Creative AI: a New Avenue for Semantic Web?

Tracking #: 2246-3459

Agnieszka Lawrynowicz

Responsible editor: 
Guest Editor 10-years SWJ

Submission type: 
Computational Creativity (or artificial creativity) is a multidisciplinary field, researching how to construct computer programs that model, simulate, exhibit or enhance creative behaviour. This vision paper explores a possible impact Semantic Web can have on the future of creativity, especially when it deals with AI creativity. Possible uses of SemanticWeb and semantic technologies are discussed, regarding three types of creativity: i) exploratory creativity, ii) combinational creativity, and iii) transformational creativity and relevant research questions. For exploratory creativity, how can we explore the limits of what is possible, while remaining bound by a set of existing domain axioms, templates, and rules, expressed with semantic technologies? To achieve a combinational creativity, how can we combine or blend existing concepts, frames, ontology design patterns, and other constructs, and benefit from cross-fertilization? Ultimately, can we use ontologies and knowledge graphs, which describe an existing domain with its constraints and, applying a meta-rule for transformational creativity, start dropping constraints and adding new constraints and see what emerges? Together with these new challenges, the paper also provides pointers to emerging and growing application domains of Semantic Web and knowledge graphs related to computational creativity and design: from food and cooking (recipe generation), fashion and style to program synthesis, software composition and scientific discovery.
Full PDF Version: 

Major Revision

Solicited Reviews:
Click to Expand/Collapse
Review #1
By Jérôme Euzenat submitted on 12/Jul/2019
Major Revision
Review Comment:

The paper is dedicated to explore the 'impact Semantic web can have on the future of creativity' adding 'especially when it deals with AI creativity'.
It starts by introducing creativity and computational creativity, then considers three types of creativity introduced by Margaret Boden and discussing how a semantic web approach could contribute to them.

Artificial creativity is a very exciting topic and relating it to the semantic web may have amazing outcomes.
However, the paper is somewhat disappointing.
To say it shortly: the really interesting part of the paper are the 'Potential for Semantic Web' paragraphs.
This is new and interesting and we would like to know more about that.
However, they end into questions when the reader is expecting to read answers, or at least directions.

Another frustrating aspect lays in the description level.
In order to understand how the semantic web could contribute to computational creativity, we do need to know what CC currently does, not really what creativity is, but how it is/could be computed.
This is missing from the paper.

The argumentation is also sometimes not very solid.
For instance, let consider the structure of the conclusion:
'In this position paper,... intelligent agents': It seems that all this is in Paper [1], thought they tried to avoid talking about AI which was not fashionable at the time.
'Therefore': not sure that this is the appropriate word here, especially that way many AI researchers have participated in SW development
'we explored... creativity': that's new! This is the topic of the paper. Except that there is a twist: 'the growing area of AI, namely in computational creativity' We jump from SW vision to AI and from AI to creativity without much justification, but that creativity is a part of AI. Most of the applications described in the Semantic web paper were 'modest', they require more problem solving than creativity. Showing how creativity would provide them a new dimension would be great, but this is not in this paper.
'We have briefly surveyed... knowledge bases.': outline of the paper
'We conclude... techniques.': a very general conclusion
'These include... and others': all this is good old AI and has nothing to do with the semantic web.

I do not object that there could be a place for semantic web resources into computational creativity, and vice versa.
On the contrary, I think that there is and looking into it is very interesting.
But it should be presented in a different, and clearly more concrete, way.

I do not know how to improve on this in such a limited space.
It may be more convincing to leave out the general introductions and to focus on a pair of examples, but to treat them more deeply, with the AI side and with the semantic web contribution, providing hints about this possible contribution.

The Scientific american paper was written as presenting applications that the semantic web would provide.
What would connecting semantic web and creativity provide?
This is not the only way to write such a paper, but since this one is taken as an example, this may be one inspiration source.

Other remarks:
- p2, 1.1: when deductive is opposed to inductive, and especially inductive reasoning, it is not related to statistical approaches.
There are relations, obviously, but one can perform symbolic or statistical, induction or deduction.
- line 41: what is the link between the two sentences? It is unclear.
- It seems that bisociations should be 2.2.1 instead of 2.3?
- It is surprising that in the combinatorial part, no mention is made of 'selection'. It seems a more challenging issue in creativity than circumscribing the solution space which is not usually considered creative.
- adequatly -> adequately
- CWA/OWA: the debate existed before the web and description logics were OWA before the semantic web.
- rather then -> rather than

Review #2
By Anna Lisa Gentile submitted on 19/Aug/2019
Major Revision
Review Comment:

This positional paper explores the field of Computational Creativity and suggests the possible impact that Semantic Web can have on this field.
It discusses three types of creativity: i) exploratory ii) combinational and iii) transformational and for each of them points the possible inputs from semantic technologies.
While the topic is very interesting I find that the "Semantic Web" input is quite limited and that the "Potential for Semantic Web" paragraphs offer a few links to available examples but without giving a strong message on the unique contribution of the SW technologies. So, for example: "Evolutionary computation has recently been used to generate recipes and using a graph-representation of the recipes [26]." - which leaves with the question "and...?". I find this true for many of the pointers, there is a lack of in-depth discussion of how SW help/can help/will help.
Also the open questions are way too general questions, without some sort of "hinted pathway".

