Getting Creative With Evolution Peter J. Bentley University College London.

28
Getting Creative With Evolution Peter J. Bentley University College London
  • date post

    22-Dec-2015
  • Category

    Documents

  • view

    218
  • download

    1

Transcript of Getting Creative With Evolution Peter J. Bentley University College London.

Page 1: Getting Creative With Evolution Peter J. Bentley University College London.

Getting Creative With Evolution

Peter J. BentleyUniversity College London

Page 2: Getting Creative With Evolution Peter J. Bentley University College London.

Evolutionary Computation

• Allows us to harness the power of evolution for non-living designs.

• We can use evolution to generate music, art, and design.

• It endows our computers with skills which suspiciously resemble creativity.

• Even for such mundane tasks as evolving the design for a coffee table, evolution shows surprising originality…

Page 3: Getting Creative With Evolution Peter J. Bentley University College London.
Page 4: Getting Creative With Evolution Peter J. Bentley University College London.
Page 5: Getting Creative With Evolution Peter J. Bentley University College London.

Explorative Evolution

• Recently, views of evolution have begun to change.

• Although evolution will behave like an optimiser if properly constrained, many now believe that its natural behaviour is more like exploration.

• And we are starting to use evolution for exploration instead of optimisation.

Page 6: Getting Creative With Evolution Peter J. Bentley University College London.

Explorative Evolution

• Differences:

• Optimisation - knowledge-rich representations, fixed fitness functions, global optimum required in minimum time.

• Exploration - knowledge-lean representations, possibly variable fitness functions, new and interesting solutions required (but optima may be unavailable)

Page 7: Getting Creative With Evolution Peter J. Bentley University College London.

Exp

lora

tive

Evo

luti

on

A new urban space in Milan, the “Caravanserraglio”. Evolved using Argenia by Celestino Soddu(From Creative Evolutionary Systems)

Page 8: Getting Creative With Evolution Peter J. Bentley University College London.
Page 9: Getting Creative With Evolution Peter J. Bentley University College London.
Page 10: Getting Creative With Evolution Peter J. Bentley University College London.

Explorative Evolution

• Recently, views of evolution have begun to change.

• Although evolution will behave like an optimiser if properly constrained, many now believe that its natural behaviour is more like exploration.

• And we are starting to use evolution for exploration instead of optimisation.

Page 11: Getting Creative With Evolution Peter J. Bentley University College London.

Evolved 60dB amplifier, by John Koza, Forrest Bennett, David Andre, Martin Keane

(From Evolutionary Design by Computers)

Exp

lora

tive

Evo

luti

on

Page 12: Getting Creative With Evolution Peter J. Bentley University College London.

Explorative Evolution

(From Creative Evolutionary Systems)

Page 13: Getting Creative With Evolution Peter J. Bentley University College London.

Evolved competing creatures by Karl Sims

(From Evolutionary Design by Computers)E

xplo

rati

ve E

volu

tion

Page 14: Getting Creative With Evolution Peter J. Bentley University College London.

“Pearly Gates”

© 1999 Steven Rooke

(From Creative Evolutionary Systems)

Exp

lora

tive

Evo

luti

on

Page 15: Getting Creative With Evolution Peter J. Bentley University College London.

Explorative Evolution

Produced by GenJam (Al Biles)

(From Creative Evolutionary Systems)

Page 16: Getting Creative With Evolution Peter J. Bentley University College London.

How is this creativity achieved?

Through component-based representations

Page 17: Getting Creative With Evolution Peter J. Bentley University College London.

Component-based representations

Instead of optimising selected elements of a given solution, we allow evolution to build new solutions from scratch, using component-based representations

Page 18: Getting Creative With Evolution Peter J. Bentley University College London.

Component-based representations

My work used primitive shapes to construct novel designs

Page 19: Getting Creative With Evolution Peter J. Bentley University College London.

Component-based representations

sin() pdiv() pminus() mandelstalk() pqj4da2013() pln() M_PI 0.022307 x y

Steven Rooke uses GP function and terminals to generate his art

Page 20: Getting Creative With Evolution Peter J. Bentley University College London.

Component-based representations

John Gero used ‘wall fragments’ to generate house floor plans

Page 21: Getting Creative With Evolution Peter J. Bentley University College London.

How is this creativity achieved?

• When evolution is told to build solutions from components, it becomes creative.

• Only those approaches that use component-based representations provide sufficient freedom.

• Evolution now explores new ways of putting components together to construct innovative solutions.

Page 22: Getting Creative With Evolution Peter J. Bentley University College London.

Creative Computers - What does this mean?

• We are now beginning to understand the benefits and pitfalls of creative evolutionary computation.

• Evolution can find solutions that disregard our conventions and theories.

• Efficient new designs have been evolved, and unusual art.

Page 23: Getting Creative With Evolution Peter J. Bentley University College London.

Creative Computers - What does this mean?

• Whilst we know that the solutions do perform better, sometimes we do not know how they work.

• Circuits, neural networks, computer programs - all evolved to perform brilliantly, but their functioning is bizarre and difficult to understand.

Page 24: Getting Creative With Evolution Peter J. Bentley University College London.

Creative Computers - What does this mean?

• Principle extraction is one way of overcoming the fears.

• Rather than use directly the wacky evolved designs, we can learn new design techniques and then apply them ourselves.

• Some even predict that new mathematical understandings will arise from the electronic circuits that defy current theories.

Page 25: Getting Creative With Evolution Peter J. Bentley University College London.

Creative Computers - What does this mean?

• Other issues begin to arise when we do use computers as creative machines.

• If the artist uses evolution to generate the art, who created the art - the artist or the computer?

• Most evolutionary artists claim to be the originator, but if the piece is more the product of the software than the person, is this appropriate?

Page 26: Getting Creative With Evolution Peter J. Bentley University College London.

Creative Computers - What does this mean?

• Legal issues arise when computers are used as composition machines.

• As we have found from our evolutionary music work, the law only recognises people as capable of music composition.

• When using a computer to evolve novel music, someone must be nominated to be the composer…

• J13 CD tracks: 10-13, (2), 4, 6, (7)

Page 27: Getting Creative With Evolution Peter J. Bentley University College London.

Creative Swarms

• Evolution is not the only way to be creative.

• The new field of swarm intelligence (related to evolutionary computation) also shows the same features.

• Our work on swarm music has provided an unusual and exciting way to perform life music improvisation.

Page 28: Getting Creative With Evolution Peter J. Bentley University College London.

Conclusions

• Creative computers allow more innovative ideas to be explored in a shorter time.

• Evolution is enabling our technology and arts to develop in surprising and exciting new ways.