Lecture 5 Flashcards
two types of Connectionism
- scientists who build brain-like networks
- philosophers who argue that connectionist model provides insight into how the mind works
neurons
send information to other neurons and other types of cells. the messages the neurons send consist of electrical pulses and neurotransmitters
structure of a neuron
neurons all have a cell body (soma) and two extensions: dendrites and axons
- dendrites receive information from other neurons and the location at which they receive this information is the synapses
artificial neurons (unit)
called a unit, it receives input and sends out output. an artificial neural network consists of an input layer, an output layer, and a hidden layer in between
how does artificial neural network process information?
it is processed in a parallel way, and not in a serial manner, as it was in functionalism
face recognition
Garrison Cottrell made a connectionist network that could scan pictures of faces. it consisted of an input layer, hidden units, and output units
- snapshot reasoning
backpropagation
a gradient estimation method used to train neural networks. the step-by-step way this process is conducted is called gradient descent learning, where the computer fine-tunes the connectionist network
is connectionism better than the cognitivist model?
it should be because connectionism is much closer to the way our brains are structured. This makes connectionism more biologically realistic than cognitivism.
issues of functionalism
- it took up to 8 hours for Shakey’s computer to process information
- information was only processed in a serial fashion and it was intolerant to damage
Does connectionism also hold up with the Chinese room problem?
in a connectionist network, representations are distributed over the system. in the face recognition, the same units and weights are used to represent the faces of Maddie, Janet, etc. this is a very economical way of representing things because you don’t need a large system to store information
structure and semantics
Searle argued that we can’t get semantics from mere syntax. we can’t get meaning from mere rules
structure and semantics - cognitivist view vs connectionist view
- in the cognitivist view, two pictures take up two separate parts of the memory
- in the connectionist view, this is not the case because representations are distributed over the system and there is no need to use a new part of the memory
this means that syntax and semantics are closely related in the connectionist networks, but unrelated in the cognitivist model
How does connectionism imply that propositional attitudes don’t exist?
in a connectionist network in which representations are distributed over the system, it’s impossible to learn something new without altering the network. If the network contains the knowledge that dogs have paws, this knowledge is distributed over the network and represented by the weight of the connections. If the same network has to learn that cats have paws, then the weights of the networks would need to be altered, also altering the way in which the information that dogs have paws is represented
problems for connectionisms
an issue for connectionism that is related to the idea that PA’s are real and discrete is that thought is systematic. this means that thought is rule-based and uses discrete concepts or words, just like language.
brain centric
Functionalism is a brain-centric view of the conscious mind. this is also true for connectionism
