Brainscape's Knowledge GenomeTM

Browse over 1 million classes created by top students, professors, publishers, and experts.


See full index

MAN AND MACHINE > TASK 2 - CREATIVITY + AI > Flashcards

TASK 2 - CREATIVITY + AI Flashcards

1
Q

creativity

A

= ability to produce 1) novel and 2) useful ideas

  • quantity of ideas (fluency) + quality of ideas (originality)
    1a) p(sychological) idea: new to person who generated it
    1b) h(istorical) ideas: never occurred in history before; includes P idea
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
2
Q

types of creativity

- little C creativity

A

= individual capacities for doing things in novel ways (= everyday procedures)

  • much of it depends on emotional input, affective co-regulation and human agency (= complex operations of human consciousness)
  • matters most for developing computer based algorithms that aim to imitate human thinking patterns
  • -> not applicable to AI yet as the input is generated by humans –> nothing new generated by the AI
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
3
Q

types of creativity

- big C creativity

A

= generation of a product that is judged to be creative (1+2) by a suitably knowledgeable social group

  • function of groups
  • doesn’t assume that it must resolve extremely difficult problems
  • important for large scale innovations, paradigm shifts
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
4
Q

types of creativity

- pro-c creativity

A

= creativity that is developed within existing domains of knowledge but isn’t “paradigm busting”

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
5
Q

approaches to creativity

- individualist approach

A

= individual as unit of analysis

  • creativity = new mental combination that is expressed in the world
  • creativity = combination of mastered concepts in unique ways
  • externalisation = ideas and thinking must be expressed externally (must be something new based on something that already existed)
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
6
Q

sociocultural approach

A

= individual isn’t the sole component in the creative process; collective generation and acceptance of new ideas
- interrelated elements with big C: individual talent, existing info + knowledge; judgement by experts

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
7
Q

creative process

A
  1. problem finding: identify existing problem space
  2. knowledge acquisition about problem space: familiarise with potential solutions
  3. information collection
  4. incubation: allow time to process info; previously irrelevant info important ant for insight
  5. divergent thinking: formulation of multiple possible answers to a problem that has more than one possible solution; allows for generation of diverse new ideas
  6. combination
  7. convergent thinking. select most appropriate idea
  8. verification: final form
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
8
Q

creativity

- meta-level/facets

A
  1. unconventionality
  2. recognition of similarities and differences
  3. appreciation for ability to draw, write and compose
  4. flexibility to change directions
  5. willingness to question nodes and assumptions
  6. motivation and energy
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
9
Q

adaptive creativity

A

= creativity that (1) delivers surprising solutions to a problem + (2) changes the way we view the problem itself (adaptation)

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
10
Q

computational creativity

A

= model, simulate or replicate creativity using a computer

  • creative impetus comes from the machine
  • hybrid CC: joint impetus from human and machine
  • creativity arises from how an intelligent agent knowingly exploits or subverts the conventional pathways of a conceptual space of possibilities
  • incongruity and contradiction are opportunity for further search through problem space –> can be resolved to yield surprisingly meaningful results
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
11
Q

computational creativity

- goals

A

1) construct computer capable of human-level creativity
2) better understand human creativity + formulate an algorithmic perspective on creative human behaviour
3) design programs that can enhance human creativity without necessarily being creative themselves

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
12
Q

computational creativity

- creativity judgments

A

answer of computer is creative if it:

  1. has novelty and usefulness (for the individual or for society)
  2. rejects previously accepted ideas
  3. results from intense motivation and persistence
  4. comes from clarifying originally vague problem
    - AI must possess ability to filter its outputs for quality + articulate why its outputs may have interesting and unexpected value for its audience
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
13
Q

computational creativity

- pasticche

A

= creativity is due to programmer’s own creativity having been imprinted onto the algorithm

