Interpretable AI as Curation Akinori Abe12 , 1 Division of Behavioral Science, Faculty of Letters, Chiba University 1-33 Yayoicho, Inageku, Chiba 263-8522, JAPAN 2 Dwango Artificial Intelligence Laboratory, Tokyo JAPAN ave@ultimaVI.arc.net.my, ave@chiba-u.jp Abstract Their main task of curator is a curatorial task, which is mul- tifaceted. Curator comes from a Latin word “cura” which “Interpretable AI” is an artificial intelligence (AI) whose ac- means cure. Then originally it used for a person who take tions can be easily understood by humans. This is a theme of the workshop. For this theme, I will show several possi- care of a cultural heritage. bilities. For the “Interpretable AI,” the key factor is curation. In the report by American Association of Museums Cu- One of the main function of curation is to show something rators Committee (AAMCC) [AAMCC, 2009], they pointed (person’s concept etc.) to other persons. In addition curation out “curators are highly knowledgeable, experienced, or ed- should be conducted according to users’ cognitive level. From ucated in a discipline relevant to the museum’s purpose or this viewpoint in this paper I discuss two types curations and mission. Curatorial roles and responsibilities vary widely the effectiveness of curation. In addition, from the cognitive within the museum community and within the museum it- bias, I will also discuss two types curations. self, and may also be fulfilled by staff members with other titles.” Then they showed the definition of curator as follows; Introduction • Remain current in the scholarly developments within their “Interpretable AI” is an artificial intelligence (AI) whose ac- field(s); conduct original research and develop new schol- tions can be easily understood by humans. This is a theme arship that contributes to the advancement of the body of of the workshop. For this theme, I will show several possi- knowledge within their field(s) and within the museum bilities. profession as a whole. For the “Interpretable AI,” the key factor is curation. De- • Make recommendations for acquiring and deaccessioning tails of curation are shown in the next section. But briefly objects in the museum collection. illustrated, one of the main function of curation is to show • Assume responsibility for the overall care and develop- something (person’s concept etc.) to other persons. In addi- ment of the collection, which may include artifacts, fine tion curation should be conducted according to users’ cog- art, specimens, historic structures, and intellectual prop- nitive level. For this, as interpretable AI strategies, I will erty. show a dementia person care system inspired by affordance • Advocate for and participate in the formulation of insti- and a new curation system. Where “Interpretable AI” can tutional policies and procedures for the care of the col- be achieved. In addition from the cognitive bias, I will also lection that are based on accepted professional standards discuss the the dementia person care system inspired by af- and best practices as defined by AAM, CurCom, and other fordance and the new curation system. relevant professional organizations. In the following, I will show two types of interpretable AI • Perform research to identify materials in the collection strategy based curation. and to document their history. • Interpret the objects belonging or loaned to the museum. Curatioin • Develop and organize exhibitions. Recently, in several situations, the word “curation” has been • Contribute to programs and educational materials. used. For instance, in the marketing strategies, for fashion • Advocate and provide for public use of the collection. shows, and for a DJ etc. This section reviews various types • Develop or contribute to monographs, essays, research pa- of curation. Actually curatorial task is usually used for tasks pers, and other products of original thought. in (art) museum. In the previous papers I several times re- viewed curation, so I introduce very typical curation which • Represent their institution in the media, at public gather- makes the visitors think themselves. ings, and at professional conferences and seminars. • Remain current on all state, national, and international (General) curation laws as they pertain to objects in the museum collection. There is at least a person who is responsible as “curator” In addition, AAMCC showed curatorial responsibilities as in (special) exhibitions, galleries, archive, or (art) museums. follows; A. Research, Scholarship, and Integrity more surprising and significant than the achievements B. Interpretation of a specialized scholarship in its own field of action, C. Acquisition, Care, and Disposal an art raised to the status of an enigma like Bacon’s D. Collection Access and Use seems to require the gamble of provoking these par- E. Replication of Objects in the Collection allels. And since at the time, and again because of its qualitative greatness, Caravaggio’s art deserves a simi- Thus curators have responsibilities for various aspects of lar provocation, the juxtaposition thus satisfies a legiti- exhibition activities. However, the most important activity mate aesthetic desire. On the other hand, the juxtaposi- will be a plan of exhibition. For