Computational creativity

Computational creativity (also known as artificial creativity, mechanical creativity, creative computing or creative computation) is a multidisciplinary endeavour that is located at the intersection of the fields of artificial intelligence, cognitive psychology, philosophy, and the arts (e.g., computational art as part of computational culture).

The goal of computational creativity is to model, simulate or replicate creativity using a computer, to achieve one of several ends:

* To construct a program or computer capable of human-level creativity.
* To better understand human creativity and to formulate an algorithmic perspective on creative behavior in humans.
* To design programs that can enhance human creativity without necessarily being creative themselves.

The field of computational creativity concerns itself with theoretical and practical issues in the study of creativity. Theoretical work on the nature and proper definition of creativity is performed in parallel with practical work on the implementation of systems that exhibit creativity, with one strand of work informing the other.

The applied form of computational creativity is known as media synthesis.

Theoretical issues

If eminent creativity is about rule-breaking or the disavowal of convention, how is it possible for an algorithmic system to be creative? In essence, this is a variant of Ada Lovelace's objection to machine intelligence, as recapitulated by modern theorists such as Teresa Amabile. If a machine can do only what it was programmed to do, how can its behavior ever be called ''creative''?

Indeed, not all computer theorists would agree with the premise that computers can only do what they are programmed to do—a key point in favor of computational creativity.

Defining creativity in computational terms

Because no single perspective or definition seems to offer a complete picture of creativity, the AI researchers Newell, Shaw and Simon developed the combination of novelty and usefulness into the cornerstone of a multi-pronged view of creativity, one that uses the following four criteria to categorize a given answer or solution as creative:
# The answer is novel and useful (either for the individual or for society)
# The answer demands that we reject ideas we had previously accepted
# The answer results from intense motivation and persistence
# The answer comes from clarifying a problem that was originally vague

Whereas the above reflects a top-down approach to computational creativity, an alternative thread has developed among bottom-up computational psychologists involved in artificial neural network research. During the late 1980s and early 1990s, for example, such generative neural systems were driven by genetic algorithms. Experiments involving recurrent nets were successful in hybridizing simple musical melodies and predicting listener expectations.

In his book, Superhuman Creators, Al Byrd argues that the primary source of creativity in humans and other animals is affordance awareness – awareness of the action possibilities in an environment. Superhuman creativity can be achieved by increasing the affordance awareness of artificial entities dramatically, and integrating that awareness tightly with the systems capable of capitalizing on the action possibilities.

Machine learning for computational creativity

While traditional computational approaches to creativity rely on the explicit formulation of prescriptions by developers and a certain degree of randomness in computer programs, machine learning methods allow computer programs to learn on heuristics from input data enabling creative capacities within the computer programs. Especially, deep artificial neural networks allow to learn patterns from input data that allow for the non-linear generation of creative artefacts. Before 1989, artificial neural networks have been used to model certain aspects of creativity. Peter Todd (1989) first trained a neural network to reproduce musical melodies from a training set of musical pieces. Then he used a change algorithm to modify the network's input parameters. The network was able to randomly generate new music in a highly uncontrolled manner. In 1992, Toddextended this work, using the so-called distal teacher approach that had been developed by Paul Munro, Paul Werbos, D. Nguyen and Bernard Widrow, Michael I. Jordan and David Rumelhart. In the new approach there are two neural networks, one of which is supplying training patterns to another.
In later efforts by Todd, a composer would select a set of melodies that define the melody space, position them on a 2-d plane with a mouse-based graphic interface, and train a connectionist network to produce those melodies, and listen to the new "interpolated" melodies that the network generates corresponding to intermediate points in the 2-d plane.

Key concepts from the literature

Some high-level and philosophical themes recur throughout the field of computational creativity.

Important categories of creativity

Margaret Boden refers to creativity that is novel ''merely to the agent that produces it'' as "P-creativity" (or "psychological creativity"), and refers to creativity that is recognized as novel ''by society at large'' as "H-creativity" (or "historical creativity"). Stephen Thaler has suggested a new category he calls "V-" or "Visceral creativity" wherein significance is invented through neural mapping to raw sensory inputs to a Creativity Machine architecture, with the "gateway" nets perturbed to produce alternative interpretations, and downstream nets shifting such interpretations to fit the overarching context. An important variety of such V-creativity is consciousness itself, wherein meaning is reflexively invented to activation turnover within the brain. Value driven creativity gives more freedom and autonomy to the AI system.

