Cognitive Science

What are the two components of a representation?

*vehicle* - a physical (or formal) object *content* - what the vehicle represents or is about; the information it carries

How did William James define Psychology and Psychological phenomena, and what method did he consider central to its studies and why?

- "the Science of Mental Life, both of its phenomena and of their conditions." - "The phenomena are such things as we call feelings, desires, cognitions, reasonings, decisions, and the like". - "Introspective Observation is what we have to rely on first and foremost and always...Because minds are a private theater of one, only way to access is through owner"

Describe David Hume's principle of uniformity of nature, and give an example of this in statistics

- *The future will resemble the past.* - Or, more precisely, if X percent of the observed objects A have property P, then X percent of the unobserved As have P. - linear regression

Name and define the 3 components of logic

- *language*: a recursively defined collection of strings on a fixed alphabet - *deductive system*: rules for deriving conclusions on the basis of premises - *model-theoretic semantics*: provides meaning (interpretation, truth-conditions) for at least part of the language

What is the difference b/t lexical and compositional semantics? What are the 3 semantic relationships?

- *lexical*: meaning of words; synonyms and antonyms - *compositional*: meaning of phrases, sentences, etc. - semantic relationships: 1. paraphrase (two differently structured sentences that are very close in meaning) 2. entailment (truth of one sentence implies the truth of another sentence) 3. reference and sense (eg. Michael Phelps is most decorated athlete)

What are the differences b/t Psychological and Philosophical Behaviorism?

- *psychological*: * An empirical research program * Goal is to explain behavior in terms of stimuli and the history of reinforcement and punishment * Tries to eliminate mental vocabulary altogether - *philosophical*: * mental terms express behavioral dispositions and/or neurophysiological states (behavioral translation) * Primarily concerned with the semantics of our common mentalistic vocabulary * The goal is to translate mental terms into terms that speak only of behaviors and propensities

What is Marr's tri-level hypothesis?

- 3 levels at which any machine carrying out an info-proc task must be understood: 1. *computational theory* - what is goal of computation, why is it approp., and what is the logic of the strategy by which it can be carried out? 2. *representation and algorithm* - how can comp. theory be implemented? what is rep. for input & output, what is algorithm for the transformation? 3. *hardware implementation* - how can rep. and algo. be realized physically?

What does monism say? What are the three main types?

- All is one, there are no fundamental divisions 1. Idealism (Berkeley) 2. Neutral monism (Spinoza's Dual-Aspect Theory) 3. Materialism (or, more specifically, physicalism)

What is the simple Hebbian Learning rule and its efficacy?

- Change in the weight between sender xi and receiver yj = Small learning rate parameter x Presynaptic activation x Postsynaptic activation - In this simple form, the rule does not work because the weights get infinitely large.

What are the 4 main functions of the central executive?

- Controlled updating of short-term memory - Setting goals & planning - Task switching - Stimulus attention & response inhibition

How can the Divide and Conquer method be used to solve the infinite regress problem of the PFC?

- Decompose the big and complex function (e.g., cognitive control) into multiple simpler subfunctions. - Each subfunction is carried out by a separate circuit (e.g., brain area). - An entire interconnected network is needed to solve the big function. - Infinite regress is avoided because one big task is divided into many simpler subtasks, which can be divided in turn into a multitude of even simpler sub-subtasks, etc. - At some point, the sub-subtasks get simple enough to be tackled head-on. •The basal ganglia (BG) play an important role in this network

What criticisms arose over introspection, and why were many psychologists becoming dissatisfied with the method?

- Different results using same methods for Imageless Thought Controversy - Kulpe: i think verbally, imageless - Wundt/Titchner: when i introspect, i see images - two *unable to reconcile b/t they had no notes/data to compare* - Many psychologists became *dissatisfied both with the focus of psychology being the "mind," and with the use of introspection as a scientific technique*.

How are associations stored and what role do connections play in distributed memories?

- Each association is stored in the relative strengths of multiple connections. - Each connection participates in the reconstruction of multiple memories.

What reasoning did Descartes use to come to his conclusion "I think, therefore I am"?

- Even if there is an all-powerful, evil genius bent on deceiving me, he cannot deceive me about my own existence because I must exist in order to be deceived. - Cogito ergo sum: self-evidently true

What is interactionalist dualism, and what problem does this present?

- Even though they are separate substances, minds and bodies interact causally with each other - If minds are not matter and bodies are, how can minds influence bodies?

What is physical causal closure?

- Every physical effect has a sufficient physical cause .-That is, the physical universe contains within itself the resources for a full causal explanation of any of its (caused) elements.

What are turing machines, what are their 2 main components in terms of formal systems, and what are they defined by?

- FSMs w/unlimited external memory - theoretical tool/def'n of computation - machine head is FSM, potentially infinitely long ticker tape is external memory - defined by their tables; program infused in the hardware

Describe the functional equivalence of ARTHUR _1 and _2, and what this means for turing machines

- For any input word x, ARTHUR1 halts if and only if ARTHUR2 halts - For any input word x for which they do halt, the two machines leave the same answer on the tape: for all (∀) x (f1(x) = f2(x)) - Two TMs with very different control tables can nevertheless implement the same input-output correspondence.

What is the fatal flaw of substance dualism?

- For anything to cause a physical object to move, or cause any change in one, there must be a flow of energy, or transfer of momentum, from the cause to the physical object. - But how could there be an energy flow from an immaterial mind to a material thing? - How could anything "flow" from something outside space to something in space?

Learning: its behavioral definition, product, and potential

- Learning is the process by which relatively long-lasting changes occur in behavioral potential as a result of experience. -Process of change; Memory is the product. -Potential: It is not necessary that learning leads to an immediate change in behavior; the effects can manifest themselves later, when an appropriate behavioral test is given .

How are dendritic trees roughly analogous to logic gates?

- Logic gates contain multiple transistors (that act as gates) into a relatively small circuit that calculates a definite logical function over multiple inputs. - Dendrites combine multiple ion channels (gates) into a relatively small circuit that calculates an algebraic sum of multiple inputs (called dendritic integration)

What is epiphenomenalism?

- Mental phenomena "float above" the physical substrate, like a shadow. - There is physical-to-mental but no mental-to-physical causation.

Summarize Leibniz's mill argument, as well as the counterargument against it

- Mill argues that perception is inexplicable in terms of mechanical reasons bc if you were to enlarge a "perception machine" and enter it, you would only find parts that push one another and never find anything that explains perception - Counterarg: if you were to enlarge your laptop and go in and walk among the circuitry and transistors, you would never see the spell-checking functionality of the word-processing program -- category mistake

How are memory traces stored in the brain and what then does retrieval entail?

- Patterns of activation come and go, leaving traces behind when they have passed. - Traces are changes in the weights of the connections between units. - Each memory trace is distributed over many different connections, and each connection participates in many different memory traces. - The traces of different mental states are therefore superimposed in the same set of weights. - Retrieval amounts to partial reinstatement of a mental state, using a cue which is a fragment of the original state.

Describe the 2 phases of the Error Backpropagation Learning Algorithm (aka Backprop) (see slide 41 of lec 22 for pics)

- Phase 1: Feedforward propagation of activation from inputs to outputs: - Phase 2: Back propagation of error (blame assignment) from outputs to inputs:

What does autonomous semantics posit? How does it define the relationship b/t semantic and syntactic structures?

- Phrase and sentence meanings are composed from the meanings of the words plus independent principles for constructing meanings, only some of which correlate with syntactic structure. - some syntactic structures express elements of meaning (not just arrangements of elements) that are not conveyed by individual words.

What is Inference to the Best Explanation or Abduction? What is an example of this in science?

- Postulate unobservable (theoretical) entities (f-rules in this case) to systematize and explain the pattern of interlocking regularities in the data. - correct inference of atomic theory by 19th century physics on basis of regularities found in 1000s of chemical rxns

How did Wundt define psychology, distinguish b/t physicists and psychologists, and what was his main goal in studying it?

- Psychology is the study of "conscious processes" -Physicists study the external world without being part of the situation while psychologists study the psychological processes by which we experience the external world (from the inside; eg. Green is defined by its characteristic wavelength vs. the greenness of the percept) - *Structuralism*: Wundt's goal was to determine the elements and structure of conscious experience (analogous to chemistry's search for the "building blocks" of compounds: "thoughts ~ molecules, both have more basic elements w/in them)

Who was Gottlob Frege and what did he develop?

- The greatest logician since Aristotle - developed formal language of logic; introduced variables, quantifiers

What is mind-body supervenience?

- Things (objects, events, organisms, persons, etc.) that are exactly alike in all physical properties cannot differ with respect to mental properties. - That is, physical indiscernibility entails psychological indiscernibility.

What characterizes valid reasoning? What does Logic do and not do in this context? What does valid reasoning requires in terms of language?

- Transitions in thought that aim to preserve truth from their premises to their conclusion - Logic does not itself aim at the truth, rather it merely transmits truth: premises -> conclusion - Requires an unambiguous language -- stable meaning

What are the 4 main differences b/t Turing Machines and Von Neumann Machines?

- VNMs can accomplish in one step what requires a long sequence of TM steps - TM has only one register; VNM has several dozen - TM has linear, sequential access memory; VNM has random, addressable access memory -- cells of memory marked by addresses/coordinates that can be jumped to - TM has only a few hardware instructions (e.g., "move L", "move R"); a modern CPU has several dozen instructions (e.g., "increment", "add", "jump")

What is linguistic performance and its role in linguistics?

- What a person actually does, the occasional imperfections have nothing to do with one's competence. - linguists maintain that their focus should be on competence and psycholinguists should focus on performance -- psychologists and cog scientists don't entirely agree

What examples do Buddhist doctrine put forth to support the idea of the anatman (no-self)?

- When a lute is played upon, there is no previous store of sound; and when the music ceases it does not go anywhere in space. - There is a path to walk on, there is walking being done, but there is no traveler. There are deeds being done, but there is no doer. There is blowing of the air, but there is no wind that does the blowing. The thought of self is an error...

Describe the "Rider and the Elephant" preliminary

- a rider (rational mind) sits atop an elephant (emotional mind) traversing a path (environmental factors) - "The rider acts as a spokesman for the elephant, even though it doesn't necessarily know what the elephant is really thinking."

What is the critical/sensitive period of language acquisition, and what is particularly problematic to learn after this period?

- ability to acquire a first language in an effortless and ultimately successful way begins to decline from age six and is severely compromised by the onset of puberty -- *syntax and morphology are particularly problematic*.

What is intentionality, and what context does it put beliefs into?

- aboutness: intentional state is about something, represents something, is directed @ something, or means something -- has content - Beliefs are paradigmatic intentional states.

What are the aims of Shannon's Information Theory, and how are semantic aspects regarded in relation to it?

- aimed at answering how to measure info content, how to compress data, and how to communicate reliably over noisy comm. channels (in effect, how to send info down telephone wires) - semantic aspects of comm./info are irrelevant to eng. prob., but doesn't mean eng. aspects are irrelevant to semantic aspects -- if interested in the semantics of the message, the content is still important

What is a heuristic search?

