.
Post and Johnson (2000). "Chaos prevailing on every continent": towads a
newtheory of decentralized decision-making in complex systems. Chicago-Kent
Law Review.
Rivkin (2000). Imitation of complex strategies. Management Science 46, pp.
824-844.
Rosenberg (1969). The direction of technical change: inducement mechanisms
and focussing devices. Economic Development and Cultural Change 18, pp 1-24.
Scharnhorst (1998). Citation networks, science landscapes and evolutioniary
strategies. Scientometrics 43, pp. 95-106.
Saraceno and Barr (2002). A computational theory of the firm. Journal of Economic
behavior and Organization.
Schilling (2000). Toward a general modular systems theory and its application to
interfirm product modularity. Academy of Management Review 25(2) pp. 312-334
Schrodt (1994). A landscape model of rule-based co-adaptation in international
behavior.
Schumpeter (1934). The Theory of Economic Development.
Simon (1969). The Sciences of the Artificial.
Sipser (2006). Introduction to the Theory of Computation. Second Edition.
Sukuraman et al. (2008). Teaching Engineering Design with a focus on the
developing world. Proceedings of the ASEE 2008.
Wagner and Altenberg (1996). Perspective: Complex adaptations and the evolution
of evolvability. Evolution 50(3), pp. 967-976
Westhoff et al. (1996). Complexity, Organization, and Stuart Kauffman's The
Origins of Order. Journal of Economic Behavior and Organization 29, pp. 1-25
Wolfram (2002). A New Kind of Science.

Explaining the evolution of technologies, firms, and industries: how complexity and constraints can be simulated

More Related Content

Similar to Explaining the evolution of technologies, firms, and industries: how complexity and constraints can be simulated (20)

More from J. M. Korhonen (14)

Recently uploaded (20)

Explaining the evolution of technologies, firms, and industries: how complexity and constraints can be simulated