  • such systems explore a pre-defined sweet spot in the space of possible outputs; mimic creativity
  • take no risk; always produce well-formed outputs
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
14
Q

computational creativity

- combinational creativity

A

= combining links that were not linked beforehand

- easy for humans + hard for computers

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
15
Q

computational creativity

- exploratory creativity

A

= conceptual spaces can be explored

  • different combinations can be explored
  • AI best for this type; current state of algorithms
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
16
Q

computational creativity

- transformational creativity

A

= you have a conceptual space but the rules for the space are amendable (thinking outside the box)

  • genetic algorithm: computers are allowed to change their own code
  • -> better versions can be obtained through random mutations
  • -> evolutionary computation: = an initial set of candidate solutions is generated and iteratively updated
  • each new generation is produced by stochastically removing less desired solutions + introducing small random changes
  • creative algorithms are not advanced in transformational creativity
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
17
Q

creativity

- dual model of creativity

A

= cyclic movement between (1) process of associative generation of ideas + (2) process of evaluating these ideas against a benchmark of standards

  1. idea generation = bottom-up (generative) process associated with diffuse attention; default and control networks were negatively correlated
    - -> no central control needed
  2. idea evaluation = top-down (executive) process involving focused attention and cognitive control; correlation between default and control networks increased
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
18
Q

neurological substrates of creativity

- divergent thinking task

A

= generating several possible solutions to an open-ended problem

  • all networks were associated with divergent thinking
  • during divergent thinking, posterior cingulate cortex (PCC) shows increased coupling with dorsolateral prefrontal cortex (control network) and bilateral insula (salience network)
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
19
Q

neurological substrates of creativity

- default mode network

A

= generation process

  • vmPFC; PCC; midline and posterior inferior parietal regions; temporo-parietal areas
  • self-mediated mental activity; active in low cognitive control situations
  • spontaneous and self-generated thought (mind-wandering, autobiographical retrieval, deriving useful information from episodic LTM)
  • evaluation of emotional reactions
  • monitoring internally generated intuition
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
20
Q

neurological substrates of creativity

- salience network

A

= switching between networks

  • anterior insula (AI); anterior cingulate cortex (ACC)
  • conflict monitoring + attention
  • reward processing + emotion
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
21
Q

neurological substrates of creativity

- (central) executive control network

A
  • DLPFC; VLPFC; PPC (lateral parietal cortex); lateral prefrontal and anterior inferior parietal regions; occipital-parietal area
  • cognitive processes that require externally directed attention (working memory, relational integration, task-set switching)
  • -> monitoring and manipulation of information in WM
  • -> problem solving
  • -> decision making for goal-directed behaviour
  • selects ideas by evaluating efficacy of each idea + modifying to meet constraints of task-specific goals
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
22
Q

neurological substrates of creativity

- interplay of networks

A
  • graded antagonistic relation: if one is activated, the other one is deactivated
  • goal-directed and self-generated thought involve both default and control network
  • -> cooperation during top-down modulation of self-generated processes
  • competition during creative tasks: contributing factor to their coupling is the degree of goal-directedness of a given creative task
  • highly creative participants show increased coupling of default network regions with left inferior frontal gyrus (cognitive control)
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
23
Q

originality

A

= key feature of creativity; uniqueness of the response

  • creativity = measure of (1) quantity of ideas (fluency) + (2) their quality (originality)
  • original individuals:
    1) have more flexible associative network = are able to abandon more conventional thinking to connect different stimuli in new, diverse ways
    2) exhibit enhanced activation in ventral anterior cingulate cortex (vACC) + enhanced functional connectivity between ACC and left occipital-temporal area (evaluative process)
  • individual must at the same time inhibit more common automatic ideas (executive) + move flexibly (salience) to alternate associations that produce indirect associations (default mode)
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
24
Q

neurological substrates of originality

- default mode network

A
  • successfully generate new ways
  • activity in mPFC and PCC mediates ability to produce original ideas
  • -> mPFC central for originality NOT for fluency
  • frontal lobes: important for developing alternative strategies
  • frontal-parietal connections: inhibit activated associative similar networks while exciting the more weak remote association networks
25
Q

neurological substrates of originality

- salience network

A
  • more original people may be better at shifting between networks
26
Q

neurological substrates of originality

- (central) executive control network

A
  • successfully inhibit common ideas
27
Q

creativity

- systems model (humans)