that the above activities tion is a modest and prudent solution, not so much for such as research, interpretation and acquisition are neces- demonstrating, but for offering the attribute of “genius” sary. They should properly exhibit a truth which is result of — which the expressive common language attributes to their researches and interpretations. the great artist of the past — opportunities to manifest In addition, their researches and interpretations properly itself. And the juxtaposition is induced by the Galleria communicate to the visitors. In the next section, however, I Borghese itself, one of the most sensitive spaces with introduce the curation example with implicit communication the simultaneous presence of genius.” of curator’s researches and interpretations. Besides the importance in aesthetics and philosophy, I Exhibition “Bacon and Caravaggio” think the most important point is that “There are parallels that appear by themselves to the visitor’s sensibility and are An exhibition “Bacon and Caravaggio” was held in Museo e not imposed by a theory of the curator.” That is, though ac- Galleria Borghese, Roma, Italy during October 2 2009 and tually a curator has a certain philosophy, he/she does not January 24 2010. insist his/her philosophy but audiences will be able to dis- The display policy of this exhibition is rather different cover additional meanings as well as the curator’s intended from the general special exhibition. First, the special exhibi- philosophy. tion was not separated from the space for permanent collec- This type of curation used to be rare, but recently the im- tions. Of course several Caravaggio’s works were exhibited portance of this type of curation has been recognised. in their original places. The other Caravaggio’s works and From the viewpoint of “Interpretable AI,” this type of Bacon’s works were exhibited between the permanent col- curation is meaningful. In fact the curation does not seem lections. For this type exhibition, usually exhibition is ed- to make effort to make visitors understand what should in- ucational and two painters are compared in various point, form. Furthermore this type of curation reduces detailed ex- for instance days and society painters lived. Before arriving planations. However the merit of this type of curation is at the exhibition, my expectation was that it would be an make visitors think themselves. That is visitors try to inter- exhibition to address the contrast between the drawing poli- pret/understand as many as possible, and they can under- cies of Bacon and Caravaggio. Caravaggio usually painted stand more than what are shown in the captions. a perfect body of human beings. On the other hand, Bacon usually painted a flesh of human beings most of parts are re- Curation as Chance Discovery moved to express the essence of human existence. However, in the catalogue of the exhibition “Bacon and Caravaggio,” In [Abe, 2010], I proposed and defined a concept of curation at first, Coliva wrote “This exhibition proposes a juxtaposi- in chance discovery1 . tion of Bacon and Caravaggio. It intends to offer visitors an Though in various articles, the definition of a “chance” opportunity for an aesthetic experience rather than an edu- is described which was introduced by Ohsawa [Ohsawa and cational one... [Colvia, 2009].” Then Coliva continued McBurney, 2003], I wish to introduce it here again. In fact, it rather differs from the original definition in [Ohsawa and “An exhibition of generally conceived and prepared McBurney, 2003] to reflect the recent research interests. with a historicist mentality, but when it materializes, the simultaneous presence of the works — in the sense A chance is rare, hidden, potential or novel event(s) / precisely of their hanging — opens up parallels and situation(s) that can be conceived either as a future op- poses very complex and spontaneous questions, which portunity or risk. may even be unexpected and not all stem exactly from Then “chance discovery” research is a type of research to questions initially posed by art-historical motives and establish methods, strategies, theories, and even activities to theses. There are parallels that appear by themselves discover a chance. In addition, it aims at discovering human to the visitor’s sensibility and are not imposed by a factors for chance discoveries. theory of the curator. This is certainly one aspect of Accordingly a definition of curation in chance discovery the vitality of exhibitions, which make the works live is: and in this are necessary for the works. The display it- self, in the sense of the presentation of the works that • Curation is a task to offer users opportunities to discover appear in an exhibition —the spectacle of their being chances. on display — creates trains of thought that are inde- 1 In [Abe, 2011] and [Abe, 2012a], I extended the definition of pendent of the interpretations provided by art-historical curation in chance discovery, which introduced a concept of holis- scholarship. And since for a profound experience of un- tic communication [Akiyama and Sugiyama, 2004]. In this paper, derstanding a work these ramifications sometimes are I used the extended version of curation. • Curation should be conducted with