CGS 001 Final
What is the principal challenge for neural network models of language learning?
replicating rule-governed behavior without the rules being explicitly taught
What is one crucial idea that came out of Tolman's studies of cognitive maps and has continued to plan an important role in cognitive science?
representations
What claim is made by the systems reply to Searle's Chinese room?
the Chinese room, as a whole, understands Chinese
Which of the following is a central feature of info processing in neural netoworks?
the ability to learn from experience
What gives rise to the integration challenge?
the fact that cognition can be studied at many different levels.
Which of the following most accurately describes what is measured by PET?
the flow of blood
What was Newell and Simon's important contribution to cog sci?
the physical symbol systems hypothesis
Why did some cog scientists interpret Shepherd and Metzler's results as evidence for non-digital info?
the processing time for digitally coded information should be a function only of the amount of info encoded
What is the local field potential (LFP)?
the sum of inputs to neurons in a particular area
Which of the following most accurately describes the 2 visual systems, as proposed y Ungerleider and Mishkin?
the ventral system recognizes and identifies objects. The dorsal system locates objects in space
What are 3 examples of cognitive systems that can be understood heirarchically?
vision, smelling, hearing
What is the main difference between Plunkett and Marchman's model of past tense learning and that of McClelland and Rumelhart?
-Plunkett and Marchman use hidden units -Plunkett and Marchman's network learns by backpropagation -Plunkett and Marchman's network more accurately replicated over-generalization erros
What is one way to build a neural network that doesn't require feedback?
-construct a competitive network -use inhibitory connections, such that any one output can inhibit other outputs -construct a network such that the "winning" unit is the one that fires the most
Why is Marr's top-down approach difficult to apply the domain-general processes?
-hard to identify relevant algorithms for domain-general processes -hard to identify the relevant functions of domain-general processes -hard to identify a determinate set of tasks performed by domain-general processes
What 3 questions do mental architectures attempt to answer?
-in what format do particular cog systems carry info? -how do those systems transform info? -how is the mind organized so that it can function as an info processor?
Allen Newell and Herbert Simon claimed that the essence of intelligent thought is the ability to solve problems. One challenge facing computer scientists is the difficulty of limiting the "search space" when trying to find solutions to a problem. What is one way to reduce the search space?
-means-end analysis and -heuristic search
What claim is made by the robot reply to Searle's Chinese room?
-that Searle is right in claiming that the Chinese room doesn't really understand Chinese -that the Chinese room is too thinly described, and does not account for the complexity of actually understanding Chinese -that, with the proper description, it would be right to claim that the Chinese room understands Chinese
Suppose you want to explain to your friend how olfaction (the sense of smell) works and you say "Olfaction takes info from sensory systems about odors and transforms it into a representation of discrete smells. But I'm not sure how the odors are transformed into discrete smells or what representations are involved in the process. "What level of explanation do you need to answer these latter questions, according to Marr?
Algorithmic
What did the Rees-Friston-Koch hypothesis propose?
there is a linear relationship between the average neuronal firing rate and the strength of the BOLD signal
What position does the physical symbol system hypothesis take on linguistic rules?
they must be represented explicitly
Behaviorists thought that even complex behaviors like language could be explained as a series of conditioned responses
true
Suppose I always walk South to get to the cafeteria but just moved to the other side of campus and now have to walk North. If I were a rat in Tolman, Ritchie, and Kalish's study of spatial learning, what would I likely do? (assume that trips to the cafeteria are rewarding for me, and I go every day)
walk north
What is one feature that is present in ID3, WHISPER, and SHAKEY?
-heuristic search
Which of the following can be used to support unity in science?
-reduction -functional decomposition -identifying bridge principles
All of the signals reaching a neuron's dendrites will increase or promote firing
False
Cog scientists have worked out implementational level analyses for most info processing problems.
False
Each unit in a parallel processing network is physically unique.
False
Neurons in the brain are all basically the same physically, just like artificial neural networks
False
None of the functions of the human body are multiply realizable.