Exploratory and transformational creativity

Boden also distinguishes between the creativity that arises from an exploration within an established conceptual space, and the creativity that arises from a deliberate transformation or transcendence of this space. She labels the former as ''exploratory creativity'' and the latter as ''transformational creativity'', seeing the latter as a form of creativity far more radical, challenging, and rarer than the former. Following the criteria from Newell and Simon elaborated above, we can see that both forms of creativity should produce results that are appreciably novel and useful (criterion 1), but exploratory creativity is more likely to arise from a thorough and persistent search of a well-understood space (criterion 3) -- while transformational creativity should involve the rejection of some of the constraints that define this space (criterion 2) or some of the assumptions that define the problem itself (criterion 4). Boden's insights have guided work in computational creativity at a very general level, providing more an inspirational touchstone for development work than a technical framework of algorithmic substance. However, Boden's insights are also the subject of formalization, most notably in the work by Geraint Wiggins.

Generation and evaluation

The criterion that creative products should be novel and useful means that creative computational systems are typically structured into two phases, generation and evaluation. In the first phase, novel (to the system itself, thus P-Creative) constructs are generated; unoriginal constructs that are already known to the system are filtered at this stage. This body of potentially creative constructs is then evaluated, to determine which are meaningful and useful and which are not. This two-phase structure conforms to the Geneplore model of Finke, Ward and Smith, which is a psychological model of creative generation based on empirical observation of human creativity.

Combinatorial creativity

A great deal, perhaps all, of human creativity can be understood as a novel combination of pre-existing ideas or objects. Common strategies for combinatorial creativity include:
*Placing a familiar object in an unfamiliar setting (e.g., Marcel Duchamp's ''Fountain'') or an unfamiliar object in a familiar setting (e.g., a fish-out-of-water story such as ''The Beverly Hillbillies'')
*Blending two superficially different objects or genres (e.g., a sci-fi story set in the Wild West, with robot cowboys, as in ''Westworld'', or the reverse, as in '' Firefly''; Japanese haiku poems, etc.)
*Comparing a familiar object to a superficially unrelated and semantically distant concept (e.g., "Makeup is the Western burka"; "A zoo is a gallery with living exhibits")
*Adding a new and unexpected feature to an existing concept (e.g., adding a scalpel to a Swiss Army knife; adding a camera to a mobile phone)
*Compressing two incongruous scenarios into the same narrative to get a joke (e.g., the Emo Philips joke "Women are always using men to advance their careers. Damned anthropologists!")
*Using an iconic image from one domain in a domain for an unrelated or incongruous idea or product (e.g., using the Marlboro Man image to sell cars, or to advertise the dangers of smoking-related impotence).
The combinatorial perspective allows us to model creativity as a search process through the space of possible combinations. The combinations can arise from composition or concatenation of different representations, or through a rule-based or stochastic transformation of initial and intermediate representations. Genetic algorithms and neural networks can be used to generate blended or crossover representations that capture a combination of different inputs.

Conceptual blending

Mark Turner and Gilles Fauconnier propose a model called Conceptual Integration Networks that elaborates upon Arthur Koestler's ideas about creativity as well as work by Lakoff and Johnson, by synthesizing ideas from Cognitive Linguistic research into mental spaces and conceptual metaphors. Their basic model defines an integration network as four connected spaces:

*A first input space (contains one conceptual structure or mental space)
*A second input space (to be blended with the first input)
*A ''generic space'' of stock conventions and image-schemas that allow the input spaces to be understood from an integrated perspective
*A ''blend space'' in which a selected projection of elements from both input spaces are combined; inferences arising from this combination also reside here, sometimes leading to emergent structures that conflict with the inputs.

Fauconnier and Turner describe a collection of optimality principles that are claimed to guide the construction of a well-formed integration network. In essence, they see blending as a compression mechanism in which two or more input structures are compressed into a single blend structure. This compression operates on the level of conceptual relations. For example, a series of similarity relations between the input spaces can be compressed into a single identity relationship in the blend.

Some computational success has been achieved with the blending model by extending pre-existing computational models of analogical mapping that are compatible by virtue of their emphasis on connected semantic structures. In 2006, Francisco Câmara Pereira presented an implementation of blending theory that employs ideas both from GOFAI and genetic algorithms to realize some aspects of blending theory in a practical form; his example domains range from the linguistic to the visual, and the latter most notably includes the creation of mythical monsters by combining 3-D graphical models.

Linguistic creativity

Language provides continuous opportunity for creativity, evident in the generation of novel sentences, phrasings, puns, neologisms, rhymes, allusions, sarcasm, irony, similes, metaphors, analogies, witticisms, and jokes. Native speakers of morphologically rich languages frequently create new word-forms that are easily understood, and some have found their way to the dictionary. The area of natural language generation has been well studied, but these creative aspects of everyday language have yet to be incorporated with any robustness or scale.

Hypothesis of creative patterns

In the seminal work of applied linguist Ronald Carter, he hypothesized two main creativity types involving words and word patterns: pattern-reforming creativity, and pattern-forming creativity. Pattern-reforming creativity refers to creativity by the breaking of rules, reforming and reshaping patterns of language often through individual innovation, while pattern-forming creativity refers to creativity via conformity to language rules rather than breaking them, creating convergence, symmetry and greater mutuality between interlocutors through their interactions in the form of repetitions.