considering to offer events. Thus he defined the affordance of something as “a implicit and potential possibilities. specific combination of the properties of its substance and • Chances should not be explicitly displayed to users. its surfaces taken with reference to an animal.” For instance, the affordance of climbing a stair step in a bipedal fashion • However, such chances should rather easily be discovered has been described in terms of the height of a stair riser taken and arranged according to the user’s interests and situa- with reference to a person’s leg length [Warren, 1984]. That tions. This can be achieved for instance by affordance. is, if a stair riser is less than 88% of a person’s leg length, • There can be a certain holistic communication environ- then that means that the person can climb that stair. On the ment. This type of holistic communication might function other hand, if a stair riser is greater than 88% of the person’s as media to discover chance for novice users. leg length, then that means that the person cannot climb that • There should be a certain freedom for user to interpret stair, at least not in a bipedal fashion. For that Jones pointed a key person, matter, thing or event, which should only out that “it should be noted also that this is true regardless stimulate or assist users’ thinking procedure. of whether the person is aware of the relation between his or • There should be a certain freedom for user to arrange her leg length and the stair riser’s height, which suggests chances. further that the meaning is not internally constructed and stored but rather is inherent in the person’s environment sys- The main point of curation is how to display data to users tem” [Jones, 2003]. and how not to insist on any interpretation to users. But crit- ical point to understand the data should be presented. Ac- Dementia care inspired by affordance cordingly, users can rather easily deal with data and can in- In [Abe, 2012b], I proposed the framework of a demen- terpret data flexibly, freely and properly. And as I pointed tia person support system. Bozeat and Hodges analyzed out before, the merit of this type of curation is make visitors the feature of mapping between objects and their meaning think themselves. That is visitors try to interpret/understand for semantic dementia person from four factors — affor- as many as possible, and they can understand more than dance, presence of recipient, familiarity, and problem solv- what are shown in the captions. Thus “interpretable” can be ing [Bozeat et al., 2012, Hodges et al., 2000]. They showed kept. very interesting results For instance, they pointed out “as a group, the patients did not achieve better performance on Dementia person support based on the concept a subset of affordable objects when use of these was com- odf curation. pared with a familiarity-matched subset of objects lacking This section discusses “Interpretable AI” from the aspect of such affordances. This absence of a general group benefit how to make understand an object especially to dementia applied both to overall use and to the specific component persons. of use afforded by the object’s structure.[...]it became clear Thanks to the advanced and innovative medical treatment, that there was a reliable benefit of affordance on the specific we are able to live longer. It will be happy to live long, but components of use, but only for the most impaired patients.” the other problems are caused by such long lives. One of the They also pointed out “The impact of recipient, like affor- most serious problems is increasing patients who are suf- dance, was found to be modulated by the degree of semantic fered from cancer. It will be able to be overcome by the ad- impairment. The patients with a moderate level of concep- vancement of medical treatment and is a problem for indi- tual impairment demonstrated significantly better use with viduals. Furthermore the more serious problem for a person the recipient present, whereas the patients with mild and se- and even for his/her family and surroundings will be de- vere impairment showed no effect. [...] It was not surpris- mentia. It is the progressive decline in cognitive function ing, therefore, to find that familiarity also influenced perfor- due to damage or disease in the body beyond what might mance on object use assessments.” These observations and be expected from normal aging. Dementia persons cannot analyses show that proper affordance might give a certain reasonably live their lives. It is said that the current medi- support to dementia persons understanding (meanings of) cal treatment cannot cure dementia completely. Even in the objects. near future, it will be negative to cure dementia. Dementia is For normal persons, it is not so difficult to provide such caused by problems in a brain. Accordingly, it is more diffi- a guidance. They can also understand analogy, so that they cult to cure dementia than cancer. Currently, some methods can extend the meaning to the other materials. For instance, to delay the progress of dementia are proposed. For instance, after finding that a tree stump functions as a chair, they can a therapy room or house will be one of the solution to take also understand a wooden