LECTURE 5

Neural Networks - Training the system

Nouvelle AI

Non-symbolic, non-representational modelling

PRESSURES FOR A DIFFERENT KIND OF MODELLING

Dissatisfaction with fragmented approach:
language processing, vision, planning, learning
Is this the right subdivision? How fit processes together?

Failure to address the ``situatedness'' of many human capabilities

Failure to model those capabilities we share in common with other animals

THE ISSUE OF REPRESENTATION

Modelling at much lower level -- below ``representation''

simple creatures

links between sensors and effectors

object avoidance, predator/prey behaviour, path finding

accumulation of many simple behaviour patterns can give rise to overall complex behaviour

Rodney A. Brooks (1991) Intelligence Without Reason, Proceedings of the Twelfth International Conference on Artificial Intelligence Sydney, Australia, pages 569-595

NOUVELLE AI

Start by modelling simplest creatures and work upwards.

Pay attention to the interaction between creature and its environment

Take account of evolutionary pressures and mechanisms in understanding how behaviours (and creatures) got to be how they are

Note that complex (intelligent?) behaviour can be produced by interactions of simple behaviours without need of internal ``representation''

Honeycomb, termite nest

Robotics rather than mental tasks

BRAITENBERG VEHICLES - EXAMPLE 1

Single sensor and single effector (wheel)

More the sensor is stimulated the faster the wheel goes

BRAITENBERG VEHICLES - EXAMPLE 2

Two sensors directly connected to two wheels

Tends to turn away from sources of stimulation

BRAITENBERG VEHICLES - EXAMPLE 3

Two sensors cross connected to two wheels

Tends to turn toward from sources of stimulation

BRAITENBERG VEHICLES - EXAMPLE 4

Multiple kinds of sensors (e.g. light, temperature, proximity)

Multiple kinds of connections

Complex (surprising) interacting behaviour patterns

Real vs. simulated robotics

LECTURE 5 CONCLUSIONS

Neural Networks - Training the system

Nouvelle AI

Non-symbolic, non-representational modelling