The Computational Metaphor for Mind

What are some common problems with theories?

Ill-specified + too vague → difficult to apply + no predictive power (explains a large range of possibilities)

What is a model?

Logical representation of a system in a formal language that describes + is simpler than the actual system (i.e. formal, math representation) 

Why did Turing propose his machine?

Conceptually allows one to think about what it would take to solve a problem

What are the 4 components of thee Turing machine?

• Machine table

• Machine state

• A tape of unbound length

• A read-write head

What is a machine table?

A list of instructions of the form, “if in state x, and input is y, then do z” → program or algorithm 

What is a machine state?

Current state of the machine represented as a number → input/output

What is a tape of unbound length?

A tape divided into discreet parts, each part containing or has the ability to contain a 0 or a 1

What is a read-write head?

A device that can read the current slot on the tape, move the tape forwards or backwards + erase or write a 0 or a 1 to a slot on the tape 

What can be done by a Turing machine?

Anything that can be done by a finite set of precise steps + isn’t mathematically random (not compressible/can’t be shortened e.g. pi) → anything that can be reasoned out by formal rules

What did Alan Turing prove the existence of?

A universal Turing machine

What is a universal Turing machine?

Can imitate any Turing machine -> is a single machine that can compute anything that is computable/doable

What are some imitations of a universal Turing machine?

Modern digital computers

Why can the mind be imitated by a universal Turing machine?

It only carries out actions that can be carried out in principle (tautological)

How can the relation between the mind and a universal Turing machine be used?

We can do what the human mind does on a UTM:

• Figure out what functions the mind is computing → can simulate them with a computer program 

What is computational theory?

We can understand the human mind if we can understand the computable problems the mind is solving → mind is best conceived of as a computational system 

What is a computational system?

System that operates by performing computations 

What is computation?

Any type of info processing that can be represented mathematically

What is an information processing system?

System that takes in info in some state + processes/transforms it into another state by an algorithmic process

What is an algorithm?

Formally describable (can be written in the language of maths) procedure/set of instructions for performing some operation in a finite number of steps (primarily for the purposes of computation) 

How can the mind be understood computationally?

1. Represent + process info all the time 

2. Algorithm is a specification for representing + manipulating info 

3. An algorithmic account of mind will specify the means by which the mind represents info + the processes by which it manipulates these representations 

What could the mind be if not computational?

• Input-output system (info via sense → behaviour)

• Info-processing system (input-output is accomplished by an algorithm)

What must be the case for the mind to be an information processing system? What happens if this isn’t the case?

• Must be algorithmic by definition if the input is transformed into output via a regular set of steps

• If the mind doesn't do this → must be an irregular (mathematically random) set of steps → can never be understood in a systematic way (can’t be scientifically studied)

Why (or why not) might the mind be understood computationally?

• For science to work it must be pointed at something that works in an understandable (i.e., systematic) way 

• If the mind is amenable to scientific study it must follows a systematic + understandable set of processes 

• If these processes are systematic they can be described via an algorithm 

• Therefore, mind is necessarily computational 

How can the mind be deductively concluded to be modelled as an algorithm on a universal Turing machine?

P1) The mind is computational

P2) Is computational → is describable by an algorithm 

P3) All algorithms can be carried out by a some Turing machine 

P4) A UTM can simulate any Turing machine 

Conclusion: can model the mind as an algorithm on a UTM 

What does it mean if the mind isn’t computational?

• Is random

• Doesn’t have a computable solution

What happens if the mind is random?

Will never know it is random:

• Random → the shortest description of the mind is the actual mind itself

• In order to understand that a pattern is random, you have to follow a procedure (series of specific steps in a specific order) → if we are random, we can’t follow a series of steps (by definition)

• We’ll never be able to look at + understand a definition of the mind in a systematic way (beyond our cognitive resources)

What happens if the mind doesn’t have a computational solution?

We’ll never know → no way to prove that an algorithm does not terminate

What is the conclusion if we assume the mind is computational?

We’ll never know if it isn’t

What is a computational model?

A model implemented in a computer program, models that specify the process, the representations in some form of programming language/computer programming 

What is needed for a theory to be modelled by a computer?

Must explicitly specify all aspects of the theory (knowledge, capacity, perceptual capabilities etc)

Why is it necessary to explicitly specify a theory?

