CH. 9 AI Flashcards
cognitive model
a computational system used to describe information processing aspects of cognition.
turing test
a behavioral test of machine intelligence. If a machine communicating with a human judge via a text interface cannot distinguish it from a human, then the machine passes the test. Some believe that passing the Turing test is an adequate definition of machine intelligence.
weak AI
the view that some machines can behave intelligently.
strong AI
the view that at least some machines that behave intelligently really
are thinking.
symbolic logic
an approach to automated reasoning in which knowledge is
encoded as sequences of symbols called sentences, and new knowledge arises by manipulating and recombining components of sentences according to a set
of formal rules.
predicate calculus
a symbolic logic that supports the representation of general
knowledge about objects, properties, and relationships. Also called first-order
logic.
computability theory
a formal account of computation, providing a means to discriminate between functions that can be computed and those that cannot.
turing machine
an idealized machine equipped with an infinite memory that manipulates symbols in accordance with rules.
algorithm
a finite sequence of formal instructions that can be executed by a
computer.
computational theory of mind
the idea that the mind is a computational
system that converts sensory inputs into symbols and process these symbols in
accordance with rules in order to produce behavior
symbolic systems/symbolic AI
an approach to AI that focuses on formal
rules applied to explicit symbol structures. This is the most popular computational
theory of mind and is closely related to the representational theory of mind. Associated with the view that the mind is to the brain as software is to hardware
physical symbol system hypothesis
the claim that the ability of a machine to
manipulate symbolic structures is foundational for general intelligence. “A physical symbol system has the necessary and sufficient means for general intelligent
action” (Newell & Simon, 1976).
sentential knowledge representation
an approach to knowledge representation inspired by sentences in natural languages. Propositions are represented as
sequences of symbols. Symbolic logic uses sentential knowledge representations.
compositional knowledge representation
an approach to knowledge representation in which the meaning of a complex representation is a simple function
of the meanings of its parts.
sound
a property of an automated reasoning system that guarantees that any
proposition inferred by the system is justified by the system’s body of knowledge.
complete
a property of an automated reasoning system that guarantees that any proposition that is justified by a body of knowledge can be inferred by the
system when given that knowledge.
locality
a property of automated reasoning systems using symbolic logic that
allows individual conclusions to be reached with certainty using only a small
amount of relevant knowledge.
detachment
a property of automated reasoning systems using symbolic logic
that allows the justifications for conclusions to be discarded once the conclusions
are determined to be true
truth-functionality
a property of automated reasoning systems using symbolic
logic that makes the meaning of a complex sentence a simple function of the
meanings of the components of the sentence.
bounded rationality
automated reasoning that takes into account limitations on
computational resources such as memory and processing time
heuristic
an approximate method that is not universally valid but may still support strong performance
planner
an AI system that outputs action sequences expected to result in the
attainment of specified goals.
frames
an approach to symbolic knowledge representation in which knowledge
concerning common situations (e.g., dining at a restaurant) is aggregated and
indexed by situation.
semantic networks
an approach to knowledge representation using mathematical graphs, with nodes in the graph representing objects or concepts and
connections between nodes representing relationships.
reasoning under certainty
automated reasoning when propositions are
believed to be true with graded levels of certainty.
probability theory
a mathematical framework for describing stochastic processes and quantifying uncertainty.
utility theory
a mathematical framework for describing and quantifying preferences between options. More preferred outcomes are associated with higher utility than less preferred outcomes.
bayesian decision theory
the combination of probability theory and utility theory, providing mathematical methods for evaluating possible actions when there is uncertainty about the outcomes that will be produced by those
actions.
expert systems
AI systems with capabilities associated with an area of professional expertise, such as medical diagnosis, financial market prediction, and
application of case law.
knowledge-based systems
expert systems that depend upon domain-specific knowledge that is explicitly represented using symbolic structures, such as logical sentences
functionalism
the view that mental states can be defined in terms of functional
relationships between sensory inputs, the internal states of a system, and behavioral outputs.
multiple realization
in philosophy of mind, the view that types of mental states
can be realized in (that is, produced by) many different kinds of physical systems.
Desire and hunger are types of mental states that can be realized in human brains, octopus brains, computer hardware, etc.
language of thought
a hypothesized system of mental representations with a structure that corresponds to the grammar of language. According to the language of thought hypothesis, cognition can be understood in terms of sentences unfolding in this mental language.
productivity
the ability to produce an unlimited number of meaningful propositions.
systematicity
the existence of regularities in the set of propositions considered
meaningful based on the compositional structure of the propositions.
levels of analysis
distinctive ways to study the mind, focused on specific phenomena and using specific methodologies. In cognitive modeling, the neural level of analysis is often distinguished from the computational level of analysis, where the former is studied by neuroscience and the latter is studied by cognitive psychology.
theoretical autonomy
the view, associated with symbolic AI, that the computational level of analysis need not draw on or be aware of developments
in the neural level of analysis, in the same way that software engineers need not be aware of how computer hardware works in order to write computer programs.
connectionism
an approach to AI that is inspired by how the brain works; connectionism emphasizes the production of intelligent behavior by processing
inputs through the weights and connections of an artificial neural network that
is trained by exposure to inputs rather than being programmed.
artificial neural networks
a kind of AI system inspired by the brain, involving large numbers of interconnected simple processing elements.
graceful degradation
a property of AI systems in which performance declines only mildly in the face of small errors in the system or small amounts of damage.
distributed representations
an approach to knowledge representation in
which the meaning of a representation depends on the pattern of activation over a collection of processing elements, in contrast to meaning being assigned
to the activation of individual processing elements.
deep learning
an approach to machine learning that makes use of deep neural
networks.
deep neural networks
artificial neural networks with a large number of successive layers of processing elements mediating between inputs and outputs.
reinforcement learning
a framework for machine learning in which performance feedback only appears as infrequent rewards and punishments.
deep reinforcement learning
an approach to machine learning involving
performance feedback in the form of infrequent rewards and punishments, incorporating deep neural networks.
computational cognitive neuroscience
an approach to cognitive modeling that describes the information processing properties of cognition in terms of the
physical properties of the brain.
cognitive architectures
formal frameworks for producing cognitive models, embodying commitments about cognition as constraints on the structure of such
models.
Bayesian belief networks
a method for representing probabilistic relationships
between variables, supporting reasoning under uncertainty
predictive processing
a theory of mind and brain according to which the brain maintains a system of internal models that constantly predict inputs (from
the external world or from other internal models). Errors between predicted and observed inputs are used to update the models, so that, over time, predictions become increasingly accurate. Neural and cognitive activity can be understood largely as constructions of these models.
dynamical systems
systems with properties that change over time, often in complex ways. Dynamical systems theory provides formal tools for describing
and analyzing such systems.
reactive control
an approach to robot control lacking explicit representations
of knowledge about the environment or about the effects of considered actions. Useful behaviors arise from the interplay between how the robot reacts to the sensed environment and how the environment responds to robot actions.
explanatory pluralism
the view that different approaches to the study of cognition framed at different spatial and temporal scales and drawing on diverse methodologies can complement one another. Contrasted with theoretical autonomy