- an evaluation fcn calculates an educated guess about the value of each position - an algo. relies on these eval.s to prune the search tree

What kind of network do the entorhinal cortex and hippocampus form, and what does this do?

- auto-associator network - Pattern completion: When a partial pattern is presented, the network fills in the missing parts. - Content-addressable memory

What are broadmann areas based on?

- based on histological properties (how (stained) slices of brain tissue look under a microscope)

What is the productivity of language?

- basically infinite number of sentences can be generated/understood from finite knowledge

How has the brain evolved, and what does this say about the deeper structures and evolution as a process?

- brain has evolved from inside out - deeper structures perform more basic, fundamental functions - evolution is a conservative process: everything must keep functioning as it grows

What does Rumelhart's brain-style computation say about the brain, what is it based on, and how does it attempt to use the brain as a template?

- brains are parallel, distributed, analog computers - BSC not based on logic, but rather imitates biological computation of human brain, therefore makes no sharp distinction b/t processors and memory - begin @ neurally-inspired level and ask what kind of info-processing tasks easy/natural for brain computation -- model of it good for what? how can we get systems like this to exhibit emergent intelligence?

TRANSISTORS: What can they act/work as, how many wires do they have, what are the 2 main types and when are there logic values 1, when does power flow through them, how do they operate, and what are they analogous to?

- can act/work as switches; - has 3 wires; - two types: nMOS and p(positive)MOS; - 3 components: source, gate, and drain; - power flows when there's contact b/t source and drain, and gate controls this contact (when open elec flows b/t them); - operate a lot like EM switches; gate closes when contacts GND, zone of positive charge acts as barrier to negative charge transmission across gate; when gate not on GND channel of negative charge connects S & D to convey neg charge across them - for nMOS, when S & D connected logic value is 1 (opposite for pMOS) - analogous to gated channels of neurotransmitter receptors

What is the minimal LoT hypothesis and its 3 main claims?

- capacities to think depend on *representational system* such that: 1.*Compositional syntax*: rules blindly followed -- complex representations are built from a stock of basic elements 2. *Compositional semantics*: The meaning of complex representations depend on their structure and the representational properties of the basic elements. 3. *Context insensitivity*: The basic elements reappear with the same meaning in many structures; meaning is relatively stable

How are creole languages formed, what evidence is there of there relative independence from the pidgin, and how do they serve as evidence of UG?

- children growing up in pidgin-speaking comm.y and use pidgin as raw-material for more grammatically rich system (improvement on parent's lang.) - grammatical devices not traceable to any of the parent languages of the pidgin - parents can't learn or create it bc they're outside the critical period, but the children create the grammar based on their expectations of "what a language has to look like" (ie. UG)

Describe language in an anthropological and evolutionary context [REVISE]

- chimps have diff.t mouth shape/vocal mechanisms, cannot produce speech - humans co-evolved w/their lang. -- natural selection: those who were better at producing language were more likely to cooperate with others to survive and reproduce

Why is semantics regard as a combinatorial system, and what are its units?

- clear from the many approaches to semantics in the literature that semantics is a combinatorial system independent of, and far richer than, syntactic structure. - units are not nouns and verbs but rather entities like individuals, events, predicates, variables, and quantifiers.

What is the fundamental problem of representation?

- cog scientists have not quite articulated how (by virtue of what) mental representations represent - several theories (like the so-called informational/causal co-variance theory), but no consensus

What is synchronous discipline, how does it work, and what maintains the current state of the FSM?

- deliberately restrict choices made, introduce global clock - waits for slowest path to complete comfortably before going to next step; - output of registers maintains current state of FSM (memory) -- input to them are specifications of what next state should be

What can combinatorial circuits implement, what is the limitation of this, and what 2 limitations are there of combinatorial circuits?

- digital circuit w/m input and n output wires can implement any finite boolean function with m binary inputs and n binary outputs after some propagation delay imposed by the physical substrate - ie. outputs are determined by their current inputs - limited by fact that they can only progress at one step at a time and have a finite # of patterns

What do decorticate animals show about the cortex?

- doesn't eliminate any functions but seems to reduce them all to some extent - don't seem to add much in way of new movements to animal's behavioral repertoire - extends usefulness of all behaviors and makes them adaptive in new situations

What is combinatorial explosion, and what does it in turn necessitate?

- even w modest branching factor, # of nodes in search tree grows exponentially w/depth of the search ("height of tree") - tf search needs to be guided and selective (eg. when lose something don't check everywhere, but previous places remember having them)

How does white matter comprise a massively interconnected network?

- fibers interconnecting neurons (myelin sheath covering axons) - connections loop under the cortex (u-shaped) - Principle of re-entry: Most connections (though not all) are bidirectional

How did Chomsky successfully argue against behaviorist FSG

- finite boxes don't represent infinite possible sentence combinations - words in a sentence may depend on a word much earlier in the sentence (accg to FSG, intervening words b/t them would disrupt this dependence, short-term memory of the initial word)

How does the finite memory of FSMs limit them, and what does this allow in terms of definition and implementation of FSMs?

- finite memory: the number of their internal states is limited. - because they are finite, FSMs can be defined with transition tables and implemented in hardware (why the control unit of every digital computer is a FSM) - FSMs are limited in what they can compute: as the inputs get larger and larger, every given FSM eventually runs out of memory and fails to keep track of the intermediate results needed to solve the problem

how does ARTHUR_1 work?

- for every valid input it's guaranteed to HALT, substitutes A's for B's and vice versa - it moves to the left first then starts changing letters rightward - when done leaves tape in same condition as ARTHUR_2

how does ARTHUR_2 work?

- for every valid input it's guaranteed to HALT, substitutes A's for B's and vice versa - only diff from A1 is it moves to right first then starts changing letters leftward - when done leaves tape in same condition as A1

Describe the general communication system of shannon's info theory and what the objective of it is

- from some info source, message is encoded into some code, transmitter sends signal carrying code, which gets corrupted by some noise in transit, receiver receives this corrupted signal -- objective becomes removing the noise and smoothing the received signal - receiver does this using probabilistic inferences: this always involves Bayes Theorem

What kind of consciousness can functionalism account for, and what debate remains as a result?

- functionalism can account for access consciousness, at least in principle. - the debate is whether it can account for phenomenal consciousness as well.

What does a function approximation task entail?

- generalize from training examples to obtain target value of new instance - target = function(predictors) - *study pics on first few slides of lec 22*

What is a common heuristic formula for evaluation of chess positions, and why is it heuristic?

- give values to piece, add up all your pieces and subtract opponent's - there are exceptions to this being beneficial, so formula is not infallible

What are diencephalic animals, what do they respond to, and how is their movement and affective behavior carried out?

- have intact brainstem but removed forebrain - respond to simple stimuli and thermoregulate effectively - voluntary movements occur spontaneously and excessively but are aimless - well-integrated but poorly directed affective behavior

What does one ask when formulating an effective procedure, how does one formulate one, and what does formulating it allow?

- how far can we go w/o intuition, ingenuity, invention and *instead w/a rule-governed, mechanical approach?* - formulate everything so no guess work/intuition is involved, allowing them to be mechanized

What reality does productivity by recursion explain, and what is recursion?

- how there's a finite # of words and rules but infinite possible sentences - *recursion*: something contains or is defined by a (smaller) version of itself (eg. noun phrase w/in a noun phrase)

What is protocol analysis and verbal reports?

- human participants given puzzles/arithmetic probs and asked to think aloud while doing them - verbal reports are considered introspection "done right" -- useful if identified properly

What does "the physical facts fix all the facts" mean in regard to physicalism?

- if one was to design a world in which the laws of nature and initial conditions are specified as 2 different physical facts about the world and physicalism holds in this world, then their job as a Creator would be finished and they just sit back and watch - the remaining facts would emerge from the physics

How do the study of thought and language shine a light on one another?

- if you can formulate a thought and then output it to the world, that proves that you are also correctly mentally representing that thought - if we can study language, we can indirectly study thought, although perhaps not comprehensively (not all thought can be articulated) - any theory of the semantics of thought would more than encompass the semantics of language

Describe the PDP/Connectionist neuron

- ignores cell geometry, ion-current dynamics, and spike timing - operate in discrete time cycles - simple unit, very limited in function by itself - "Activation" models firing rate (0-100 spikes/s) - "Weights" model synaptic efficacies.

What is a finite state machine/automaton, what does it encode states in, and what is its transition table?

- in each cycle it is in a particular state, it moves from state to state based on its input - can encode states in binary bit patterns - transition table is boolean function : takes finite Boolean values as input, outputs finite Boolean values

What are the two basic intuitions of information?

- info is extensive quality: longer messages potentially contain more info; can add together chunks (bits) of info to create more [ for independent A and B: I(AB) = I(A) + I(B) ] - info reduces uncertainty (inversely related to it): improbable messages contain more info

What is the Big Picture accg to LoT theory, and the major questions that arises from this?

- inside mind exists natural lang and LoT, connected to each other, but LoT has an intentionality/aboutness - major question: how do these get their content?

What are propositional attitudes?

- involve a thinker S taking an attitude ψ {belief, doubt, hope, fear, ...} towards a proposition p; intentional states about a proposition

What are constraints in linguistics, and what may they consist of? Give an example

- kind of rule that places extra conditions on structures created by formation rules and derivational rules. - may consist of conditions that structures must necessarily satisfy, or of conditions that help make a structure more "favorable" or "stable." - ex. transitive verbs, such as dislike, must be followed by an NP (in underlying structure)

What levels of the brain do we have varying levels of knowledge for, and why does this then necessitate creating models?

- know lots about macro- and microscopic org..n of brain, but comparatively little about circuit level in b/t - therefore try to compensate by theorizing about this level, create models to bridge across levels of abstraction

Describe the connectionist network model and what it is an example of

- layered structure of info processing in brain is simplified and idealized in connectionist models - millions of simple units (PDP neurons) form networks - example of a feedforward network: inputs -> hidden -> outputs

What is the main idea of population code, and what advantage does this offer?

- main idea: use not one, but multiple neurons to increase the info-carrying capacity of the channel - an entire pop'n of neurons encodes a variable of interest - plays absolutely central role in computational neuroscience

Summarize Tolman's study in which he found latent learning in rats

- mice made to run a maze - 3 groups: 1. immediate reward 2. delayed reward 3. no reward - when group 2 was given reward, they showed fastest maze times -- suggested before reward they were learning the layout of the maze

According to functionalist theory of mind, what is the mind, how are mental states characterized, and what distinction (if any) is made b/t the mind and the brain?

- mind is info-processing system - mental states are functional states, multiply realizable (both humans and octopi feel pain through diff.t structures -- in principle robots can feel bc it's functionally defined); cause/relate to other mental states - mind/brain are same, just two different perspectives by which to look at it 1. neurological "hardware" 2. psychological "software"

How does functionalism define minds in terms of structure?

