Philosophy and Psychology L3: Computationalism, Chinese room experiment, Turing test, Connectionism

Computationalism = to think is to perform computations based on syntactic properties of a symbolic language

Mashines and computers:

- (can/ cannot) manipulate symbols: based on syntactic properties and a set of instructions

- (can/ cannot) understand semantics

- (can/cannot) respect semantics: by programming

When can we state that computers can 'think'?

- thinking = causal and relational

1. all metalls expand when heated

2. iron is a metall

3. iron will expand when heated

A person in a room produces answers by manipulating symbols. It receives input from outside in chinese, and although not understanding the message, the person gives output in chinese due to symbolic manipulation. The person behaves like a computer, but the symbols remain uninterpreted. This is the Chinese room experiment.

Proponents of computationalism claim that the computer in the Chinese Room experiment:

- (does/ does not) understands the input

- this (does/ does not) explain human understanding

Searle argues to computationalism that the computer in the Chinese Room experiment:

- (does/ does not) understands the input

- this (does/ does not) explain human understanding, only that the computer is functioning

"To think is to perform computations based on syntactic properties of a symbolic language"

A) computationalism

B) Searle

"Thinking is not computation, but a requirement. Human beings understand semantics and computers do not"

A) computationalism

B) Searle

A human questioner asks a series of questions to a computer and another human. If she cannot tell the difference between human and computer then the computer can think. This is the Turing Test.

Connectionism is (top-down/bottom-up), whereas computationalism is (top-down/bottom-up).

starting point is a simple computational unit

A) Connectionism: bottom-up

B) computationalism

starting point is a higher-order cognition and expands to it's implementation

A) Connectionism

B) computationalism: top-down

Connectionism:

- (rejects/accepts) the requirement of symbolic language: starting point is a simple computational unit

-  (focuses on/ ignores) artificial neural networks: complex behaviors require many interconnected neurons

Connectionism:

- neurons (do/ do not) compute a simple transfer function, which determines the neurons output given some input

- the (combination/ weight) of the connection determines the effect on the ouput of a previous neuron on the receiving neuron

Connectionism:

- the combinations of connections, weights and transfer determine the (behavior/rules) of the network

- output is determined by the activty of many neurons, there (are/ are no) explicit higher order rules

Connectionism:

- weights are modified by (backward/inward) propagation of error

- networks (do/ do not) learn to generalize from input to categorize novel input

Connectionism

This is an example of (unsupervised /supervised) learning:

an input (dog picture) is presented to a network of random weights

- the network produces some input 'szdbakss' which is composed to the right answer

- the errors are 'back propagated' through the network, updating weights to minimize error

teaching a network to distinguish between generated rocks and mines

A) supervised learning

B) unsupervised learning

teaching a network to pronounce English words 97.5% accuracy compared to DECtalk 78% accuracy

A) supervised learning

B) unsupervised learning

Connectionism:

This is (unsupervised/ supervised learning)

- relies on statistics of the input to organize the output space into categories

- the network learns to pick up the most salient patterns in the input

An example is K-means clustering

Connectionist networks are

- (good/ bad) at handling noisy input

-after damage they (do/do not) function

- (good/ bad) at perceptual recognition/classification

= properties of connectionist networks

Connectionism:

- explain how tasks are done (yes/no)

- investigate what tasks are done (yes/no)

- connectionism can explain higher order human cognition as abstract symbolic thinking (yes/no)

Connectionism is good at A)perceptual recognition and classification, whereas computationalism is good at B) calculation and memorizing data sets.

A)perceptual recognition and classification

B) calculation and memorizing data sets