A

= creativity is systemic and can be found when the three elements (cultural context, individual, social context) interact

  • instead of locating creativity in an individual’s production process
  • -> confluence/componential approach = more than one element must be present in order to produce a creative outcome
  • traditional psychological view of creativity as a mental process is an injustice to the complexity of creativity (needs to be expressed within cultural and social contexts)
28
Q

systems model

1. domain

A

= domain of knowledge = cultural context; element that encompasses existing traditions and conventions
- set of structures and rules (conventions, past works, standard way of working) that an individual learns and draws on to produce a creative product
- multitude of domains in a culture (chess, mathematics, cooking…)
+ encourages creative contributions if (1) domain has clear rules and procedures, (2) it is easier to learn past knowledge, (3) it is important within a culture, (4) domain information is accessible
- inhibits creative contributions if (1) symbol systems are not clear and accurate (2) knowledge is opaque and difficult to decipher

29
Q

systems model

2. individual

A

= person who understands and uses that knowledge to produce a novel change; person’s background, personal traits and motivation to produce
- enables person to generate creativity along with their ability to internalise the rules of the domain + the expectations of the field
- necessary but not sufficient to account for emergence of creativity
- accounts for lack in universal characteristics across all creative individuals: personal contribution varies according to states of other sub-systems
+ personal qualities contributing to creativity (1) curiosity, (2) intrinsic motivation, (3) cognitive abilities (divergent thinking), (4) other traits vary depending on field and historical period

30
Q

systems model

3. field

A

= social context that understands the domain and uses that knowledge to judge if an individual contribution is novel and appropriate; social group responsible for the verification of creativity
- all individuals who possess domain knowledge (network of experts with varying expertise, status, and power)
- power to determine the structure of the domain: preserve the domain, help to evolve the domain (selection of new content)
+ influences creative production if (1) it is proactive (actively seeking novelty), (2) using a broad filter (changes domain at faster rate), (3) it is connected to the rest of society
- does not influence creative production if (1) it is reactive (better able to preserve status quo), (2) uses a narrow filter (avoids chaos and collapse), (3) it is not connected to the rest of society

31
Q 
systems model (machines)
4. AI
A
  • based on Big C creativity
  • humans must first impart rule-based algorithms before an AI is capable of recognising patterns + generating something creative
    = field influences AI indirectly through individual
    –> BUT, as an AI can draw info directly from domain, it improves its predictions and begins to make new combinations increasingly independent from the original info provided by individual programmers
32
Q

Ai + creativity

- systems view

A
  • AI creativity is procedural
  • deep learning technologies allow to make intelligent predictions based on trial + error
  • current limitations
  • -> humans still have a lot of input
  • -> more complex forms of pattern recognition associated with creative combinatorial processes are still in its beginnings
  • -> absence of emotional arousal systems removes an important relational component that humans need for creativity (moral + emotional reasoning beyond utilitarianism cannot yet be replicated)
33
Q

AI + creativity

- problem space search

A

= state space search = successive configurations or states of an instance are considered, with the intention of finding a goal state with a desired property (used in computer science) = consider states of an instance to find the goal state with the desired property

  1. problems are modelled as a state space (= set of all possible configurations of a system)
  2. set of states forms a graph
  3. two states are connected if there is an operation that can be performed to transform the first state into the second
    a) S = set of all possible states
    b) A = set of possible action (regarding all of the state space)
    c) action(s) = function that establishes which action is possible to perform in a certain state
    d) results (s, a) = function that returns the state reached performing action a in state s
    e) cost (s, a) = costs of performing an action a in state s
34
Q

problem space search

- state-space search algorithms

A
  1. uninformed search methods: (1) traditional depth-first search, (2) breadth-first search, (3) iterative deepening, (4) lowest-cost-first search
    - used when we do not know information about the goal’s location
  2. heuristic search: (1) heuristic depth-first search, (2) greedy best-first search, (3) A* search
    - when we are taking into account information about the goal node’s location
35
Q

connectionism

A

= approach to the study of human cognition that utilises mathematical models, known as connectionist networks or artificial neural networks

