Machines can learn from fables – USC Viterbi
If a friend told you that he has the blues, do you think he changes color?
Although this question may seem facetious, it offers a simple entry into the rich world of analogical reasoning, a tool that allows humans to generalize knowledge from familiar situations to new ones. We see it in areas ranging from politics to medicine; it is a cornerstone of our daily cognition. It can be as simple as a child throwing a beach ball, which they recognize is similar to a basketball; and as complex as a physician using past case studies to determine a care plan for a patient.
Now, researchers at the USC Information Sciences Institute (ISI) are extending this thinking process to machines.
A new paper, “Understanding Stories Through the Dimensions of Analogy,” featured at the Qualitative Reasoning Workshop, which is co-located with the International Joint Conference on Artificial Intelligence taking place on July 23 2022, teaches artificial intelligence to make creative analogies through an ancient art form… fables.
“People working in AI have tried to achieve the same level of reasoning as humans in AI systems, and it’s a very difficult challenge to try to mimic the analogical reasoning that humans take for acquired,” says Jay Pujara, USC ISI research manager and assistant research professor at the USC Viterbi School of Engineering..
The current dominant AI paradigm is machine learning, which relies on applying existing knowledge to new situations. This framework cannot support analogical reasoning, which has created a significant challenge for researchers. This is partly because analogical reasoning relies heavily on relational thinking, which is how humans discern meaningful connections between items or situations that lack superficial similarities.
For example, despite the apparent differences between summer and winter, one can reasonably conclude that the following is an analogous pair of words: “the sun is to summer what snow is to winter”. In essence, analogical reasoning unifies different concepts, allowing us to extract meaning from the myriad stimuli we encounter daily. Instead of drawing connections from appearance alone, we can thus make creative connections between existing and new scenarios.
Since AI models lack analogical reasoning mechanisms, they struggle to understand, explain, or generalize novel stimuli. Creating technology with analog reasoning capabilities would allow AIs to assess the relevance and meanings of language, which has many real-world applications. AI with human-like conversation and interpretation capabilities could be used to teach new concepts to students or create new products based on consumer marketing data. These AIs have a transformational impact on society, such as reducing traffic by analyzing gaps in current infrastructure to generate improved highway models.
Previous studies have had limited success in developing an AI with the ability to make analogies. However, the technology did not understand the implications of such analogies and could not make large-scale generalizations. The USC team tackled this problem by experimenting with different techniques to train AIs to understand the analogies found in Aesop’s fables, a collection of simple short stories that convey moral ideas. Using natural language processing (NLP) methods, they analyzed the fables to generate story pairs based on lexical and semantic similarities or on the words and meanings present in the text.
Said Pujara: “We chose short stories with a moral purpose because often you find several fables with the same moral purpose and the same message at the end but which are told in very different ways. So that means there’s a semantic meaning to a fable, which is very different from the surface form it takes, and humans can see those connections.
While humans recognize that the same theme of greed connects the stories of a stealing fox and a hoarding merchant, the study found it difficult for AI systems to identify these analogies. .
“Although the techniques we have developed can be used to construct analog frames, recent advances in NLP are still not sufficient to achieve human-level precision,” says Thiloshon Nagarajah, co-author of the paper and MSc student in Computer Science from USC Viterbi. Despite this, the team managed to catalog the different ways humans approach analogies, which is a promising step towards the goal of creating an AI with analog reasoning capabilities.
A secondary conclusion of the study was that the researchers themselves needed to engage deeply with each other to determine whether specific stories constituted an analog pair. It turns out that analogical reasoning is a more subjective and interpretive space than expected, which “points to the fact that there are other phenomena that have not been fully explored in this article,” said Filip Ilievski, another co-author and researcher at USC Viterbi. assistant teacher.
One such phenomenon is the influence of prior knowledge on analogical reasoning tasks. When an individual encounters a new situation, he uses his personal experiences to establish connections between the unknown and the known. Since each individual has a unique repertoire of knowledge and skills, analogical reasoning inevitably varies from person to person. Ilievski suggests that by further exploring the nuances of human analog reasoning, changes can be implemented to improve the design of AI technology.
This study offers a promising start to developing artificial intelligence with analogical reasoning capabilities. The usefulness of such AI technology goes well beyond short stories – potentially improving everything from education and public policy to art and urban design.
Posted on July 25, 2022
Last updated July 25, 2022