- minds functionally/relationally defined in way that abstracts away from structure that implements them - in principle can be implemented by non-physical "thinking substance", although most functionalists are physicalists

BCM: what is it, why better hebbian rule, and what is its likely implementation

- modification threshold adjusts dynamically as a fcn of the avg postsynaptic activity (change in weight b/t sender and receiver = learning rate parameter(n) x presynaptic activation x non-monotonic fcn of the postsynaptic activation for which the receiver is the arg.) - this BCM learning rule fits a lot of data on synaptic plasticity and has nice system-level properties. - Likely implementation: LTP+LTD

How did General Finite-State Grammar build on FSG?

- parts of speech replace arbitrary words in behavioral language

What is a category mistake?

A semantic error in which things of one kind are presented as if they belonged to another or, alternatively, a property is ascribed to a thing that could not possibly have that property.

The authority of the past was being shaken by skeptical arguments that were undermining the received wisdom

What was occurring during the epistemological crisis of the 16th century?

Briefly summarize Wittgenstein's private language argument

i can mark a certain sensation some way, but it will only serve as an ostensive definition so i can remember the connection in the future; but in marking it i have no criterion of correctedness against which to check it, so in the future any use of the mark is going to seem right, and that means that we can't talk about 'right' in this context

What is the problem of consciousness?

if the mental is physical, how can we make sense of the phenomena of consciousness?

What is the function/purpose of a buffer (logic gate)?

important to maintain digital discipline due to voltage degrading on wire, signal could attenuate and flip sign (eg. bad cell reception/signal); buffers repair signal and convey it down the line

According to John Watson, what topics should psychologists never study and which should they, and why?

scientists should only study events they can directly observe (ie. environmental stimuli and behavior), must never use the terms consciousness, mental states, mind, introspectively verifiable, content, imagery, etc/

How did Chomsky alternatively explain sentence structure formation?

sentences have a hierarchical design in which words are grouped together into successively larger structural units called phrases

What is an ostensive definition?

show what a term means by pointing to the thing it refers to

What are the 2 components of input for universal TMs?

software - Description of the machine table for the TM being simulated data - Input for the TM being simulated

Functionalism: how public/private is the mind and what is the role of inputs, outputs, internal states, material composition, and functional roles?

mind neither absurdly public in way behaviorism says, nor completely private as cartesianism says -- halfway house b/t them inputs and outputs still central, but added inferences of real internal states, both of which have causal powers (analogical to FSMs) material composition of the components are irrelevant (medium independence), functional roles define them (eg. mousetrap, thermostats)

How does the brain have a hierarchical functional organization?

new structures add up on top of old ones, don't replace their functions but rather modulate and augment them to improve overall functioning

What are phrase structure rules?

paradigmatic examples of formation rules

What is discipline in design (generally-speaking)?

the act of intentionally restricting your design choices *so that you can work more productively at a higher level of abstraction.*

What are ionotropic receptors?

transmitter binds to the binding site and the pore opens, allowing the influx or efflux of ions (ligand-gated)

What is Moravec's paradox, and some possible explanations for it?

• "It is comparatively easy to make computers exhibit adult level performance on intelligence tests or playing checkers, and difficult or impossible to give them the skills of a one-year-old when it comes to perception and mobility." • Possible explanations: - "We're least aware of what our mind does best." - Sensorimotor skills took millions of years to evolve. - Logic machines (= computers) are good at logic, not so much for other things

What is causation by content, and what problem arises from this?

• *Causation by content*: propositional attitudes cause behavior and propositional attitudes cause behavior as a function of their content • *Problem*: How can propositional attitudes have causal effects as a function of their semantic properties - propositions stand in inferential relations to other propositions, not in causal relations to perceptions or behavior?

Name and describe the two main restrictions of the Fodorean module

• *Domain specificity* is a restriction on the representations that a cognitive mechanism can take as input - that "trigger it" or "turn it on." - Roughly, a mechanism is domain specific to the extent that it can only take as input a highly restricted range of representations. •*Informational encapsulation* is a restriction on the kinds of information a mechanism can use as a resource once so activated - paradigmatically, though not essentially, information stored in memory. - A cognitive mechanism is encapsulated to the extent it can access, in the course of its computations, less than all of the information available to the organism as a whole.

Physical symbol systems: define their memory, symbols, operations, interpretation, and capacities

• *Memory* - Contains structures that contain symbol tokens - Independently modifiable at some grain size • *Symbols* - Patterns that provide access to distal structures - A symbol token is the occurrence of a pattern in a structure - Required if enough variety is required of the system • *Operations* - Processes that take symbol structures as input and produce symbol structures as output • *Interpretation* - Processes that take symbol structures as input and execute operations • *Capacities* -Sufficient memory and symbols -Complete composability and interpretability

What is the inductive principle?

• Any hypothesis found to approximate the target function well over a sufficiently large set of training examples, will also approximate the target function well over other unobserved examples. • However, there are no guarantees...

What is the Frame Problem?

• Applying an operator to a state in a state-space problem will affect some relations between objects in the state but not others. •Frame problem: How is the system to keep track of which relations have changed and which not? •Proofs would be very complicated if there were framing axioms for each operator specifying the relevant changes and non-changes.

What is means-ends analysis and the problem-solving steps it goes through? What are two complications of it?

• General algorithm for solving problems: 1. Evaluate the difference between the current state and the goal 2. Identify a transformation that reduces a salient difference 3. Check that the transformation in (2) can be applied to the current state: a.If it can, then apply it and go back to step 1. b.If it cannot, then return to (2) and try a different transformation •Two complications: - Sometimes one needs to back-track to an earlier decision point - Operators have preconditions that generate subgoals

Summarize GOFAI: it's guiding intuition, it's emphasis, synonyms, and how it's regarded in the field as a whole

• Guiding intuition: The mind *is* a computer. • Emphasis on reasoning, problem solving, heuristic search, and knowledge representation. • A.k.a. "classical", "symbol manipulation", or "language-of-thought" AI. • Track record of extravagant predictions that are now remembered as cautionary tales, but its accomplishments are undeniable -- still going strong.

In principle, how can Finite-State Grammar be learned, and what two limitations are there of this idea?

• In principle, such grammar can be learned by associating pairs of words (or categories). • FSGs have extremely limited memory - only the last word that was produced - but dependencies between words in a sentence can extend over many words (can't account for recursion) • would take children >30yrs to learn lang due to ~1 bil associative links among specific word combos if they were learning @ 1 assoc./sec.

What is representational theory of mind, and what is an influential proposal that is along these lines?

• Main idea: Have a representational system in which syntax tracks semantics: - Connect the causal properties of representational vehicles with their semantic properties via their syntax. - Fodor's language of thought hypothesis

How does General Problem Solver (GPS) problem solve? Where does it start, what is its task, what allows intelligent action, and what kind of knowledge does it have?

• Problem specification: - Initial state (premises, position, data, ...) - One or more goal state(s) - A set of operators for transforming one state into another. • The task is to find (search for) a sequence of operators that transform the initial state into the goal state. • Regularities allow intelligent action • The agent has knowledge, suppose it is supplied with heuristics of two sorts: -Procedures to detect salient differences between states -Rules of thumb about which operations typically reduce differences of which sorts.

How do hidden units solve the XOR problem?

• The XOR problem can be solved if we add a hidden layer to the network. • The hidden layer re-represents the inputs in a way that makes the problem linearly separable and thus solvable by the output unit.

How is fMRI data usually analyzed?

• The most basic method subtracts the fMRI maps of an experimental and control condition. • Most methods use multiple linear regression. • FMRI data are inherently correlational.

What is the physical symbol system hypothesis, what are its necessary and sufficient conditions, and how does it define general intelligent action?

• The necessary and sufficient condition for a physical system to exhibit general intelligent action is that it be a PSS. • Necessary: Any physical system that exhibits general intelligence, whatever additional structures and processes it might have, will contain a PSS. • Sufficient: Any PSS can be organized further to exhibit general intelligent action. (Universality) • General intelligent action: The same scope of intelligence seen in human action; Use knowledge to achieve goals and adapt to the environment.

What ensures a valid argument? What does validity guarantee and not guarantee? What kind of conclusion would suggest a false premise?

• The validity of an argument is its logical strength: whether the conclusion really follows from the premises. •Validity does not guarantee truth, but it does guarantee truth preservation. • Conversely, if we have reached an absurd conclusion via a valid argument, then at least one of the premises must be false

What is arbitrary I/O mapping, and how does Hebbian Learning contrast this?

• There are no constraints on which pattern maps to which pattern. • Example: Names -> phone #s • Example: Concepts -> words • In the limit: Random associates task • By contrast, in Hebbian (aka covariance) learning, similar inputs must be mapped to similar outputs.

How has the boom and bust cycle of emergent technologies affected AI?

• Went through several cycles: -Hype and extravagant promises and predictions -Influx of funding -Disappointment and criticism -> "AI winter" -Funding cuts -Renewed interest years or decades later • Cycles more about perception of AI by government bureaucrats and venture capitalists. • Despite rise and fall of AI's reputation, it has continued to develop new and successful technologies.

What is the Universal Turing Machine and how does it work?

• a particular TM that can simulate any TM upon specification: -Specify the desired TM on the tape -Provide input pattern on the tape -UTM will produce the output pattern of the specified TM

How do interpreted formal systems build on formal systems?

• assign meaning (or interpretation) to the tokens of a formal system. •The rules according to which the tokens in a formal system may be manipulated and what those tokens mean are closely related - compositional semantics

What is Belief-Desire Psychology, and what are some synonymous terms?

• explaining and predicting other people's behavior in terms of propositional attitudes, in particular: -What they believe to be true. -What they desire to achieve. • A.k.a. commonsense psych., folk psych., theory of mind, naïve psychology

What are formal systems? What are some components of token-manipulation systems?

• like a game in which tokens are manipulated according to definite rules, in order to see what configurations can be obtained. (eg. generative grammar, algebra) •Token-manipulation systems -A set of types of formal tokens or pieces -One or more allowable starting positions -A set of formal rules specifying how positions may or must be changed into other positions.

What is the difference b/t the mentalist and abstract perspective of language study?

• mentalist - language is a product of a mind, it is a window to the mind - language is a cognitive structure; words, sentences, etc. express conceptual representations in the mind • abstract - language has nothing to do with minds - refers to objects, states, etc. in possible worlds (eg. logic language of philosophy, math)

What are the 3 major types of rules in linguistics, and the 2 interpretations of these rules?

• rules: 1. formation rules 2. derivation rules (aka transformations) 3. constraints • *interpretations* 1. an abstract way of describing a class of structures -- no claims about processing 2. algorithmically -- generate grammar (eg. sentences)

What is an automatic formal system?

• system that "moves" by itself. • More precisely, it is a physical device or machine such that: -Some configuration of its parts or states can be regarded as the tokens and positions of some formal system -In its normal operation, it automatically manipulates these tokens in accord with the rules of that system.

How are digital circuits disciplined, and what disadvantage and advantage does this offer?

• use discrete voltages -- leave a forbidden zone, (logical 0), and arrange circuits to never operate in this zone • Therefore, digital circuits are a subset of analog circuits and in some sense must be capable of less than the broader class of analog circuits, however, digital circuits are much simpler to design

What are the "easy" and "hard" problems of consciousness?