Review #3
By Harald Sack submitted on 20/Aug/2019
Major Revision
Review Comment:

Paper Summary:
In this vision paper the author presents her ideas on how Semantic Web technologies could be applied to further exploit and progress computational creativity. The author follows the categorization of Boden and accordingly differentiates (i) exploratory creativity, (ii) computational creativity, as well as (iii) transformational creativity in combination with Wiggins’ unifying formalization of creativity as search. Thereby exploratory creativity refers to search within a predefined search space (i.e. a knowledge graph) to mine new (hidden) concepts of potential value. Combinational creativity refers to the combination of already known concepts within a search space to form new and potentially useful concepts. To achieve this, conceptual blending as well as bisociation are introduced, the application of which are suggested to be extended for ontologies. At last, transformational creativity is referred to by modification of the search space itself by “modifying rules and constraints”. However, transformational creativity is further exemplified on the example of word embeddings including recent approaches to improve expressivity ba using Non-Euclidian Spaces. In the end, the trivial conclusion is drawn that “to achieve its full potential the Semantic Web must be accompanied with valuable applications”, and potential applications for future research are pointed out.

General Comments:
In general, the topic of computational creativity for the Semantic Web is - from the reviewer’s perspective - new and rather interesting. To apply the potential of Knowledge Graphs, ontologies, and rules to the different types of creativity pointed out in the paper seem to be rather promising. However, in the paper only research questions have been phrased and some already existing approaches have been used to further illustrate the underlying concepts, whereas ideas regarding the solution on how to address or even solve the raised research questions are missing. Especially on transformational creativity the relation of latent knowledge representations as e.g. word embeddings to explicit knowledge representations and how to leverage latent representations to achieve/evoke transformational creativity could have been illustrated in more detail.

Detailed Comments:
p.1, c.2, l.36-51:Starting with the first groundbreaking Semantic Web article of Tim Berners-Lee the connection to knowledge graphs is directly drawn via the earlier approach of semantic networks. Here the role of RDF (esp. its graph interpretation) & Linked Data for the popularity of knowledge graphs could have been emphasized more.

P.2, c.1, l.27-44: The author claims a development in Semantic Web from deductive reasoning over inductive reasoning (both analytical approaches) towards abductive reasoning (synthetic and thus generative approaches). This hypothesis should be supported by further evidence in terms of references. In addition the paragraph would be much more helpful, if it could be endorsed with examples and further explanations.(Word) Embeddings are not directly related to Semantic Web technology. They suffer from the same polysemic problems as natural language does and are a latent (implicit) representation of knowledge. However, there are approaches to embed different word senses into separate vectors, as e.g. polysemic aware word embeddings or distributed multi sense embeddings, which would serve better for the task.

P.2, c.2, l.15-38. In this paragraph you stress the example of Ada Lovelace (1815-1852), who could not imagine the computer to be creative. This is a nice anecdote. However, I don’t think that the experiences of one of the first programmers in history are sufficient to motivate contemporary reservations against the possibility of “computational creativity”. Please also note that Ada Lovelace being referred to as “the first programmer” still is a subject of dispute [1].

P.2, c.2, l.44-51: Please maintain the same sequence for the three types of creativity all over the paper.

P.3,c.2,l.1-9. An example of the mentioned refinement operators would be helpful. Also the question of quality measures for creativity could have been further discussed (wrt. Usefulness, novelty, serendipity, etc.)

P. 4, c.2, l.20-35. How can generic spaces be created (also manually). Please give an example. Concerning the “large number of possible combinations to create blends”, here also a quality assessment to come up with some kind of ranking for the most promising candidates should be discussed.

P.5, c.1, Fig.4 This figure is not rather helpful. Please provide node labels to better understand the concept of BisoNet.

P.5, c.1, Fig.45-46. How are bisociations helpful to discover analogies and to create metaphors. Please elaborate on that and give a justification.

P.5,c.2,l.17-22 For transformational creativity no example is given as for the other creativity types.

P.5,c.2,l.32ff Again plain word embeddings are used this time to illustrate(?) transformational creativity, while an improvement in performance is pointed out for non-Euclidian vector spaces. To better transform the (explicit) search space of a knowledge graph, probably the concept of graph embeddings would be better suited. However, as already pointed out, an improvement of plain word embeddings would simply be to use polysemic aware word embeddings instead that take one vector per meaning. Overall the entire paragraph is not clear to me. How can a potential “transformation” to embedding space leverate transformational creativity for knowledge graphs? Plase be more explicit.

P.6, c.2, l.21-23. The paper only raises research questions (which are helpful). However, it does not really “explore under-explored and rising opportunities”. For that I would have expected some ideas and directions in a vision paper.

P.6, c.2, l.27. I missed the aspect on reasoning in general (besides the hypothesis in the beginning that there has been a shift from deductive over inductive to abductive reasoning)..

Minor Issues:
P.2, c.2, l.10: after “etc.” there are overall 2 periods. This should only be one period.
P.4, c.1, Fig.2: “housebot” -> “houseboat”

[1] Simonite, Tom (24 March 2009). "Short Sharp Science: Celebrating Ada Lovelace: the 'world's first programmer'". New Scientist. Archived from the original on 27 March 2009.