board or box can also function care of dementia person [Sloane et al., 2002]. Actually, it is as a chair. That is, they can extend or map the meaning rather a support system for dementia person’s everyday life. to the other situations. However, for dementia persons, it is not easy to provide a proper guidance with which they Affordance can afford the function of an object. Actually, for person Gibson ecologically introduced the concept of affordance who does not have common knowledge or context, it is also for perceptional phenomena [Gibson, 1977, Gibson, 1979]. not easy to provide a proper guidance for affordance dis- It emphasizes the environmental information available in ex- covery. For them affordance is something rare or novel. Ac- tended spatial and temporal pattern in optic arrays, for guid- cordingly, it is rather difficult to be aware of “affordance” as ing the behaviors of animals, and for specifying ecological an afforded matter. In therapy houses, there should be many things which are not able to properly used by dementia per- sons. In the case, it is necessary to provide certain guidance’s F ∪ Object ∪ af f ordance ̸|= 2 (5) to lead the user to the correct direction to use things prop- A shikake can function as a the other object with map- erly. The simplest method will be to attach the name and ping function. Object′ ∪ shikake means that Object′ is a usage of things. It will functions well for normal persons. shikake to select a proper affordance for guessing the mean- However, for impaired persons, sometimes even such attach- ing or function of Object. For instance, if somebody does ment will not function well. For them, it will be necessary not understand the meaning or usage of a folding bed, a fold- to apply the other strategy to suggest or instruct the mean- ing wallet can be placed near the folding bed or shown to the ing or usage of things. For semantic dementia persons, it is user. Perhaps it will be better to show the wallet in an unfold- observed that they did not achieve better performance on a ing style. Thus by the help of a shikake, poeple can select a subset of affordable objects when use of these was compared proper affordance to understand the meaning of the object. with a familiarity-matched subset of objects lacking such af- fordances. Therefore, when we design an environment for dementia persons, it is necessary to consider such unhappy situations. It is necessary to prepare specialized affordances to dementia person. Even if they can detect affordance, they might not understand what it will emerge. For affordance, according to the Gibson’s definition, an Object is observed and affordance is detected in the envi- ronment to understand its meaning. Then, when meaning is fixed, by using abduction framework, the affordance deter- mination situation will be logically described as follows: F ∪ Object ∪ af f ordance |= meaning (1) F ∪ Object ∪ af f ordance ̸|= 2 (2) The above is described based on the formalization of The- Figure 1: Shikake in the proposal system. orist [Poole et al., 1987]. Affordance can be regarded as a set of hypothesis. We can select consistent affordance (equation In some case, Object′ is a shikake. From the viewpoint of (2)) in the environment (hypothesis base) to explain mean- “Interpretable AI,” as I mentioned above, this type of cura- ing. In addition, for understanding subset of or similar af- tion tries to make (dementia) persons think themselves. That forded objects (Object′ ), the affordance determination situ- is if all information is shown (dementia) persons think no ation will be logically described as follows: more, then dementia may proceed. However by the mapping via a shikake the message will be interpretable. F ∪Object∪Object′ ∪M ∪af f ordance |= meaning (3) This study was assumed for dimentia persons. However it can be applied to normal person as well. In fact, the above description is based on Goebel’s formal- ization of analogy [Goebel, 1989]. M is a mapping function Curation for museum visitors. from Object to Object′ . That is, to understand the same This section discusses “Interpretable AI” from the captions meaning of the subset of or similar afforded objects, an ad- in museum. ditional mapping function M is required. Thus if M can be determined and the usage of Object is known, Object′ can Captions in museums also be understood. In fact, for normal persons, M is easy In museums, especially in art museum, there are several vis- to understand. However, for dementia persons, it is pointed itors only reading captions (short explanation displayed next out that it is rather difficult to understand and determine M . to artwork in exhibition room) without seeing any artworks. Then the issue becomes how to suggest a mapping func- Tadaki and Abe tried to add psychological triggers by adding tion M as an additional hypothesis. One of the solution is some features to captions. The presence of change in how to introduce a shikake. According to Matsumura’s defini- they see artwork was measured by time spent to see art- tion [Matumura, 2013], a shikake is an embodied trigger for works, movement from caption to artwork, and participants’ behaviour change to solve social or personal issues. As a re- impressions to each artworks