•"Easy problem": - Explain how and why mental states are reportable. - Explain how a cognitive system can access its own internal states. - Explain the difference between wakefulness and sleep. - etc. ... [Note that all these will be functional explanations.] •"Hard problem": - The problem of subjective experience; Qualia. [Phenomenal "glow"] - There is something it is like to a conscious organism.

Solipsism: what does it say, how did Descartes open the door to it, how is it related to idealism, and how do most philosophers regard it?

•"I am the only mind which exists." •Descartes's method of radical doubt and his "Cogito ergo sum" opened the door. •Many types of idealism lead quite directly to solipsism. •Most philosophers regard solipsism as a philosophical dead end.

Why can't language be entirely private?

•"Language games" are rule-governed practices that people engage in -- language itself is rule-governed •[Verificationsim:] A rule must have a criterion of correctness to verify that the rule is being followed in each specific case •*An isolated mind is fallible: mistakes can occur during reasoning, e.g., memory lapses.* •Without some resource external to the mind, there can be no criterion of correctness.

What is the Problem of Other Minds, and what other questions does it raise?

•"What justifies the almost universal belief that others have minds very like my own?" •It also is an effective, if roundabout, way of asking, "What is the nature of mind?" and "Can machines (or robots) have minds?"

What are the three terms and three rules that are relevant to learning?

•(More than) Three terms are relevant to learning: - Presynaptic activation xi - Network-generated postsynaptic activation yj=yj(x,w.j) - External training (teacher) signal tj - Other: the current weight wij, sliding thresholds, ... •Three learning rules: -Unsupervised Hebbian (no tj): -Supervised Hebbian (ignores yj): -Delta rule: uses all three terms

What do the Church-Turing Thesis and the Mathematical Church-Turing Thesis posit about computation?

•*CT*: any fxn that is intuitively computable is computable by some TM (Turing-computable) •*Mathematical CT*: Any function that is computable by following an effective procedure is Turing-computable.

What are the two main modalities of synaptic transmission?

•*Chemical synapse*: - Involves neurotransmitters - One-directional - Requires sophisticated molecular machinery - Metabolically expensive - Allows great flexibility •*Electrical synapse*: - Via gap junctions - Two-directional

What are the 2 main characteristics of sequential circuits? What is an example of this type of circuit from language theory?

•*Circuits with memory*: Their output depends not only on the current inputs but also on their internal state; the internal state contains information about previous inputs. •*Computational process*: Disciplined sequence of state transitions. • Finite-State Grammar

What are the restrictions on the Fodorean module?

•*Domain specificity* •*Information encapsulation* •Shallow outputs •Mandatory application •Fast operation •Fixed neural architecture •Specific breakdown patterns

What two components must a theory of semantic structure include?

•*Expressiveness*: must be able to express all the semantic distinctions made by a natural language. •*Universality*: Languages are (largely) inter-translatable -- hence the stock of semantic structures available to be used in particular languages must be universal. (inter-lingual)

What is the difference b/t feedforward and recurrent networks?

•*Feedforward networks* can be drawn so that all connections point in one direction. - There are no loops. - The information flows from inputs to hidden units to outputs. •*Recurrent networks* have loops. - It is possible to start at some node, travel around the links, and return back to the same node. - The recurrent networks are dynamical systems whose behavior depends not only on the stimulus but also on their internal state.

What are the 2 main types of representations?

•*Localist representation* - Each item is represented by one unit - Each unit represents one item ("grandmother cell") - Example: Traffic lights •*Distributed representation* - Each item is represented as a pattern of activity over multiple units - Each unit participates in the representation of multiple items - Example: letters and words

What are the neurological complications of the Error Backpropagation Learning Rule?

•*Nobody knows how to implement backprop in a biologically plausible manner.* - Some researchers claim it can't be done. •Other learning rules exist that accomplish the same job and are more (although still not quite) biologically plausible. •The search for a biologically plausible error-correcting learning rule is an active research area. - Some neuroscientists doubt that the brain does error-correcting learning at all - others hope that evolution must have come up with something, given that error-correcting learning is so advantageous for survival.

What are the characteristics of hippocampal formation?

•*Specialized for rapid learning of arbitrary associations ("binding")* •Sparse, non-overlapping representations minimize interference •Temporary storage of arbitrary current states for later retrieval •Context-specific and episodic memories

How can synaptic strength and firing rates affect neuronal transmission?

•*Strong synapse*: Each presynaptic spike causes a large post-synaptic potential. •*Weak synapse*: Each presynaptic spike causes a relatively small PSP. •*Alternative formulation in terms of firing rates*: Assume the presynaptic neuron fires at a constant rate, then the stronger the synapse, the higher the postsynaptic firing rate

What is the necessary condition for learning? What can this be compared to?

•*The test instances come from the same source that had generated the training instances.* - If this condition does not hold, we may just as well ignore the observations and consult an astrologer. •Compare with the encoding-specificity principle of episodic memory.

What is the working hypothesis regarding error-correcting learning in humans?

•*Working hypothesis: The brain implements some approximate version of error-correcting learning via stochastic gradient descent.* •We don't yet know how the brain does it, but there are good reasons to believe that evolution has found a way.

Summarize the 2 main competing theories of language acquisition

•*universal grammar*: - something in the genes code for syntax - Rule-based grammar - autonomous, modular syntax •*general learning*: - genes code for proteins, not concepts like syntax; genetic endowment specifies general learning mechanisms, not language-specific rules - lang is inextricably intertwined with the rest of cognition, it's not special, like learning to ride a bike, etc.

What/when are the major milestones of language acquisition? (12 wks - 4 yrs)

•12 wks: cooing, smiling •6 mos: babbling, e.g, ba-ba-ba-ba •10 mos: no longer able to discriminate non-native phonemic contrasts •12 mos: first words, understands simple commands •18 mos: can pronounce up to 50 words •24 mos: two-word phrases, productivity •30 mos: explosive increase in vocabulary, steady increase of mean length of utterance (MLU) •3 yrs: ~1000 words, grammatical complexity •4 yrs: well-established language

What is an operationally complete level, and what implications does this have on describing a system at a given level?

•A level is operationally complete (or "strong") to the extent to which its behavior is determined by laws formulated at that level, applying to its state, as described at that level. •The strength of a system level is a property of nature, and not just something in the head of the observer. •Therefore, being able to describe a system at a given level carries an empirical claim -- that abstraction to this level still preserves all (or most) that is relevant for future behavior described at that level of abstraction.

Describe the leaky integrate-and-fire neuronal model

•A mathematical model of an idealized neuron •The input current I(t) is integrated across time •Analogy: Leaky bucket •When the membrane potential V(t) reaches a predefined threshold, the neuron fires a spike and resets V← Vreset

What is content-addressable memory and how can it be activated? What role does the system play?

•A memory of an episode that "binds together" multiple components - e.g., faces of several people, voices, what was said, where, when, ... •We can re-activate the memory by giving a retrieval cue that contains some of the original contents. Ex: Google keywords •The system fills in the missing parts.

What was BF Skinner's simplistic view of sentences?

•A sentence is merely a sequence of words. •sentences are learned chains of responses: -A stimulus S0 elicits the first word. -The first word serves as the stimulus for the second word, second for the third, and so on

What is the difference between access and phenomenal consciousness?

•A-consciousness: - A representation is A-conscious if it is poised for free use in reasoning and for direct "rational" control of action and speech. •P-consciousness is experience: - Experiential properties of sensations, feelings, and perceptions. - Also of thoughts, wants, and emotions.

What is AI the science and study of?

•AI is the science of making machines do things that would require intelligence if done by humans. •AI is the study of the computations that make it possible to perceive, reason, and act.

Compare and contrast the AMPA and NMDA receptors

•AMPA Receptor - Opens when Glutamate binds - Permeable to Na+ - Transmission of activation •NMDA Receptor - Opens when BOTH Glutamate binds and membrane is depolarized - Permeable to Na+, Ca2+ - Functions as co-activation detector, triggers learning

How does the Na-K pump power the brain?

•Active mechanism (consumes energy) •Drives Na+ and K+concentrations away from thermodynamic equilibrium •Moves 3 Na+ out of the cell for every 2 K+ into the cell •Sets the resting membrane potential Vrest < 0. •Separate transporter moves Cl- out of cell

What were Newell's and Anderson's definitions of cognitive architecture?

•Alan Newell's (1990) definition: The fixed (or slowly varying) structure that forms the framework for the immediate processes of cognitive performance and learning. •John Anderson's (1983) definition: A theory of the basic principles of operation built into the cognitive system.

How are orthogonal, localist, and distributed patterns related and what does this allow?

•All localist patterns are mutually orthogonal. •By rotating the coordinate system, we can turn a set of localist patterns into a set of distributed patterns. •Orthogonality is preserved -> the resulting distributed patterns are still orthogonal. •Thus, there will be no interference and multiple distributed memory traces can be stored in the same weight matrix

What is input-output mapping?

•An I/O mapping is a (long) list of correspondences, w/ each correspondence mapping one pattern to another pattern •In mathematics, such mappings are called functions: y = f(x) •This is a quintessential learning task, essential building blocks of any behaving system.

What is Input-Output mapping?

•An I/O mapping is a (long) list of correspondences. •Each correspondence maps one pattern to another pattern •In mathematics, such mappings are called functions: y = f(x)

Go sequentially through the steps of neurotransmitter release

•An action potential reaches the axonal terminal. •Voltage-gated Ca2+ channels open up and Ca2+ enters. •Ca2+ binds to synaptotagminproteins incorporated into vesicle membranes. •These proteins interact with SNARE complex proteins, causing fusion of the two membranes. •Exocytosis of neurotransmitters follows •The vesicle membranes are recycled back into the interior.

How are the hippocampus and neocortex in constant interaction?

•An episodic memory consists of many components stored in separate cortical areas. •The hippocampal region binds these components into a coherent episode. •Over time, the components become linked to each other and hippocampal involvement is no longer required.

What is the role of prior knowledge in induction, what may the conditions of this knowledge be, and how can this affect the generalizations made?

•Any system that is prepared to make inductive leaps must enter the learning situation with some a priori assumptions about the world. •This prior knowledge may be present in the learner by design, by evolution, or as a result of learning during earlier stages of development. •The learner's generalizations tend to be successful to the extent to which its inductive assumptions actually do hold in the environment (at least approximately).

What is Global Workspace Theory? Describe its basic architecture, stance on consciousness, and definition of attention

•Basic architecture: - Domain-specific specialized processors - Domain-general global workspace •Consciousness is restricted to the information within the global workspace. - Modular processing is unconscious. •Attention as a gate-keeper: which results of modular info processing are allowed into the global workspace.

Go through the process of storage and retrieval for learning

•Begin with an empty network (W=0) •Store (write) an association: -Present cue x = [ 1 0 1 0 1 0 ] on input layer -Present associate y = [ 1 1 0 0 ] on output -Make small weight changes accg to HLR: W= W + ΔW •Retrieve (read) by reconstruction: -Present cue on the input layer -Calculate netinput and use that as argument of function to find output activations

What were the titles and achievements of Alan Turing?

