Technology and the Evolution of Complexity W. Brian Arthur

Technology and
the Evolution of Complexity
Human Complex Systems, Feb 23, 2007
W. Brian Arthur
External Professor, Santa Fe Institute
How does the collective of technology
evolve over time?
And what does this say about the evolution of complexity?
Consider …
1. Novel technologies are constructed from existing
2. … and offer themselves as components—building
blocks for the construction of further technologies
Technology Builds Itself from Itself
Technology is autopoietic or self-creating:
– New elements build from existing ones
– Complication builds from simplicity
W. F. Ogburn’s Claim (1922)
“It would seem that the larger the equipmen t of material cultu re, the greater
the number of inv entions. … The more there is to inven t with , the greater
will be the number of inv entions . When the existing material cultu re is
small, embracing a stone techniqu e and a knowledge of skin s and some
woodwork, the number of inv entions is more limi ted than when the culture
consists of a knowledge of a variety of metals and chemicals and the use of
steam, electricity , and various mechanical princip les such as the screw, the
wheel, the lever, the piston, belts , pulley s, etc. The street car could not have
been inv ented from the material cultu re existing at the last glacial period.
The discovery of the power of steam and the mechanical technology existing
at the tim e made possible a large number of inv entions.” -- Social Change,
Exponential buildup of technology?
• Dubious: Many combinatorics possible
– (Simplest gives 2N – N – 1 possible technologies)
• Diminishing returns to novelty
or to demand possible
An Experiment at FXPAL
W. Brian Arthur and Wolfgang Polak
- Create an artificial world in which the technologies are logic circuits
- Give this world “logic needs” to be potentially fulfilled
- Allow the system to create technologies to fulfill these
by combining previous technologies
Rules of the Experiment
1. Start from one primitive (NAND gate)
2. Make circuits by random combination of existing elements
3. Check to see if any needs (target logic purposes) are fulfilled
4. If so, these novel circuits become new building blocks
5. Better versions (simpler ones) replace previous versions
Tech 20
1-bit adder
2-bit adder
Tech 33: And-3
3-bit adder
4-bit adder
We Find …
1. Quite complicated circuits evolve
– An 8-bit adder (16 inputs, 9 outputs) is one
of 10177,554 possible circuits
We Find …
2. Complicated circuits require intermediate steps
– … and intermediate needs
(Cf. R. Lenski et al. Evol. Origin of Complex Features)
We Find …
3. Buildup is history dependent
– New technologies build on what is
Figure 5: Implication, being invented before negation in this example, is used mo re heavily. Usage
declines over time as better technologies are invented.
We Find …
4. Sudden appearance of key circuits (enabling technologies)
then quick use of these
– Full
adder appears after 32,000 steps; 2,3,4-bit adders
quickly after that
Gales of Destruction
When a technology disappears (is replaced) a technology it
used may have no further use. That tech then disappears
… etc.
Q. Are these gales “sand-pile avalanches”?
–I.e. is the system at self-organized criticality?
Avalanches of destruction follow a power law
Summary: a different view
• Technology is a self-constructing “chemistry”
(-cf. genetic regulatory networks)
• Technology bootstraps in numbers and complication
• Innovation: places of generative interaction important
– Where a chemistry is generated, harbored, nurtured
The Process as Evolutionary Algorithm
• Our algorithm creates a library of functionalities
by constructing new objects from previously
existing ones
• Differs from genetic algorithm or GP in several