Incorporating Abstract Knowledge Structures in Machine Learning: Improving Question Answering, Problem Solving and Teaching in Personal Assistants and Educational Software .
Year: 2021
Type: article
Abstract: Artificial intelligence (AI) and machine learning(ML) are finding their way into many applications, ranging from personal assistants to learning systems to educational software.We believe that each of these applications has inherent limitations that keep it from realizing its full potential.Personal assistants answer questions by matching queries to text strings or metadata from websites.When responses are complex, users are often given links to websites rather than answers to their queries.Many learning systems use ML algorithms that focus on statistical data analysis, thus limiting its usefulness to classification or statistical predictions.Most AI-based educational software is labor intensive to create as the underlying concept knowledge must be hard-coded into the software.The limitations that all of these applications have in common is an inability to process information (learn) and form abstract knowledge representations that support higher level cognitive tasks.The present paper describes a framework and resulting software that can read a mathematics lesson on 2-step equations that is written in English (simulating the result of a search engine retrieval), learn the underlying concepts and then apply its knowledge to answer questions, solve problems, correct the work of others who are solving similar problems, and solve problems and correct the work of others for the topic of 1-step equations, even though the material it learned made no mention of this topic.
Authors: John Leddo, Ivy Liang
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
SDG: Quality education
Open Access status: diamond
OpenAlex ID: W4206309101