Embodied Artificial Evolution:

the Next BIG Thing?Gusz Eiben

VU University Amsterdam

Popular version in the

TED talk: "Tech Kangaroos: Evolution at Work”Search for “Eiben” on youtube

Past, Present, Future 19th century:

evolution is a theory, an explanatory concept that helps us understand things

20th century:

evolution is a tool, an active, creative concept that helps us produce things (solutions for problems) in digital spaces

21st century:

evolution is a tool, an active, creative concept that helps us produce things (solutions for problems) in physical spaces

WETWARE HARDWARE

EVOLUTIONEVOLUTION

Embodied Artificial Evolution

SOFTWARE

Biosphere Evolutionary Computing

Evolutionary Computing lessons+• Evolution can solve problems we don’t fully understand and cannot clearly

specify

• Evolution can cope with changing situations

• Evolution can come up with original, unexpected solutions (that can be reverse-engineered)

–• Designing good EAs can be difficult (representation, operators, parameters)

• No good theory

• Could take too long (can change with new hardware)

Evolutionary computing and embodiment

• Evolution in digital space, result in digital space e.g., time tables, consumer models, robot controllers

EVOLUTIONARY OPTIMIZATION

• Evolution in digital space, result in physical space e.g., jet nozzle, satellite boom, P Bentley’s coffee table EVOLUTIONARY DESIGN

• Evolution in physical space, result in physical space EMBODIED ARTIFICIAL EVOLUTION

Embodied Artificial EvolutionOld: Evolution of code (blueprint)

New: Evolution things

1.Takes place in physical objects

2.Reproduction = real birth = new object made, survivor selection = real death = object deleted

3.Fitness is environmental or hybrid environmental and task-based

4.Reproduction and survivor selection are decoupled, autonomous decisions, distributed management

Application examples

• Mars explorers, artificial pets, robot companions, …

• Functional organisms, medical nano-robots, personal virus scanners, …

• Personal replicators (networked, evolutionary, Internet of Things)

Active vs. passive entities • Two kinds of artefatcs

• Active: needs controller to govern its activities (e.g., robot)

• Passive: needs no controller (e.g., smartphone cover)

• In biological terms

• Active: needs body + mind

• Passive: only body

Different approaches, under different umbrellas

Physical medium, embodiment ?

Bottom-up, chemistry Top-down, biology Mechatronics, plastics

A new game for “passive” stuff

Manufacturing

Testing (Re)design

Embryonic development

EvaluationSelection

Reproduction

propagule blueprint

A new game for “active” stuff

Robots

• very small to big

• programmable

• controllable

• sensors/actuators/controllers

Cells

• very small

•self-replicating

• self-repairing

• self- …

WETWARE HARDWARE

Sweet spot

Grand challenges• Body type

• Reproduction - how to start

• Kill switch - how to stop

• Evolvability and evolution speed

• Process control and methodology

• Co-evolution of body & mind

• Lifetime learning (self-awareness, mind must match body)

Closing remarks• Evolution in the real world is different – matter matters

• Computation in this new paradigm is different

– Non-von-Neumannian architecture for information processing/computing

– Redefines ICT, e.g., “program”, “programming”, software engineering, testing, validation, ect.

• Big questions are implied:

– Scientific (fundamental issues regarding evolution, notion of Life)

– Technological (feasibility)

– Applicability (remember IBM in the 1940ies)

– Desirability (ethics, think of GMOs)

• Different communities, unifying vision / umbrella needed

• Next BIG thing ?