36
Q

neural network theory

A

= tries to explain how mental processes could be explained by neurone-like components

1) processing units/nodes (neurones)
2) activation state (activated nodes correspond to what the focus of attention is)
3) connection pattern among nodes (excitatory or inhibitory)
4) input and output rules
5) learning rules (what fires together wires together)
6) environment for the network (the network should be split into modules devoted to specific tasks, similar to the way the brain works)

37
Q

theories of creativity

- selective retention and blind variation (campbell)

A

= with selective retention + blind variation one can learn to get a creative solution

  • blind variation = random ideas (e.g. sequences of letter strings)
  • selective retention = remembering what has been a good idea
  • nothing special about creativity: quasi-random thoughts might just by chance lead to something creative
  • -> everyone should be able to be creative
  • -> BUT, most people never have a creative idea or some have many
  • Darwinian evolution: random variants (= blind variation) that are fit are retained (= creative ideas are remembered)
  • remembering depends on connectivity between nodes
    a) nothing will happen if connections already are at maximum strength or if there aren’t any
    b) weak or indirect connection, we get a quick sequence of events
    1. arousal system bombards cortex with nonspecific activation
    2. activated nodes become extremely activated –> connection strength between them is quickly increased (= selective retention)
    3. “creative insight” or “conditioning”
38
Q

theories of creativity

- defocused attention (mendelsohn)

A

= differences in attentional capacity can explain individual differences in creativity (with a positive correlation between two factors)

  • to have a creative idea one has to be conscious of it –> combine elements in focus of attention
  • creative people: less focused attention (= defocused attention)
  • -> more nodes can be simultaneously activated in creative people than in uncreative people (regardless of whether nodes are in focal attention)
  • neural networks: consciousness can be divided in attention (= most activated nodes) + short-term memory (= nodes are activated less)
    1) preparation phase: attention too focused = few highly activated nodes dominate consciousness (encode ideas that are relevant)
    2) incubation: creative person activate more nodes that encode ideas that were thought out be irrelevant but turn out useful = nodes remain partially activated
    a. uncreative person = deactivate nodes encoding the problem and forget about it
39
Q

theories of creativity

- associative hierarchies (mendica)

A

= when there is less lateral inhibition, nodes are more equally active (= flat) and one probably finds more creative solutions

  • lateral inhibition = when a few nodes are highly activated, they inhibit other nodes to prevent them from becoming active
    1) attention focused: few nodes highly active and inhibit others from becoming activated (= high LH)
  • -> steep associative hierarchy = uncreative, stereotyped response
    2) attention defocused: activation more equitably spread out among nodes (= less LH)
  • less lateral inhibition –> allows short-term memory to be activated more –> relatively flat associative hierarchy gives opportunity to find creative solutions
  • -> flat associative hierarchy (= nodes are equally active) = creative, variable response
40
Q

theories of creativity

- primary process thinking (kris)

A

= creative people are more able to alternate between primary process and secondary process cognition

  • primary process-secondary process continuum: dimension along which consciousness varies
    1) primary: free associative, analogical –> creative inspiration thinking (makes discovery of new combinations more likely)
  • creative people: use primary processes to deal with all sorts of info (abstract or neutral material)
  • uncreative: can only use it for affect-laden, personally relevant ideas
  • neural network: primary process thinking corresponds to a state in which large numbers of nodes are weakly and about equally activated (= flat associative hierarchy)
    2) secondary: logical, goal-oriented, abstract –> verification thinking
41
Q

theories of creativity

- arousal (martindale)