• Vagary leads to interpretation 

• To know what a theory predicts you have to go through the mental process of running the theory yourself (can lead to different conclusions)

  • When I input x, y or z into my theory, what does it actually do 

  • Must figure out what your theory predicts independently of you, the theoriser

What is Turing’s test of machine intelligence?

3 rooms: computer, person, interrogator 

• Interrogator’s job is to figure out which is the real person 

• Blind test: Based on behaviour, not appearance 

How is Turing’s test of machine intelligence subjective?

Different testers might be fooled by different machines 

What was found by Turing’s test of machine intelligence?

Machine intelligence vs machine is as good as interacting with judges as the person in the other room

What are some further problems with Turing’s test of machine intelligence?

• There are other aspects to intelligence other than fooling a human (doesn't mean it can do everything the human mind can do)

• Only a performance, not competence (what the machine does vs is capable of doing)

• Is a behaviourist test + intelligence of the machine depends on the intelligence of the interrogator

What is the Chinese Room argument (John Searle)?

• An English-only-speaker in a room 

• In the room, they have: 

  • An “in” slot: Squiggles on paper come in 

  • An “out” slot: Squiggles go out 

  • A blackboard on which they write numbers 

  • A set of instructions, in English, of the form… 

    • If the number X is on the blackboard, and a slip of paper with squiggle W comes in the “in” slot, then draw the squiggle Y on a piece of paper, push that piece of paper through the “out” slot, and draw the number Z on the blackboard (erasing S) 

• The person follows these directions to the letter, but they are not told why 

• Unbeknownst to the person in the room: 

  • The room is located in China 

  • The “squiggles” coming in are questions written in Mandarin 

  • The “squiggles” he is pushing out are answers written in Mandarin 

  • The answers he is giving are really smart, and give the impression that there is an intelligent Chinese person in the room 

How does Searle’s argument challenge the computational theory of mind? What are the weaknesses of this argument?

• Chose the conditions to mimic as closely as possible the “life” of a computer (i.e., Turing Machine) passing the Turing Test 

• Neither the person, nor the person-room combination have the equivalent of human intelligence, but they still pass the Turing Test -> passing the Turing Test is not sufficient to demonstrate human-like intelligence 

• Computers operate like the person in the room -> computers cannot have human-like intelligence (computer's rules representations lack intentionality) 

  • BUT problem with consciousness AKA awareness (responding to stimuli) -> you cannot know whether any other entity is aware of anything -> can only assume (what Searle has done -> circular argument) 

  • Consciousness has no observable consequences -> can't be operationalised (cannot know whether it is present in any given entity) 

What are the 2 types of models in psychology?

• Mathematical models

• Process models

What is an example of a mathematical model?

Rescola-Wagner → many models of personality specify which traits interact to produce things like preferences or behaviours 

Which type of model is the most common in cognitive and behavioural neuroscience?

Process models

What are 2 examples of process models?

• Symbolic

• Connectionist/neural-network

What is the function of symbolic models?

• Rep knowledge as symbolic data structures → specify some format for repping knowledge + some content that can fill this format 

• Manipulate data structures with variablised rules (specified algorithm) 

  • Data structures 

    • Basic/irreducible (atomic elements) 

    • Rules for composing these elements into more complex structures (e.g. natural language, math) 

  • Processes → symbolic operations on the data structures (application of symbolic rules) 

What is cognition like (in general) in symbolic models?

A computer program

• Mental reps = data structures

• Procedures = functions/methods that operate on data structures

What is a production system?

A phototypical symbolic model

What are the components of a production system?

• Data base of known facts (represented as data structures) 

• A set of inference rules (if x, then y) 

  • If the rule fits a fact in the data base, then the rule can be fired while preserving the variable bindings 

  • Knowledge + rules = new knowledge (transitive inference) 

• Executive control structure -> decides which rules to fire (determines the emerging knowledge) 

What are the operations/rules of a production system?

• Current state → current contents of the data base (known facts about a system) 

• State space → set of all possible states 

• Goal state → the state you want the data base to be in 

• State transition → moving from one state to another (firing a rule + changing the data base) 

• Search → algorithm for traversing the state space + finding the best path from moving from current state to goal state (means-end analysis) 

What are the advantages of a production system?

Computational power -> you can define variablised + universally quantified rules 

What is the function of a connectionist model?

Rep knowledge as patterns of activation in some set of units (nodes in a network)

How is processing carried out in a connectionist model?

Passing activation between units (nodes) over weighted connections