and each captions [Tadaki and sult of the action, all or part of problem will be solved. It Abe, 2017]. We regarded a caption as a Shikake which is may not sometimes be the person’s will. Matsumura contin- aiming to make visitors see artwork, each caption already ues that the shikake should be properly designed. That is, has a physical trigger. By our experiment, we can suggest the relationship between a problem to solve and a trigger to the possibility of a Shikake displayed in text as well as ob- action should be properly designed. I proposed the system jects. In addition, in [Tadaki and Abe, 2018], we questioned with a shikake [Abe, 2017]. “would people use different strategies to evaluate the paint- ing for each abstract painting and representational paint- F ∪Object∪Object′ ∪shikake∪M ∪af f ordance |= meaning ing?.” Our experiment was to control the description of cap- (4) tions. For instance, we hided the title of the abstract painting (Figure 3) (Forêt by Jean Fautrier (Figure 2)). The painting is the effect of caption according to its contents. We used the would be rather difficult to understand or create story with- painting shown Figure 4. We prepared three levels of ques- out the title. However, if the title (Forêt (forest)) was given to tion as follows; the audience (they could open the hidden title), one of them • Please imagine and write the story occurring in the paint- created a story “I found a fairy in a forest.” Thus a certain ing. hint or guidance can support person to understand a diffi- cult matters. This type of a shikake will be able to support • The title of this painting is “New Day,” please imagine and person’s understanding. write the story occurring in the painting. If something are different from what you imagined in the previous ques- tion, please write them. • The painter who drew this painting is a female painter liv- ing in Beograd in Selvia. Her name is Ivana Živić. She was born in 1979 in Sarajevo. She drew this painting in 2018. Please imagine and write the story occurring in the painting. If something are different from what you imag- ined in the previous question, please write them. Figure 2: Forêt by Jean Fautrier Figure 4: Experiment One of the answer sets is: • Since she is ill or is confined in the house, she cannot go out of her house. She dreams to go out to the outside world and go out from the window in her mind. • The reason she could not go out of her house was not unavoidable but intentional. And she did not dream to go out of her house but did go out of her house. • She remembers her girlhood. She remember that since outside world was danger she wanted to go out of house but could not, and that she strongly dreamed that she could go out of house. Before shown the statement I think the painter drew current situation but after reading it I Figure 3: Caption with hidden title changed my mind to think that she drew in her reminis- cence. The reason why she did not go out of house was the Yugoslav Wars? Captions with several levels of explanation The other answer sets were almost similar. The partici- In the previous experiment, a title is hidden. Especially for pants seem to be able to gradually understand the art work. an abstract painting if title is missing, it is very difficult Even such a representational painting, the level of under- guess what is drawn. When a title is given, new image ap- standing is gradually changing. For abstract paintings this pears in participant’s brain. In this experiment, our question type of caption functions more like the Jean Fautrier’s case. This type of showing system according to the cognitive tia person care system inspired by affordance, I used a term level is very important. In the dementia person assistant “shikake,” it can be regarded as various types of stimuli. system shown in the previous section, shikake should also Thus it will be possible to install several types of stimuli in be prepared according to the cognitive level. The gradually the environment for the better understanding. And this helps showing strategy can be regarded as shikake. In addition, users interpretation. Especially for dementia persons by a this shikake may automatically selected according to users help of selecting affordance, an interpretation of something interest and knowledge. is powered. For the new type of curation, strategies were ex- perimented and showed good results for “Interpretable AI.” Toward cognitive bias. I also discussed cognitive bias by the proposed systems. A cognitive bias is a systematic error in thinking that af- In the future, more experiments can be conducted to show fects the decisions and judgments that people make. Some the better results and propose the better system. of these biases are related to memory. The way you re- member an event may be biased for a number of reasons References and that in turn can lead to biased thinking and decision- Abe A. 2010. Curation in Chance Discovery, Proc. of making. Other cognitive biases might be related to problems ICDM2010 5th International Workshop on Chance Discov- with attention. Since attention is a limited resource, peo- ery, pp. 793–799 ple have to be selective about what they pay attention to in the world around them. Because of this, subtle