False
SHRDLU's microworld is much more complex than the world of most human beings
False
The local field potential (LFP) is typically negatively correlated with the firing activity of neurons
False
The physical symbol structures hypothesis requires physical symbol structures to be language-like
False
The problem with decision trees is that they aren't designed to handle new cases
False
The visual system is the only cog system that can be broken down into 3 levels (as Marr did with vision).
False
There are no double dissociable functions in the human brain.
False
What is the difference between the formal and semantic properties of the brain?
Formal properties are the structural features of the brain's representations. Semantic properties are what the brain represents.
A common challenge for A.I. researchers is coming up with a way to program rules into an artificial system that will correctly identify what info and which inferences the system should pursue in certain situations. What is this problem called?
Frame problem
What is the main difference between McCullough and Pitt's conception of a neural network and Hebb's conception of a neural network?
Hebb focused on training neural networks
What is a characteristic of a multiply realizable cog ability?
It can perform the same function in many different types of physical systems
Who carried out some of the best-known experiments on the relationship between BOLD and LFP?
Logothetis
Who wrote the classic two-volume collection on training multilayer networks?
McLelland and Rumelhart
What is the main difference between single-layer and multilayer neural networks?
The units in a single-layer network receive inputs directly multilayer networks receive inputs indirectly.
What is the principal challenge for multilayer networks?
They are difficult to train
What type of analysis is illustrated by Marr's three levels of explanation?
Top-down
What is an algorithm?
a finite set of rules that can be applied systematically to transform objects in circumscribed ways
Whose ideas about the nature of human behavior are most strongly represented in the design of SHAKEY's programming?
Lashley
Who coined the term "perceptrons"?
Rosenblatt
The symbol grounding problem is best illustrated by....
The Chinese room
Which of the following claims about the language of thought would Fodor endorse?
The language of thought depends on casual interactions between physical states.
Chomsky's theory of syntax illustrates the hierarchical nature of language (as described by Lashley)
True
Fodor thought that the easies way to understand how the brain works was to think of the brain as a type of computer.
True
Marr and Ungerleider both provided functional accounts of the visual system. What is the main difference in their approach to functional analysis?
Ungerleider and Mishkin focused more on the neural structures of the visual system
The discovery of neural networks is usually credited to....
Warren McCullough and Walter Pitts
What important concept in cognitive science is exemplified by Chomsky's transformational grammar?
algorithms
In the prisoner's dilemma, the TIT FOR TAT strategy... -is an evolutional stable strategy -involves simple calculations -requires the identification of defectors and cooperators -all of the above
all of the above
Which of the following is used in designing expert systems?
detailed info about the domain of the expert system
Cog scientists generally try to unify different field of cog science by reducing all phenomena to the level of neuroscience
false
The backpropagation algorithm corrects weights in the...
hidden units
According to Miller, the advantage of recoding in an immediate memory task is that it does which of the following?
increases the bits per chunk of information
What is one way to understand sentences in the language of thought?
purely semantical, in terms of how they represent the world
What are the 3 main functions of memory?
register, store, retrieve
Which of the following claims are included in the physical symbol systems hypothesis?
-thinking consists of the transformation of symbol structures according to rules -nothing is capable of intelligent action unless it is a physical symbol system -constructing an artificial mind requires constructing a physical symbol system
Who first proposed the working memory hypothesis?
Alan Baddeley
Which of Marr's 3 stages is missing from Ungerleider and Mishkin's account of the visual system?
algorithmic
Why, according to Cosmides and Tooby, are human beings particularly good at conditional reasoning about rules, obligations, and entitlements?
because of our specialized ability to detect cheaters and free riders
Why does Jerry Fodor think that propositional attitudes, and belief-desire psychology more generally, are so successful in explaining and predicting our behavior?
because propositional attitudes, as well as beliefs and desires, are real things in our minds
How does the ID3 algorithm measure info gain?
by calculating the reduction is entropy
Searle's Chinese room presents a challenge to the physical system hypothesis because the room is not producing complex behavior.
