Location
Hilton Hawaiian Village, Honolulu, Hawaii
Event Website
https://hicss.hawaii.edu/
Start Date
3-1-2024 12:00 AM
End Date
6-1-2024 12:00 AM
Description
Generative artificial intelligence (GAI) appears useful in the creation of new data, which assists in the expansion of small, limited datasets in fields such as analogical reasoning (AR). This multidisciplinary study expands the number of AR visual datasets within the field of visual question answering. We introduce the first visual analogy dataset that includes abstract concepts by leveraging three text-to-image GAI generators, Text2Img, Craiyon, and Midjourney, to produce images for antonym and synonym analogies. Our visual dataset achieves up to 70% accuracy and performs better 84.6% of the time compared to the same evaluation on only textual information. Interestingly, results also imply that paid GAI services produce higher accuracy. This work shows the potential for GAI to aid in the development of abstract visual analogy datasets, which allows for a better understanding and incorporation of AR into cognitive-inspired AI models capable of analogy-based information fusion.
Recommended Citation
Combs, Kara and Bihl, Trevor, "A Preliminary Look at Generative AI for the Creation of Abstract Verbal-to-visual Analogies" (2024). Hawaii International Conference on System Sciences 2024 (HICSS-57). 2.
https://aisel.aisnet.org/hicss-57/da/trustworthy_ai/2
A Preliminary Look at Generative AI for the Creation of Abstract Verbal-to-visual Analogies
Hilton Hawaiian Village, Honolulu, Hawaii
Generative artificial intelligence (GAI) appears useful in the creation of new data, which assists in the expansion of small, limited datasets in fields such as analogical reasoning (AR). This multidisciplinary study expands the number of AR visual datasets within the field of visual question answering. We introduce the first visual analogy dataset that includes abstract concepts by leveraging three text-to-image GAI generators, Text2Img, Craiyon, and Midjourney, to produce images for antonym and synonym analogies. Our visual dataset achieves up to 70% accuracy and performs better 84.6% of the time compared to the same evaluation on only textual information. Interestingly, results also imply that paid GAI services produce higher accuracy. This work shows the potential for GAI to aid in the development of abstract visual analogy datasets, which allows for a better understanding and incorporation of AR into cognitive-inspired AI models capable of analogy-based information fusion.
https://aisel.aisnet.org/hicss-57/da/trustworthy_ai/2