•British logician,mathematician,cryptographer, and philosopher. •Credited as the father of modern computer science. •Invented the Turing Machine and Turing Test.

What is the Delta Learning Rule: what are the weights changed in proportion to?

•Change each weight wi in proportion to: -The activation xi of the corresponding sending unit i -The prediction error (t-y), where y is the predicted activation of the receiving unit and t is its target activation

Give an example of how football can illustrate error backpropagation rule?

•Consider a football team that tries to make a pass, but the receiver fails to catch the ball. •Linemen -> Quarterback -> Receiver •The receiver is indeed responsible for some of the blame, but not all of it. - some blame should be propagated backward to the quarterback who threw the ball too far. •The quarterback is indeed responsible for some of the remaining blame, but not all of it. - some should be propagated further back to the linemen who did not protect the QB well.

What was Descartes' geometrical method, and what Euclidian method was this analogical to?

•Construct a system of certain knowledge using the two fundamental, foolproof operations of reason: -Intuition ("I see clearly and distinctly that...") -Deduction •Analogous to axiomatic method

What is the basic framework for control in the brain?

•Cortico-striato-thalamo-cortical (CSTC) loops •The cortex represents the current state: stimuli, goals, action plans... •The projections from cortex to BG calculate the respective utilities of these stimuli and/or actions. •Winner-takes-all selection in the BG: Which action has best utility? (Noisy) •Cortico-thalamo-cortical pathways are gated (dis)inhibited by the BG.

What persistent activity is seen in the neuronal firing rates of monkeys doing delayed match-to-sample tasks?

•Delayed match-to-sample task in which monkeys received fruit juice for pressing the button that matched the sample after a 15-sec delay. •Persistent firing of spikes throughout the delay: Active maintenance of a WM trace of the stimulus. •Note that all data on this slide come from a single neuron on multiple trials.

What were some of Chomsky's distinctions and contributions to cognitive science and linguistics?

•Described human language as a sophisticated, hierarchical system that involves storage and manipulation of sophisticated, hierarchical phrase structures •Defined an algorithmic basis for this manipulation •Mathematical theorems - Chomsky's hierarchy of formal grammars •Opened up a whole new field of inquiry

What role does the Hippocampus play in binding things together and creating formations?

•Different aspects of an episode (e.g., sights, sounds, smells) are represented in disparate cortical areas. •The hippocampus binds them into a coherent episode. •It specializes in the rapid formation of arbitrary associations.

What are distributed representations and why do they support generalization?

•Distributed representation - Each item is represented as a pattern of activity over multiple units - Each unit participates in the representation of multiple items - Example: letters and words •Stimuli have overlapping representations. •Higher similarity -> more overlap. •Due to representational overlap, some connections that were modified during training are reused at test.

How does the Delta Rule solve a multiple regression problem across trials?

•Each output unit has to solve a (nonlinear) multiple regression problem •Optimal weights: minimize the summed square error (SSE) across all patterns

How are postsynaptic potentials the material substrate of WM?

•Each presynaptic spike releases some neurotransmitter. •The neurotransmitter lingers in the synaptic cleft for a few milliseconds, causing a PSP. •The residual neurotransmitter is a memory trace of the spike. •It is possible to design networks in which new spikes are emitted just in time to refresh (and thereby maintain) the decaying memory traces. •This is the molecular basis of WM!

What is parallelism?

•Each realm runs in isolation, like a clock made by God and set in motion to run in sync . •The two realms run in parallel, deterministically.

What role does medium independence play in functional systems?

•Every system must be implemented in some medium. But, as long as the medium supports the relevant structure, only the form of a formal system is significant, not its matter. •The form supervenes on the matter.

What are EPSPs and IPSPs?

•Excitatory PSPs -Depolarize the postsynaptic membrane -Make the receiver more likely to fire •Inhibitory PSPs -Hyperpolarize the membrane -Make the receiver less likely to fire

What are the main excitatory and inhibitory types of synapses?

•Excitatory: -AMPA -NMDA •Inhibitory: -GABAa -GABAb

What is the "The Gate Is Closely Guarded" preliminary?

•Few things are as anxiety-provoking as the realization that death is inevitable and to some extent unpredictable, so the physicalist hypothesis about the MBP is bound to encounter very strong resistance; rational arguments just don't cut that deep.

What kind of circuits are FSMs? As a result, what do they always contain? In synchronous circuits, how are cyclic paths broken? Finally, what do state registers do, and when do registers update?

•Finite State Machines are sequential circuits -- they always contain cyclic paths. • Synchronous circuits: All cyclic paths are broken by inserting registers somewhere on the path. •State registers store the current state (encoded as bit patterns). •A register updates its state only when the global clock changes from 0 to 1.

What breakthroughs did Brenda Milner make in memory research?

•General memory function can be dissociated from other cognitive functions such as language. •Long-term memory can be dissociated from working memory. •Within LTM, declarative (explicit) memory can be dissociated from non-declarative (implicit) memory. •Role of the medial temporal lobe.

What are the benefits of distributedness?

•Generalization to similar patterns - The network generalizes to novel patterns that are similar to known input patterns - Similarity = feature overlap •Robustness to damage - Good (slightly degraded) reconstruction from imperfect cues (noisy inputs) - Graceful degradation when units or weights get damaged

What does the red spot on Jupiter illustrate about processes, and how does this relate back to Descartes's "cogito ergo sum"?

•Giant red spot serves as vivid example that a process can persist for a very long time even though there is no "substance" underneath it. • ie. thinking can persist without a thinker

What is occasionalism?

•God causes both the physical and the mental, like a puppeteer with a puppet in each hand. •Mental and physical appear to interact, but they don't actually

What is the good and bad news of hebbian learning?

•Good news: - Hebbian learning rule is neurologically plausible: LTP, NMDA - Hebbian can learn many input/output mappings - Hebbian learning can establish autoassociative memories. •Bad news: - Hebbian cannot learn arbitrary input/output mappings.

What idea did the CT thesis along with Turing's guiding intuitions about the mind lead to?

•Guiding intuition: -Cognition is computation. -The human mind is a computer of some (yet unknown) sort. •But then, in light of the Church-Turing thesis, it can be realized on a Turing Machine -- only need way to turn this idea into a testable hypothesis (Turing Test)

How did HM show evidence for separate explicit and implicit memory systems?

•H.M has severe explicit / declarative memory disorder •H.M. is almost normal on procedural or implicit memory tasks: - Mirror tracing/Mirror reading - Priming - Classical conditioning •This dissociation shows that explicit memory depends upon the temporal lobes whereas implicit / procedural memory does not.

How are computer paradigmatic examples of functional systems?

•Hardware/software distinction •One can develop software (e.g., a Java applet) that can run on multiple hardware platforms (e.g., PC, iPhone,...) •One can study algorithms (e.g., sorting) in the abstract, bracketing hardware out. •The computer is a single system that can be described in two ways: hard- & softw.

Compare and contrast the Hebbian learning rule with the Delta learning rule

•Hebbian rule detects co-occurrences •Hebbian involves no teaching signal •Delta rule reduces the discrepancy ("error" = t-y) between the target output t and the predicted output y •Both rules make small incremental adjustments to the weight, in proportion to the presynaptic activation xi.

What is the Mind-Body Problem, and the two problems within it?

•How are our mental states, events, and processes related to the physical states, events and processes? •The problem of *mental causation*: if the mental is not physical, then how can we make sense of its causal interaction with the physical? •The problem of *consciousness*: if the mental is physical, how can we make sense of the phenomena of consciousness?

How can reinforcement learning occur in the BG?

•Hypothesis: The timing of memory updates and the utility of the stimuli and/or actions is acquired through trial and error using reinforcement learning algorithms. •This leverages the evolutionarily earlier use of BG to perform reinforcement learning for motor control. •Dopamine as a signal of reward-prediction error.

How does Cartesian Metaphysics explain the existence of God?

•I think, therefore I am. •I am and I have the idea of God in my mind, therefore God is! -Ontological proof: God is the only possible explanation of the idea of a Most Perfect Being, that is clearly and distinctly present in my mind. •God is no deceiver (God put the idea of God's existence in my mind, and since God doesn't deceive, God exists).

What are the biological implications of learning, and what philosophical idea does this harken back to?

•If long-lasting changes have occurred in the behavioral potential of an organism, long-lasting physical changes must have occurred in this organism's brain. •Recall that philosophers of mind use the term supervenience for the notion that mental changes must always be accompanied by physical changes.

What is the counterargument to the Knowledge Argument of p-consciousness?

•Imagine you learn all the physical facts about swimming by reading books and by watching the coach and the swimmers from the sidelines. •Will you be able to swim when you finally get in the water yourself? •Will you learn anything new? •Arguably, what you acquire when you get in the water is a new procedural skill. - there need be no declarative facts over and above what you learned from books.

What is the propagation speed of action potentials in myelinated and unmyelinated axons?

•In unmyelinated axons: <1 mm/ms •In myelinated axons: 10-100 mm/ms

What is inflection and some examples of it in English?

•Inflection = the modification of a word's form to indicate grammatical information of various sorts •Examples of inflection in English: -Singular vs plural: the book vs the books -Past vs nonpast: We work. vs We worked.

What are the components of the simulated neuron, and how are net input and output activation calculated?

•Input activations xi •Output activation y •Connection weights wi -- correspond to synaptic efficacy, can be excitatory (positive) or inhibitory (negative). •Activations -- correspond to firing rates •The net input to each unit equals the weighted sum of the inputs to this unit: - netinput = w1(x1) + w2(x2) + w3(x3) + ... + wn(xn) •In our model, the output activation of each unit is a threshold linear function of its net input: - y = F(netinput)

What is the task of pattern associator networks, what system is it part of which provides what, how does each output neuron calculate its activation, what do all outputs receive, and how are outputs regenerated?

•Input patterns W(aka P2) •Output patterns Y(aka P1) •The task is to map each input pattern to a specific output pattern. •Part of a larger system, which provides inputs and interprets outputs. •Each output neuron calculates its activation Yj independently of all other output neurons. •All outputs Yj receive the same inputs Xi, but with different weights Wij •Regenerate outputs given an input pattern: - netinputj = W1j(x1) + W2j(x2) + ... + Wnj(xn)

What modern objections are there to the Turing Test?

•Input-output equivalence, which is the same as behavioral equivalence, is not sufficient to guarantee equal mentality. •Only a test for human-like mentality. •Ignores embodiment and real-time perception and action. •Is language really sufficient to probe all of cognition?

How does the spirit of behaviorism live on in functionalism?

•Inputs and outputs are still central •Real internal states (cf. Descartes) •Internal states have causal powers

What are ion channels are the 3 types?

•Ion channels (or "receptors") are proteins embedded in the cellular membrane that selectively allow passage to certain ions or molecules •Ion channel types: - Ligand gated: activated by neurotransmitter - Voltage gated: activated by depolarization - Double gated: activated by co-occurrence of neurotransmitter and depolarization

Describe the neuron at rest: which ions in the solution, membrane and resting potential, and what leads to excitation?