False
What features of logical inferences make them domain-general?
The subject matter of the inference is irrelevant
Why are artificial neural networks so useful for cog science?
because they provide a bridge between algorithm and implementation
What is one feature of cog systems that requires implementational level analysis?
determining the speed of computation
Marr concluded from Elizabeth Warrington's studies that the main job of the visual system is to "derive a representation of a shape." What level of analysis is Marr using to describe the visual system here?
computational
Tanya , a cognitive scientist, studies how memory works. Suppose she tells you that memory works to store and retrieve information. Which level of analyzing a cognitive system is she using to describe memory? (in Marr's terms)
computational
What was the model for early cog accounts of how info is transformed and transmitted in the human brain?
computers
What makes connectionist models of past tense learning different from Pinker and Prince's model?
connectionist models do not have explicit rules about past tense learning programmed into them
What do bridge principles linking different scientific theories help accomplish?
explanatory reduction
What does the physical symbol system hypothesis say about the format of info encoding?
info can only be encoded in either an analog or a digital format
What is one major theme in the work of Chomsky, Miller, and Broadbent?
information
What is one crucial idea that came out of applying the Turing machine to human minds?
information processing is algorithmic
Which of the following is not a distinction usually made about the function of memory?
internal and external memory
Which of the following features of introspective psychology did behaviorism object to?
introspective psychology claimed to be studying "inner" psychological states
What is one feature the cheater detection module possesses, according to Cosmides and Tooby?
it is domain-specific
One of the groups of rats in Tolman and Honzik's maze experiment were unrewarded for 10 days and then rewarded after that. What prediction would a strict behaviorist make about these rat's performance after the first day of rewarded behavior? (or Day 11 of the study)
it should be the same as the rewarded rats' performance on their first day of reinforcement
How does ID3 work so well for something like predicting the weather?
it uses the decision tree capable of decreasing the amount of entropy at each step
What types of algorithms are considered to be more biologically plausible than the backpropogation algorithm?
local algorithms
In ch4, conditional reasoning and the BOLD signal are identified as examples of what?
local integration
The sketchpad, the phonological loop, and the central executive are all subsystems of what system in the brain?
memory
Which cog science discipline is generally thought to be unhelpful in providing an integrated approach to cognition? -psychology -neuroscience -computer science -none of the above
none of the above
Which of these behaviors only requires serial processing, in a way behaviorists envisioned? -playing golf -writing a poem -planning one's day -none of the above
none of the above
According to Marr, which of the following are involved in giving a computational-level analysis of an information-processing system?
specifying what the system is doing-- its basic task
What prediction would someone make about Shepherd and Metzler's shape rotation experiment if they thought that all information processing in the brain was digital?
that all rotation process will take the same amount of time
What is one problem identified in the book with using the Stevens Law (in psychology) as a source of unification?
the Stevens Law is not explanatory
What kind of mental architecture does WHISPER use?
diagrammatic symbol structure
What does WHISPER's algorithm look for in preventing structures from collapse?
instability for individual blocks
Which model of information processing appears to be supported by Peterson et al.'s studies of single-word processing?
parallel
What characteristic feature of neural networks is illustrated by Gorman and Sejnowski's mine/rock detector?
pattern detection
ID3 works by transforming a complex symbol structure into a simple symbol structure. Based on this info alone, what sort of info processing model does ID3 utilize?
physical symbol system
Which of the following features of language is missing from Winograd's SHRDLU? -syntax -semantics -pragmatics -all of the above
pragmatics
Marr described 3 stages of information processing in the visual system. Which stage(s) did he think must be explained by his 3 levels for analyzing cognitive systems in order to prove a complete account of vision?
primal sketch, the 2.5D sketch, and the 3D sketch
Where does WHISPER's High Level Reasoner get its info?
a retina
