AI systems often need to work with limited knowledge representations. For example, when trying to identify an object in an image, the system may only have a few pixels to work with. Similarly, when trying to identify a concept in text, the system may only have a few words to work with. In these cases, the AI system needs to be creative in order to identify the relevant features and make use of them to solve the problem at hand.
One way AI systems can be creative in limited knowledge situations is by using analogous reasoning. This involves finding other similar cases where more is known and using that knowledge to infer information about the current case. For example, if an AI system is trying to identify an object in an image, it may first look for other images of that object. If it finds an image of a cat, it may then use information about cats (e.g., they are usually furry) to help identify the object in the original image.
Another way AI systems can be creative in limited knowledge situations is by using heuristics. This involves using intuition or “rules of thumb” to solve the problem at hand. For example, when trying to identify an object in an image, the system may use heuristics such as “look for things that are usually round” or “look for things that are usually a certain color.”
In general, AI systems can be creative in limited knowledge situations by using a variety of techniques such as analogous reasoning and heuristics. By using these techniques, AI systems can often find creative solutions to difficult problems.
References:
https://www.pnas.org/content/pnas/112/34/E4683.full.pdf
https://towardsdatascience.com/handling-skewed-data-8a30129a4b76