•Ions in solution: Na+, K+, Cl− •Membrane potential: Vin - Vout •Resting potential: -65 to -70 mV (that is, the inside voltage is more negative than the outside voltage) •Excitation is depolarization(Vin becomes less negative or even positive relative to Vout)

What were the benefits and detriments of psychological behaviorism?

•It provided needed counterweight to the prior, speculative mode of theorizing. •However, its refusal to consider mental mechanisms at all was too narrow. •Within their self-circumscribed domain, the behaviorists made many important discoveries in animal learning, etc.; they established a robust experimental rigor

How are multiple associations stored in the same weight matrix?

•Just add the weight changes incrementally •The resulting matrix stores both associations •The mathematical term for this is superposition. •All changes accumulate locally in the synapses.

What is recurrent auto-associate memory and its key idea?

•Key idea: Feed the outputs back as inputs. •The connections are recurrent - they loop back onto the same units. •Repeat until convergence - that is, until the pattern stops changing.

What is the Auto-Associator Network and its key idea?

•Key idea: Learn to make the output pattern identical to the input pattern •Input patterns x •Output patterns X •Everything else is the same as any other pattern associator

What support is there for the CT theses?

•Lack of counterexamples •Confluence: Various formalisms, proposed independently by different authors all define the same class of functions. •*Turing's (1936) argument: TM seems capable of reproducing any operation that a human being can perform while following an effective procedure.*

According to UG, how do languages differ from one another, and what explains how children learn lang so quickly? What do they define as the role of external input?

•Languages differ only in the settings of a relatively small number of parameters. •The child learns language quickly because it knows most of it already -- needs external input only to discover the parameter settings of the local language and to learn words.

What are the functions of layer IV, layer V, and layers II and III of the cortical microcircuit, and what is its neuronal composition?

•Layer IV: Input •Layer V: Output •Layers II + III: Lateral •Neuronal composition: -Pyramidal 75% -Spiny stellate 10% -Smooth stellate 15%

What is a lexicon and what are they composed of?

•Lexicon: A list of morphemes and the rules for combining morphemes (The smallest units of language that carry meaning)

What is the Knowledge Argument of p-consciousness?

•Mary the color scientist: - Lives in a black-and-white room from birth. - Has complete knowledge of color science. - When Mary finally leaves her B&W room, will she learn anything new? •Qualia = subjective experiences. •There is something it is like to see red. - If Mary learns anything new, it cannot be understood in information-processing terms.

What is long-term potentiation, and why does it necessitate long-term depression?

•Massive, prolonged stimulation of one pathway leads to permanent strengthening ofthe synapses to another pathway. - input-specific - associative •With LTP only, the network would soon become saturated. •LTD weakens low-frequency connections to prevent such saturation

What are firing rates and what modulates them?

•Measured in spikes per second •The sending (presynaptic, input) neurons modulate the rate at which the receiving (postynaptic) neuron fires action potentials.

Why are models good?

•Mechanisms are made explicit. •Models deal with complexity, span levels. •Models promote theoretical consistency. •Models help understand phenomena. •Models generate predictions. •Models enable complete experimental control. •Models can be applied to practical problems, useful tools.

What is the method of exaggerated skepticism, and how is this illustrated by the indestructible object principle?

•Method of exaggerated skepticism: I will doubt everything until I reach a bedrock of certain truth. -The senses are fallible. -Assume an "evil genius" is bent on deceiving me. -Only way to find indestructible object in the room is to smash everything until you find something that cant be smashed

How distributed memories robust to damage, and what is this called?

•Minor damage tends to cause a small change in response to many inputs, rather than a total loss of some memories and no effect on others. •Distributed memories are robust to damage -- called graceful degradation.

Why are nonlinearities important in nets?

•Multi-layer linear nets are no more powerful than one-layer linear nets: - y=Ax, z=By -> z=(BA)x •Multi-layer nonlinear nets, however, are much more powerful than one-layer nets: - y=f(Ax), z=g(By) -> z can be a very complex function of x

What is the infinite regress problem of the PFC?

•OK, the PFC controls the brain. •But who tells the PFC what to do??

What are the 3 main types of non-interactive dualism?

•Occasionalism •Parallelism •Epiphenomenalism

How is learning a theoretical abstraction?

•One never observes learning (or memory) directly. One infers that learning has occurred on the basis of observations that experience changes behavior. •Behavior is the window to the mind. •Learning can cause both increases and decreases in behavior.

Describe ACT-R's Perception-Action Cycle

•Perceptual modules construct representations •Central bottleneck: only one production fires at a time •Motor modules plan and execute movements

How does pattern completion occur in recurrent auto-associate memory?

•Present a fragment of an input pattern - for example, activate 5 out of the 8 units. •The network fills in the rest of the pattern - for example, by activating the other units. •Importantly, any fragment can be used as a cue. •As long as the retrieval cue contains sufficiently many elements of the whole episode, the missing elements can be filled in.

How can generalization occur in pattern associator networks in which similar inputs give rise to similar outputs ?

•Present a novel input pattern, that is, a pattern the network has not been trained on: x = [1 1 0 1 0 0] •This pattern is similar, but not identical, to our second training pattern [1 1 0 0 0 1] •Given this novel input, the network produces:y = [0 1 0 1] -- Generalization has occurred.

How is I/O mapping carried out in pattern associator networks?

•Present input pattern x = [1 0 1 0 1 0] •Calculate the activation of each output unit, independently and in parallel: - netinputj = W1j(x1) + W2j(x2) + ... + Wnj(xn) - the weight matrix determines the mapping from inputs to outputs •The network produces the output pattern y = [1 1 0 0]

What is property dualism? Provide an example from physics to illustrate

•Property dualism: mental properties are diverse from and irreducible to physical properties. •Example from physics: The electron has gravitational mass and electrical charge.

What is property dualism and Panprotopsychism?

•Property dualism: mental properties are diverse from and irreducible to physical properties. •Panprotopsychism: Everything is conscious to some extent - electrons just a little, dogs a lot...

What are the components of a general learning experiment, and what is this analogous to in computer science?

•Put an organism in an environment •Provide experience: stimuli •Some time later, test the organism with some aspect of the original experience •Based on the outcome of the test, infer if learning occurred •Analogous to the WRITE and READ operations in computer memories

What are the main types of physicalism?

•Reductive physicalism -Behaviorism •Nonreductive physicalism •Functionalism

Why may representations not be necessary for intelligence?

•Representations (or "models") are inaccurate and obsolete. •Are they really necessary for intelligence? •"The world is its own best model." •The phenomenon of change blindness suggests that our brains don't really build a detailed visual representation of the world.

What is the appeal of physical symbol processing, but what does it still not eclipse in terms of goal formulation?

•Representations are both meaningful and physical at the same time. •The processing is carried out according to formal rules and obeys physical laws. •Still, the goals formulated at the knowledge level are accomplished by virtue of the systematic correspondence between syntax and semantics .

What is representative perception and the egocentric predicament?

•Representative perception: the theory that our ideas correspond to and faithfully represent objects in the external world. •Egocentric predicament: if all we can directly know is our own ideas, then how could we ever know whether our ideas correspond to anything, or even approximate anything, "out there"?

How do the robots R1, R1D1, and R2D1 illustrate the Frame Problem?

•Robot R1: Pulled the wagon out of the room. Fail to appreciate the importance that the bomb was also on the wagon. •Robot-deducer, R1D1: Took too much time to calculate a long list of potential side effects. The bomb went off... •Robot-relevant-deducer, R2D1: Took too much time deciding which implications are irrelevant and can be ignored safely.

What are derivational rules, and how can they be thought of? Give an example of one. What do they always leave behind?

•Rules that take a fully formed structure and change some aspect of it; can be thought of more generally as a relation between two structures, one "more underlying" and one "more superficial." •Example: "Fatima really disliked that movie." -> "That movie, Fatima really disliked [it]." •Movement rules always leave behind a "trace" - a sort of unpronounced pronoun, whose antecedent is the moved constituent

What are some arguments against UG?

•Statistical learning algorithms are very powerful. Much more powerful than most people realize -- some have already demonstrated lang.-learning •Language universals might be due to non-language universals and/or historical factors. •double dissociations are never absolute. •brain is a massively interactive system. •Genes code for proteins, not word order. •Communicative demands enforce grammar.

three major brain scans

•Structural MRI - Neuroanatomy: tissue density, etc. - Widely used for medical diagnosis •Functional MRI (fMRI) - Blood-oxygenation-level dependent (BOLD), correlated with the average energy consumption of "voxels" of brain tissue. - Poorer spatial resolution but better temporal res than structural MRI •Diffusion Tensor Imaging (DTI) - Orientation of white-matter tracts - uses water flow in the white matter to measure the relative direction of white- matter tracts -- structural technique.

What are the 3 three main, hotly contested questions of the "Language War"?

•Symbolic: Are formal grammar rules expressed in the brain? •Autonomy: Is grammar independent of other brain structures? •Genetic: Is there some special genetic encoding specifically for grammar?

What was Cajal's Synaptic Plasticity Hypothesis?

•Synaptic plasticity: the strength and/or number of synaptic connections changes as a result of experience.

What was the Cartesian project, and how did Wittgenstein challenge this idea?

•The Cartesian project was to build a system of certain knowledge using nothing but the resources of the individual mind. •Wittgenstein argued that the resources of any individual mind are not up to the challenge. •Human memory is fallible and language, which is inherently social, is needed

What are the neural substrates of the Global Neuronal Workspace?

•The GNW is not a single neural location, but a distributed network of high-level processors that are massively connected to other high-level processors. •GNW is characterized by numerous, long-distance cortico-cortical connections, particularly dense in PFC.

Describe the steps of operating a turing machine

•The Operator: -Writes a "Question" on the tape. -Positions the head. -Presses Start. •The control unit takes over: -The head moves left and right. -Reads and writes symbols on tape. •Eventually the machine halts (unless it loops indefinitely). •The Operator reads the "Answer" from the tape.

What is the Identical Elements Theory of Transfer, and what does this imply about generalization?

•The amount of transfer from a familiar situation to an unfamiliar one is determined by the number of elements that the two situations have in common. •Generalization requires representational overlap.

What role do the Basal Ganglia play in action selection?

•The basal ganglia selectively facilitate one response pathway while suppressing others •The basal ganglia evolved for motor control. •Later, BG were the natural circuit to control internal actions such as working memory.

How does the brain represent objects in the world, how does info processing occur as a result, and what role do synaptic weights play under this context?

•The brain represents the objects from the external world as patterns of activation across populations of neurons. •Information processing in the brain is carried out by transformations of these patterns of activation. •The synaptic weights of the connections between neurons govern how these transformations are carried out.

How is the brain always in learning mode, and what makes this possible?

•The brain typically operates in both Read/Retrieve and Write/Store modes simultaneously. •That's possible because the synaptic changes are small and incremental. (This is modeled by setting the learning-rate parameter to a small value) •The brain uses ("Reads") the existing weights to produce behavior and also updates ("Writes") them a little according to the current activation patterns

What are the characteristics of the posterior cortex?