A

= focus of attention is related to arousal –> as we decrease arousal thinking moves from secondary (high arousal) to primary processes
1) decreased arousal: behaviour more variable
= several nodes activated (associative hierarchy)
= primary process/defocused attention
2) increased arousal: behaviour more stereotyped
= one node gets all the activation
= secondary process/focused attention
- creative: less arousal while taking creativity test; low arousal during inspiration and high arousal during elaboration
–> more variable in their level of arousal
- uncreative: more arousal while taking creativity test; same level in intelligence test; show same high level arousal during both

42
Q

theories of creativity

- simulated annealing

A

= humans alternate between intense concentration (high arousal) + break (low arousal);

  • optimisation method; used to find an optimal (or close to) solution to an optimisation problem
  • basis for periodic annealing, allowing us to escape from local minima (= false starts/dead ends in our thinking)
  • high temperature = primary process/low arousal
  • low temperature = secondary process/high arousal
  • annealing is not different but more extreme in creative insight than normal problem solving
43
Q

optimisation

A

= selection of a best element (with regard to some criterion) from some set of available alternatives

  • optimisation problems = consists of maximising or minimising a real function by systematically choosing input values from an allowed set and computing the value of the function
  • arise in all quantitative disciplines (computer science)
  • local optimum of an optimisation problem = solution that is optimal (either maximal or minimal) within a neighbouring set of candidate solutions
  • global optimum = optimal solution amongst all possible solutions
44
Q

optimisation

- local search (hill climbing) methods

A
  1. start from an initial configuration and repeatedly move to an improving neighbourhood configuration
  2. mapped in search space as movement from an initial point to a local optimum
  3. in many cases, local optima deliver sub-optimal solutions to the global problem
  4. local search method must be modified to continue the search beyond local optimality
    - modifications include iterated local search, tabu search, simulated annealing
45
Q

optimisation

- global optimisation methods

A

(1) deterministic methods
(2) stochastic methods
- direct Monte-Carlo sampling (random simulations are used to find an approximate solution), stochastic tunneling, parallel tempering

46
Q

optimisation

- combinatorial algorithms

A

= algorithms that terminate in a finite number of steps

  • combinatorial optimisation = finding an optimal object from a finite set of objects
  • used on optimisation problems where set of feasible solutions are discrete or can be reduced to discrete, and also when the goal is to find the best solution
47
Q

optimisation

- meta-heuristics

A

= higher-level heuristic designed to find, generate, or select a heuristic that may provide a sufficiently good solution to an optimisation problem (especially with incomplete or imperfect information or limited computation capacity)

  • heuristics = any algorithm which is not guaranteed (mathematically) to find the solution, but which is nevertheless useful in certain practical situations
  • sample set of solutions (too large to be completely sampled)
  • do not guarantee that a globally optimal solution can be found on some problems
  • used for combinatorial optimisation (optimal solution is sought over a discrete search-space)
  • find good solutions with less computational efforts than other methods by searching over a large set of feasible solutions
  • examples of heuristics and meta-heuristics: (1) ant colony optimization, (2) simulated annealing, (3) tabu search, (4) evolutionary algorithms, (5) differential evolution, (6) swarm-based optimization algorithms, (7) memetic algorithms, (8) graduated optimization
48
Q

meta-heuristics

- properties

A
  1. strategies that guide the search process
  2. goal is to efficiently explore the search space in order to find near-optimal solutions
  3. techniques range from simple local search procedures to complex learning processes
  4. algorithms are approximate and non-deterministic
  5. not problem-specific
49
Q

optimisation

  • approaches
    1. local search vs. global search
A

a) local search strategy: improves local search heuristic ideas in order to find better solutions; (1) simulated annealing, (2) tabu search, (3) iterated local search, (4) variable neighbourhood search
b) global search strategy: population-based; (1) evolutionary computation, (2) genetic algorithms, amongst others

50
Q

optimisation

  • approaches
    2. single solution vs. population based
A

a) single solution approach: focus on modifying and improving a single candidate solution; (1) simulated annealing, (2) iterated local search, (3) variable neighbourhood search, (4) guided local search
b) population-based approach: maintain and improve multiple candidate solutions, often using population characteristics to guide the search; (1) evolutionary computation, (2) genetic algorithms, (3) particle swarm optimisation