Story generation

Substantial work has been conducted in this area of linguistic creation since the 1970s, with the development of James Meehan's TALE-SPIN
system. TALE-SPIN viewed stories as narrative descriptions of a problem-solving effort, and created stories by first establishing a goal for the story's characters so that their search for a solution could be tracked and recorded. The MINSTREL system represents a complex elaboration of this basic approach, distinguishing a range of character-level goals in the story from a range of author-level goals for the story. Systems like Bringsjord's BRUTUS elaborate these ideas further to create stories with complex inter-personal themes like betrayal. Nonetheless, MINSTREL explicitly models the creative process with a set of Transform Recall Adapt Methods (TRAMs) to create novel scenes from old. The MEXICA model of Rafael Pérez y Pérez and Mike Sharples is more explicitly interested in the creative process of storytelling, and implements a version of the engagement-reflection cognitive model of creative writing.

Metaphor and simile

Example of a metaphor: ''"She was an ape."''

Example of a simile: ''"Felt like a tiger-fur blanket.''"
The computational study of these phenomena has mainly focused on interpretation as a knowledge-based process. Computationalists such as Yorick Wilks, James Martin, Dan Fass, John Barnden, and Mark Lee have developed knowledge-based approaches to the processing of metaphors, either at a linguistic level or a logical level. Tony Veale and Yanfen Hao have developed a system, called Sardonicus, that acquires a comprehensive database of explicit similes from the web; these similes are then tagged as bona-fide (e.g., "as hard as steel") or ironic (e.g., "as hairy as a bowling ball", "as pleasant as a root canal"); similes of either type can be retrieved on demand for any given adjective. They use these similes as the basis of an on-line metaphor generation system called Aristotle that can suggest lexical metaphors for a given descriptive goal (e.g., to describe a supermodel as skinny, the source terms "pencil", "whip", "whippet", "rope", " stick-insect" and "snake" are suggested).

Analogy

The process of analogical reasoning has been studied from both a mapping and a retrieval perspective, the latter being key to the generation of novel analogies. The dominant school of research, as advanced by Dedre Gentner, views analogy as a structure-preserving process; this view has been implemented in the structure mapping engine or SME, the MAC/FAC retrieval engine (Many Are Called, Few Are Chosen), ACME ( Analogical Constraint Mapping Engine) and ARCS ( Analogical Retrieval Constraint System). Other mapping-based approaches include Sapper, which situates the mapping process in a semantic-network model of memory. Analogy is a very active sub-area of creative computation and creative cognition; active figures in this sub-area include Douglas Hofstadter, Paul Thagard, and Keith Holyoak. Also worthy of note here is Peter Turney and Michael Littman's machine learning approach to the solving of SAT-style analogy problems; their approach achieves a score that compares well with average scores achieved by humans on these tests.

Joke generation

Humour is an especially knowledge-hungry process, and the most successful joke-generation systems to date have focussed on pun-generation, as exemplified by the work of Kim Binsted and Graeme Ritchie. This work includes the JAPE system, which can generate a wide range of puns that are consistently evaluated as novel and humorous by young children. An improved version of JAPE has been developed in the guise of the STANDUP system, which has been experimentally deployed as a means of enhancing linguistic interaction with children with communication disabilities. Some limited progress has been made in generating humour that involves other aspects of natural language, such as the deliberate misunderstanding of pronominal reference (in the work of Hans Wim Tinholt and Anton Nijholt), as well as in the generation of humorous acronyms in the HAHAcronym system of Oliviero Stock and Carlo Strapparava.

Neologism

The blending of multiple word forms is a dominant force for new word creation in language; these new words are commonly called "blends" or "portmanteau word). Tony Veale has developed a system called ZeitGeist that harvests neological headwords from Wikipedia and interprets them relative to their local context in Wikipedia and relative to specific word senses in WordNet. ZeitGeist has been extended to generate neologisms of its own; the approach combines elements from an inventory of word parts that are harvested from WordNet, and simultaneously determines likely glosses for these new words (e.g., "food traveller" for "gastronaut" and "time traveller" for " chrononaut"). It then uses Web search to determine which glosses are meaningful and which neologisms have not been used before; this search identifies the subset of generated words that are both novel ("H-creative") and useful.

A corpus linguistic approach to the search and extraction of neologism have also shown to be possible. Using Corpus of Contemporary American English as a reference corpus, Locky Law has performed an extraction of neologism, portmanteau

HOME


Content is Copyleft
Website design, code, and AI is Copyrighted (c) 2014-2017 by Stephen Payne


Consider donating to Wikimedia



As an Amazon Associate I earn from qualifying purchases