•The canonical type of processing •Sensory processing •Motor processing •Knowledge-dependent inferences •Slow integrative learning •Overlapping, distributed representations

What does the CPU of a digital computer contain and for what function? What do the CPU hardware circuits implement?

•The central processing unit (CPU) of a digital computer contains several dozen registers that keep its internal state. •The CPU hardware circuits implement several dozen instructions: -Read/write the registers from/to memory -Arithmetical, logical, and comparison operations among the registers -- equivalent to the state transitions in a Turing Machine

What is the Universal Grammar hypothesis of language acquisition?

•The child comes to the language-learning situation prepared -- they already know certain aspects of language in advance. •The standard term for the unlearned component is Universal Grammar or UG.

How do redundant codes remove noise from signals?

•The encoder encodes the source message s into a transmitted message t, adding redundancy to the original message in some way. •The channel adds noise to the transmitted message, yielding a received message r. •Using the known redundancy introduced by the encoder, the decoder infers (probabilistically) both the source message and the added noise.

What is the error gradient, what are its 3 characteristics, and how does it differ from the error function?

•The error gradient ∇E(w) quantifies how the error changes in the vicinity of a given point w. •The gradient ∇E(w) is a vector: - As many dimensions as w. - It points in the direction of the greatest change of E. - Its magnitude measures the rate of change (= steepness of the slope). •Note that the error E(w) is a scalar field, but the error gradient is a vector field - think of a tiny arrow attached to each point in space.

How do neurons integrate their inputs?

•The membrane potential at the axon hillock is determined by the weighted combination of all EPSPs and IPSPs. •There is both spatial and temporal integration •Like a tug of war: the excitatory and inhibitory PSPs pile up according to the laws for superposition of electric fields. •Thus, the postsynaptic membrane computes the net input to the neuron.

What provides the myelin sheath, which specifically in which areas of the NS?

•The myelin sheath of axons is provided by specialized glial cells: -Oligodendrocytes in the central NS -Schwann cells in the peripheral NS

What are the properties of the neocortical system of memory?

•The neocortex uses distributed representations. •This enables pattern overlap and thus similarity-based generalization. •However, it is prone to catastrophic interference •*The cortex needs to learn slowly and integrate information over multiple experiences.*

What is the output of nonlinear model neurons and how are they related to linear neurons?

•The output is a monotonically increasing, but non-linear function of the net input. •Much more powerful than linear neurons

What is the phonological loop and four pieces of evidence for it?

•The phonological loop can maintain about two seconds of speech by implicit verbal rehearsal. •It is not a halfway station to long-term memory. •Otherwise analogous to the short-term store (STS) in the multi-store model. •Evidence: - Speech rate correlates with memory span - Articulatory suppression reduces the memory span - Phonological similarity reduces the memory span - Irrelevant speech reduces the memory span

What does language of thought hypothesis posit

•The representational vehicles are sentences in the "language of thought" (aka LoT or Mentalese) -- concepts are expressions with intentionality in LoT, and language converts these from LoT to English •They have compositional syntax and semantics. •Mentalese is not English, it's (relatively) free of the imprecision, ambiguity, and vagueness characteristic of human languages.

How are distributed representations widespread in the brain?

•The same neuron responds to tones of different frequencies •Tuning curve - max response to one preferred freq - weaker responses to nearby stimuli •Thousands of neurons respond to any given stimulus.

According to the Heuristic Search Hypothesis, what are solutions represented as and how do physical symbols systems exercise intelligence?

•The solutions to problems are represented as symbol structures. •A physical symbol system exercises intelligence in problem solving by search -- that is, by progressively modifying symbol structures until it produces a solution structure.

Describe recurrent nets as dynamical systems

•The state of the network is the pattern of activations across the population of neurons. •The state evolves along a trajectory in phase space. •An attractor state has a basin of attraction.

What is the methodology and usual finding of the Brown-Peterson distraction task?

•The subjects are asked to count backwards by 3s during the retention interval. •The span drops precipitously because of retroactive interference.

What is synaptic tagging hypothesis?

•The synapse is tagged by the stimulation •Plasticity products (PPs) are synthesized in the soma •The molecular tag "captures" the PPs. •Untagged synapses do not learn.

What is the paradox of language acquisition?

•The whole community of linguists, working together for decades with all sorts of cross-linguistic and psycho-linguistic data unavailable to children, has still been unable to come up with a complete characterization of the grammar of a single natural language. •Yet every child does it by the age of 10.

Describe the 2 theorems which posit that networks can be universal function approximators

•Theorem: Feedforward networks with one hidden layer using arbitrary squashing activation function can approximate any (Borel measurable) function from one finite dimensional space to another to any desired degree of approximation, provided sufficiently many hidden units are available. •Theorem: The network weights that implement this approximation can be learned from input-output data.

What are some of the dark, mortal implications of physicalism?

•There is no immaterial soul that survives the disintegration of the brain. •Death is the end of an individual's existence. •The only thing that survives is the person's legacy, and it too will not survive for long.

What is the Gradient-Descent Optimization Algorithm, and what is analogical to?

•To find the bottom of the valley, simply follow the arrows. •An iterative optimization algorithm: - My current location is w. - Obtain a statistically unbiased estimate of the local gradient ∇Ê(w) - Make a step in the direction of (the negative of) the gradient. - Step size proportional to magnitude: Δw ∝ -∇Ê(w) -Make this my new location. - Repeat. •Analogy: Walking downhill on a foggy day in the mountains.

How are turing machines a type of automatic formal system?

•Tokens: The symbols on the tape. •Formal rules: The machine table of the control unit. •Once started, the machine runs autonomously until it (ever) stops.

What is the distinction b/t top-down and bottom-up interpretations of a rule? Give an example of each in order to answer this

•Top-down: A noun phrase (NP) has as parts a determiner (Det), and adjectival phrase (AP), and a noun (N), in that order. •Bottom-up: A Det followed by an AP followed by an N may be taken to constitute an NP.

What are graded EPSPs and IPSPs?

•Vary in magnitude depending on how much transmitter was released by the sender •Attenuated with distance

How can Auto-Associator networks "Clean Up" degraded patterns, and what advantage does this present?

•We train the system with "good" patterns so that the weight matrix accumulates knowledge about the valid "vocabulary" in the domain. •We then test with noisy or incomplete patterns •Because pattern associators can generalize and are robust to damage, the reconstructed (output) pattern can be less noisy and more complete than the test (input) pattern.

What is the difference b/t weak and strong AI?

•Weak AI - Systems that solve real-world problems. - Compared against the ideal standard of rationality or against a practical standard •Strong AI - Systems that shed light on (and even explain) various sorts of human intelligence and consciousness - Compared against human performance

Define the idea of Hebbian Learning

•When an axon of cell A excites cell B and repeatedly or persistently takes part in firing it, some growth processes or metabolic change takes place in one or both cells so that A's efficiency as one of the cells firing B is increased. •In effect, cells that fire together, wire together.

What occurs at the cell body as a result of neural transmission?

•When neurotransmitters bind to receptors on dendrites or soma, they open and close ion channels on the postsynaptic membrane -- this changes the postsynaptic potential. •Causes either Excitatory postsynaptic potential (EPSP) or Inhibitory postsynaptic potential (IPSP)

What are the main ideas of the Error Backpropagation Algorithm?

•When the input vector x(p) is presented, the network performs a forward pass in which activation flows along the links, with the current weights w. - This pass determines the activations across the hidden units y and the output units z. •A teaching signal t(p) is provided for the output units only. - The errors on the output layer are calculated: δout = t − z. •Now comes the tricky part: The network performs a backward pass in which certain error-related quantities flow backwards along the links, with the current weights w. - This redistributes part of the "blame" to the hidden units: δhid. •δout and δhid scale the error gradients w.r.t. w_out and w_hid. •Finally, all weights are updated acc. to SGD: Δ*w* ∝ -∇E^(p)(*w*)

Describe the Philosopher's Zombie and the implications it has on functionalism

•Your zombie twin: - Behaves exactly like you - Has a brain identical to yours. - The only difference is that the zombie has no experiences. •If zombies are possible, the functionalist theory of mind is incomplete in principle.

Under the context of a distributed memory system, describe mental states, alternative mental states, and information processing

•a mental state is a pattern of activation over the units. •Alternative mental states are simply alternative patterns of activation. •Information processing is the process of evolution in time of mental states which depends on the connection weights, which store the network's knowledge.

Summarize the chinese room argument against the validity of the Turing Test

- person receives pieces of paper through one window and passes out pieces of paper through another, paper have Chinese symbols written on them - person inside room has huge instruction manual telling them which papers to pass out based off papers they receive -- just need to be able to identify Chinese symbols in some sort of syntactic way - now imagine manual is written in way that inputs are all Q's in Chinese and outputs are appropriate answers -- but person in room doesn't know Chinese, just following instructions in english instruction manual - Searle: perfectly possible to have syntactic symbol manipulation without any intelligence/understanding

Where do pidgin languages generally occur, how are they formed, and what are their drawbacks?

- place of contact of people coming from different speaking areas - mix together aspects of parent languages to get some rudimentary meaning across, typically borrowing vocab from parent lang.s *often in phonologically degraded forms* - *lack stable word order, their grammatical organization* has a rudimentary "Me Tarzan, you Jane" flavor, *lacking inflection and subordination*.

What are the facts of physical life under the context of circuits? What are race conditions?

- propagation delay is caused by the capacitance and resistance in the circuit - slow down when hot and speed up when cold, thus causing delays during operation - twin chips will have non-identical delay characteristics - *A tiny difference in the timing can have dramatic impact on the rankings*

Machine BABY and MOM

- read about these in Haugeland (1985) - pink tape keeps track of current state of blue tape

What evidence is there supporting UG?

- sensitive periods for lang. acq. - limited ability of apes to acq. human lang - localization of lang fxns and aphasias - parallels b/t spoken & signed lang. - genetically-inherited lang. deficits - creole languages and NSL

Summarize William James' "Fundamental Fact"

- similar to Descartes' cogito ergo sum - 'it thinks' similar to 'it rains' or 'it blows' ('it' as a nonpersonal, grammatical placeholder) as in "thought goes on"

What are neural network models composed of and implemented on, and what functional advantages do they offer?

- simulated neurons and weighted connections - implemented on computer which enables one to carry out simulations and have contradictions exposed - promote explicit definitions and specifications - can lead to unexpected output - abstraction: ignore many aspects of real neurons, preserve functionally important properties

What does functionalist theory specify, what is each state defined in terms of, how do they relate to each other, and what role does implementation play?

- specifies a system of functional states - each state defined in terms of 3 functions w/in the system: 1. which inputs from outside system it reacts to 2. how interacts causally w/other internal states 3. what outputs if any the state produces - functional states interdefined - no commitment about implementation

What composes the Striato-Thalamo-Cortical Loops and what does it play a role in?

- striatum is collective name for several important BG nuclei - reciprocal connections b/t striatum, thalamus, and cortical areas - such recurrent networks play crucial role in controlling overall brain function

What did Introspection entail?

- systematic self-observation and self-report - reaction times - word-association

How do the cortex and thalamus compose a crucial, massively interconnected system?