51
Q

optimisation

  • approaches
    3. hybridisation vs. memetic algorithms
A

a) hybrid meta-heuristic: combines a meta-heuristic with other optimisation approaches (from mathematical programming, constraint programming, or machine learning); both components run concurrently and exchange information to guide the search
b) memetic algorithms: interaction between population-based approach (e.g. evolutionary) and separate individual learning or local improvement procedures for problem search

52
Q

optimisation

  • approaches
    4. parallel meta-heuristic
A

= uses techniques of parallel programming to run multiple meta-heuristic searches in parallel

53
Q

optimisation

  • approaches
    5. nature-inspired vs. metaphor-based
A

a) nature-inspired: algorithms inspired by natural systems; nature acts as a source of concepts, mechanisms and principles for designing of artificial computing systems to deal with complex computational problems; (1) simulated annealing, (2) evolutionary algorithms, (3) ant colony optimisation, (4) particle swarm optimisation

54
Q

AlphaGo

A

= uses value networks to evaluate board positions and policy networks to select moves

  • deep neural networks are trained by novel combination of supervised learning (from human expert games) and reinforcement learning (from games of self-play)
  • solve games by computing optimal value function in a search tree containing approximately bd moves, where b is the breadth (the games’ number of legal moves per position) and d is the depth (game length)
  • effective search space can be reduced by:
    1. position evaluation: shortening search tree at state s and replacing subtree below s by a value function that predicts the outcome from state s
    2. sampling actions: from a policy that is a probability distribution over possible moves in position a
  • -> combines the policy and value networks in an MCTS algorithm that selects actions by lookahead tree search
55
Q

AlphaGo

- monte carlo tree search

A

= uses Monte Carlo rollouts to estimate the value of each state in a search tree

  • as more simulations are executed, the search tree grows larger and the relevant values become more accurate
  • strongest GO programs are based on MCTS and enhanced by policies that are trained to predict human expert moves
56
Q

AlphaGo

- deep convolutional neural networks

A

= use many layers of “neurones,” each arranged in overlapping tiles to construct increasingly abstract, localised representations of an image

  • achieved unprecedented performance in visual domains (image classification, face recognition, playing Atari games)
  • AlphaGo uses similar architecture to reduce the depth and breadth of the search tree (evaluate position using value network, and sample actions using policy network)
57
Q

AlphaGo

- machine learning stages

A
  • neural networks are trained through several stages:
    1. supervised learning (SL) policy network: training takes place directly from expert human moves; provides efficient learning updates with immediate feedback
  • input s to policy network is a simple representation of board state
  • trained on randomly sampled state-action pairs (s, a)
  • 13-layer network: larger networks achieve better accuracy but are slower to evaluate during search
    2. reinforcement learning (RL) policy network: improves SL policy network by optimising final outcome of games of self-play; adjusts policy towards the correct goal of winning games rather than maximising predictive accuracy
  • identical in structure to SL policy network
  • play games between current policy network and a randomly selected previous iteration of the policy network
    3. value network: predicts winner of games played by RL policy network against itself; reinforcement learning of value network
  • involves similar architecture as policy network, but outputs a single prediction instead of a probability distribution
58
Q

AI + creativity

- critique

A

+ good at generating output which shares characteristics with input
+ allows humans to interact with data on higher abstraction + deal with high-level concepts, use new creative paradigms
+ great potential with new tools, create new genres to work with
- have difficulties working with meaningful parts of data and artworks –> creative responsibility still falls upon the user (pasticche)
- still dependent on humans and human-machine interaction –> human effort and agency matters for creativity
- will not be creative until they can interact with environment –> need internal model of the shared outside world

MAN AND MACHINE (3 decks)

Brainscape helps you reach your goals faster, through stronger study habits.
© 2024 Bold Learning Solutions. Terms and Conditions