biases can Abe A. 2011. Curation and Communication in Chance Dis- creep in and influence the way you see and think about the covery, Proc. of IJCAI2011 6th International Workshop on world. [Cherry, 2018] And Cherry pointed out that a cogni- Chance Discovery, pp. 3–8 tive bias occurs when people are processing and interpreting Abe A. 2012. Curation in Chance Discovery, in Ohsawa Y. information in the world around them. Also Cherry pointed and Abe A. eds.: Advances in Chance Discovery, SCI 423, out that Cognitive bias is not necessarily all bad, however. pp. 1–18, Springer Verlag Psychologists believe that many of these biases serve an Abe A. 2012. Cognitive Chance Discovery: from abduction adaptive purpose —they allow us to reach decisions quickly. to affordance, in Philosophy and Cognitive Science (Mag- This can be vital if we are facing a dangerous or threatening nani L. and Li L. eds), SAPERE 2, pp. 155–172, Springer situation. Verlag A few of the most common types of cognitive biases that Akinori Abe 2017. Abductive cognitive support for (seman- can distort our thinking are; tic) dementia persons, Proc. of 8th International Conference • Confirmation Bias: This is favoring information that con- Digital Human Modeling: Applications in Health, Safety, forms to your existing beliefs and discounting evidence Ergonomics and Risk Management (DHM 2017), Held as that does not conform. Part of HCI International 2017, (V.G. Duffy Ed.), , Part II, • Availability Heuristic: This is placing greater value on LNCS10287, Springer Verlag, pp. 119–131 information that comes to your mind quickly. You give American Association of Museums Curators Committee greater credence to this information and tend to overesti- 2009. A code of ethics for curators, http://www. mate the probability and likelihood of similar things hap- curcom.org/_pdf/code_ethics2009.pdf pening in the future. Akiyama R. and Sugiyama K. 2004. Holistic Communica- • Attentional Bias: This is the tendency to pay attention to tion, Senden Kaigi (in Japanese). some things while simultaneously ignoring others. When Bozeat S., Ralph M.A.L., Patterson K., Hodges J.R. 2002. making a decision on which car to buy, you may pay at- When objects lose their meaning: Waht happens to their tention to the look and feel of the exterior and interior, but use?, Cognitive, Affecgtive, & Behavional Neurosciences, ignore the safety record and gas mileage. Vol. 2, No. 3, pp. 236–251 Thus a cognitive bias is rooted in thought processing er- Kendra Cherry 2018. How Cognitive Biases Influence How rors often arising from problems with memory, attention, at- You Think and Act, https://www.verywellmind. tribution, and other mental mistakes. A cognitive bias is not com/what-is-a-cognitive-bias-2794963 Up- logical. The problem is it is not logical. dated November 06, 2018. If the thinking system is logical, it can be logically solved. Colvia A. 2009. Caravaggio Beckons to Bacon: The Beauty However, it is not logical. In addition, it is not always be of Sorrow, in Caravaggio Bacon (Coliva A. and Peppiatt M. solved. As pointed out in behavioral economics, sometimes heds.), pp. 17–22 it is not logical but reasonable. If it should be solved, the pro- posed methods can function well. Because they can control Gibson J.J. 1977. The Theory of Affordances, Perceiving, our attention and help our memory. Acting, and Knowing (Shaw R. and Bransford J. eds.) Gibson J.J. 1079. The Ecological Approach to Visual Per- Conclusions ception, Houghton Mifflin I showed two types of strategies for “Interpretable AI.” Goebel R. 1989. A sketch of analogy as reasoning with Which were a dementia person care system inspired by af- equality hypotheses, Proc. of Int’l Workshop Analogical and fordance and a new curation system. In fact, for the demen- Inductive Inference (LNAI-397), pp. 243–253 Hodges J.R. et al. 2000. The role of conceptual knowledge in object use evidence from semantic dementia, Brain, Vol. 123, pp. 1913–1925 Jones K.S. 2003. What Is an Affordance?, ECOLOGICAL PSYCHOLOGY, 15(2), pp. 107–114 Matsumura N. 2013. A Shikake as an Embodied Trigger for Behavior Change, Proc. of AAAI2013 Spring Symposium on Shikakelology, pp. 62–67 Ohsawa Y. and McBurney P. eds. 2003. Chance Discovery, Springer Verlag Poole D., Goebel R. and Aleliunas R. 1987. Theorist: A Logical Reasoning System for Defaults and Diagnosis, The Knowledge Frontier: Essays in the Representation of Knowledge (Cercone N.J., McCalla G. Eds.), pp. 331–352, Springer-Verlag Sloane P.D. et al. 2002. The Therapeutic Environment Screening Survey for Nursing Homes (TESS-NH): An Ob- servational Instrument for Assessing the Physical Environ- ment of Institutional Settings for Persons With Dementia, Journal of Gerontology: SOCIAL SCIENCES, Vol. 57B, No. 2, pp. S69–S78 Kotone Tadaki and Akinori Abe 2017. Museum Visitors’ Behavioral Change Caused by Captions, Proc. of the 2nd. Int’l Workshop on Language Sense on Computer in IJ- CAI2017, pp. 53–58 Tadaki K. and Abe A. 2018. The Influence of Story Writing Worksheets on Art Appreciation, Proc. of ICAROB2018, pp. 687–690 Warren W.H. 1984. Percneiving affordances: Visual guid- ance of stair-climbing, Journal of Experimental Psychology: Human Perception and Performance, Vol. 10, pp. 683–703