- thalamus comprised of ~50 nuclei which connect reciprocally with specific cortical areas - its dense loops into and out of cortex represent functionally a 7th cortical layer - interacts directly w, or is the target for, nearly every other part of the brain - can't live w/o thalamus, can live even if missing major portions of cortex

What is the optimization problem regarding the error surface of all possible weight vectors(*w*)? (Best to look at slide 27 of lec 22 w/this)

- the optimization problem is to find the weight vector w* that corresponds to the lowest (minimal) point on the error surface - these are the weights that optimize the fit to the training data.

What did Tolman describe as being the best way to explain latent learning, and how was this in contrast to behaviorism?

- the rats created cognitive maps, or mental representations, of the maze - r/t than explaining behavior through S-R psychology and behavior alone, explained it through S-O-R (stimulus-organism-response), using behavior to infer about mental states

What was Watson's goal for psychology? What two types of responses did he define?

- theoretical goal is the prediction and control of behavior. •Two kinds of responses: -Explicit response: overt and directly observable, e.g., pressing a lever -Implicit response: occurs inside the organism, e.g., glandular secretions, neural activity, visceral movements

How does Cartesian philosophy distinguish privacy b/t body and mind?

- thinking cannot occur outside the mind/consciousness, which itself is private, only accessible to owner

What is linguistic competence, and how does grammar fit into it?

- to be able to produce and understand an unlimited number of utterances, including novel and unfamiliar ones, and recognize that certain ("ungrammatical") utterances do not belong in the language - Grammar: the mental system behind our linguistic competence.

What is motivated reasoning and interpretive biases?

- tricks people use to reach the conclusion they want to reach. - "interpretive biases" that tend to operate subconsciously in the service of emotion regulation - people are quick to embrace ideas that make them feel good and resist ideas that threaten them or their worldview.

What was Aristotle's fundamental insight of validity, and what does this allow?

- validity depends only on the structure (or form) of the argument, not on the specific content. - allows validity to be studied in content-free arguments with abstract terms such as X, Y, Z

How did logic arise from Aristotle?

- validity of arguments; fundamental goal of Aristotle's logic was to capture, systematize, and formalize validity - defined syllogism (valid patterns of reasoning)

What did the Wug Test show about f-rules in language?

- young children have extracted generalizable f-rules from the language around them, rather than simply memorizing words that they have heard

What are the 4 main components of an effective procedure?

-*Executability*: procedure contains finite # of deterministic instructions, which have finite & unambiguous specs commanding the execution of a finite # of primitive operations. -*Automaticity*: The procedure requires no intuition about the domain, no ingenuity, no invention, and no guesses. -*Uniformity*: The procedure is the same for each possible argument of the function -*Reliability*: If the procedure terminates, the procedure generates the correct value of the function for each argument after a finite number of primitive operations.

What are the 2 levels at which one can describe a system? What is this analogous to?

-Abstract, formal, software -Physical, silicon, hardware - analogous to mind and brain

What is Psychological Behaviorism and its goal?

-An empirical research program -Goal is to explain behavior in terms of stimuli and the history of reinforcement and punishment; tries to eliminate mental vocabulary altogether

What are some basic facts of language acquisition?

-Any normal child becomes fluent in their local language(s) without formal instruction -No non-human animal comes close -Feral and neglected children never achieve fluency

What was Euclid's axiomatic method?

-Begin with self-evident truths ("axioms") -Prove theorems based on these axioms and previously proven theorems.

How does the peristimulus time histogram (PSTH) method of measuring neuronal firing rates work, and what does it estimate?

-Collect data from the same neuron on multiple trials. -Align all spike trains at the onset (or offset) of the stimulus. -Divide the time line into bins. -Count how many spikes fall into each bin and construct a histogram. -Calculate spikes per second. •The PSTH estimates the instantaneous firing rate r(t).

How does the peristimulus time histogram (PSTH) measure neuronal firing rates, and what does it estimate?

-Collects data from the same neuron on multiple trials. -Align all spike trains at the onset (or offset) of the stimulus. -Divide the time line into bins. -Count how many spikes fall into each bin and construct a histogram. -Calculate spikes per second. • The PSTH estimates the instantaneous firing rate r(t).

What two main types of solutions did philosophers follow for the mind-body problem of dualism?

-Give up the idea that a [causal] connection exists between the mental and the physical. -Give up Dualism in favor of Monism.

Summarize Descartes's epistemological argument that led him to dualism

-I am such that my existence cannot be doubted. -My body is not such that its existence cannot be doubted. -Therefore, I am not identical with my body. -Therefore, the thinking thing that I am is not identical with my body.

What are the key properties of incremental learning that span across learning rules?

-Learning makes small weight changes -The direction and magnitude of each change depends on the activation levels of the sending unit and receiving unit. -Multiple changes are simply added together as the network encounters new input-output pairs.

What is Philosophical Behaviorism and its goal?

-Primarily concerned with the semantics of our common mentalistic vocabulary -The goal is to translate mental terms into terms that speak only of behaviors and propensities

What was Descartes's explanation for the Mind-Body problem?

-The idea of mind-body union is a "primitive notion" that cannot be explained in terms of more basic notions -The locus of interaction is the pineal gland (where not how)

What two methods are used to test phrase structure

1. *Substitution test* - Some groups of words can be replaced with a pronoun 2. *Movement test* - Some groups of words move together

What are the two main conceptions of memory?

1. Memory Traces: Stored & Retrieved, Bottom-up (Rote, Passive) 2. (Re)constructive Processing, Top-down (Active)

Name and define the 6 components of grammar

1. Phonetics: the articulation and perception of speech sounds 2. Phonology: the patterning of speech sounds 3. Morphology: the structure of words 4. Syntax: the structure of phrases and sentences 5. Semantics: the (literal) meaning of words, phrases, sentences, passages, and texts 6. Pragmatics: how to use language in context

Definition of a problem: 5 components, and how these in turn define a solution

1. unsatisfactory present/initial state 2. more satisfactory goal state 3. operators/moves that change on state into a diff.t state 4. constraints, rules, restrictions 5. problem space: all possible states that can occur w/in the constraints - solution: sequence of valid moves w/in the problem space from initial state to goal state

What does conceptual semantics posit? Accg to it how does and doesn't lang make contact w/the world?

*Conceptual semantics* - mentalistic approach: defines meaning in terms of a relationship between an utterance and conceptual structures in the mind of the language user. (ie. word "snow" doesn't refer directly to snow but to mind's idea of snow, which is what is referring to snow in the world) - Language makes contact with the world only through the complex mediation of the perceptual and motor systems, not some mysterious mind-world relation of intentionality

What are propositions and how do they relate to declarative sentences?

*Propositions*: - home base for linguistics - origin of questions (reformulate propositions to create questions: "the snow is white" -> "is the snow white?") - language independent - "people-independent" - fundamental bearers of truth and falsity Declarative sentences express propositions

What does truth-conditional semantics posit? Accg to it, how does one know a sentence's meaning?

*Truth-conditional semantics* - mainstream, dominant approach - objectivist approach: defines meaning in terms of a logical relationship between an utterance and the states of affairs in possible worlds - to know a sentence's meaning is to know the conditions under which the sentence would be true - truth conditions as possible worlds (eg. if it hadnt rained on sunday i would have cut the grass)

What are the most basic breakdowns of language comprehension and production?

*comprehension*: phonology -> syntax -> semantics *production*: semantics -> syntax -> phonology

What is the emerging consensus of the cognitive theories of consciousness?

"Consciousness operates as a distributed and flexible system offering non-conscious expert systems global accessibility to information that has high concurrent value to the organism."

What are the 3 basic ideas behind SHAKEY and Logic Programming?

(1) states in state-spaces characterized by sets of predicate calculus formulas (2) problems are conjectures to be proved (3) problem-solving is finding proofs using proof-theoretic techniques and heuristic search techniques

What is the ontological physicalism thesis?

All that exists in this world are bits of matter in space-time and aggregate structures composed exclusively of bits of matter. There is nothing else in the space-time world.

Why do localist representations not support generalization?

Because there is no representational overlap, there is no transfer to other stimuli despite their similarity with the reinforced stimulus.

What rule solves the limitation of Hebbian learning to learn arbitrary I/O mappings, and how does it do so?

Error-driven delta rule, using the discrepancy between actual and target outputs (error)

Briefly summarize physicalism

Everything that exists in the space-time world is a physical thing, and every property of a physical thing is either a physical property or a property that is related in some intimate way to its physical nature.

How did Allen Newell describe problem solving as a search?

Formulation of last resort: If it is not known how to obtain X, then create a space that is known to contain X and search that space for X.

What is the consensus definition of intelligence?

Intelligence measures an agent's ability to achieve goals in a wide range of environments.

How can you use transistors to implement not gates?

Look at slide 16 of L09

What is behavioral translation?

Mental terms express behavioral dispositions and/or neurophysiological states.

What is neuromodulation?

Neuromodulators have spatially distributed (as opposed to targeted), temporally extended effects on the recipient neurons and circuits.

What are formation rules?

Specify how lexical items are combined into larger units (phrases), and how these larger units in turn are combined into still larger units

What is the neuron doctrine?

The neuron is the fundamental structural and functional unit of the nervous system.

Define memory span

The sequence length that yields perfect recall on 50% of the trials

What is the structure requirement of the compositional syntax claim of the minimal LoT hypothesis?

The vehicle of a given propositional attitude must be a structured object that can be put into a one-to-one correspondence (isomorphism) with the structure of the content of that propositional attitude

Provide an illustration of a mutually recursive rule

X -> ...Y... Y -> ...X... X expands into something w/Y, Y expands into something with X

What is Kludge AI?

a specialized system (e.g., Deep Blue) that is terrible at most tasks, mediocre at some tasks, and superhuman at a few tasks.

What is Artificial General Intelligence (AGI)?

a system that can do efficient cross-domain optimization

What is human-level AI?

a system that is able to do many of the things that humans are able to do (eg. Turing test)

What does Boolean mean?

all variables can take only two values: True or False

What hope for immortality does functionalism raise?

assuming functionalism is true, human mind might survive destruction of material brain bc they can be multiply realizable -- transferrable to other physical substrates

What are the brain's main subcortical structures?

basal ganglia, hippocampus, and amygdala

What is the interdisciplinary methodology of creating neural network theories?

combine behavioral, computational, and neurobiological constraints to create the theory

What is the AI Effect?

• "AI is whatever hasn't been done yet." • "A problem that proponents of AI regularly face is this: When we know how a machine does something 'intelligent,' it ceases to be regarded as intelligent. But if I beat the world's chess champion, I'd be regarded as highly bright."

logic values rep.d by voltage levels

digital opposite of continuous: black and white, 2 values driver feeding input to receiver

What reality of problem-solving underlies the inception of sequential circuits, and what does this then necessitate doing to get around this?

eventually finite mechanisms become too large for their resources, necessitate adoption of a divide-and-conquer, sequential strategy

What is the problem of induction?

what justification does one have to predict the target attribute of any new instance, unseen so far, on the basis of our past observations of *other* instances?