Dissertation
The Artificial Intelligence Quotient: Measuring Machine Intelligence Based on Multi-Domain Complexity, Similarity, and Diversity
Washington State University
Doctor of Philosophy (PhD), Washington State University
2023
DOI:
https://doi.org/10.7273/000006368
Abstract
The growth and development of AI systems and benchmarks have been rapidly increasing, yet there is a disproportionately small amount of examination into the domains used to evaluate these systems. Even the domains that consider a larger scope of evaluation are often not generalizable, and the implemented AI systems are often not usable for different domains. To address the previously discussed issues within the AI community, we are putting forward a notion of machine intelligence that can be intuitively understood and effectively utilized. This notion will allow us to compute new metrics for both the agents evaluated and the domains used in the evaluation. The intelligence of a given system is determined by measuring its performance on a multi-domain test with measurable complexity, similarity, and diversity. The Artificial Intelligence Quotient (AIQ) structures these domain side measurements into a clear and consistent framework that can be utilized to measure an AI system’s intelligence. These domain side measurements allow for the creation of an intelligence space. Once a test is located within the space, its accompanying performance metric can be used to effectively scale the location allowing for a notion of agent capacitance. That is, the agent’s capability to achieve a certain understanding of the domain. An agent that achieves a higher understanding of the domain will be given a higher AIQ score than one that achieves a lower understanding. Agent capacitance also scales with domain complexity. An agent that achieves a comparable understanding of a more complex domain will be given a higher AIQ score than an agent on a less complex domain.
Metrics
7 File views/ downloads
59 Record Views
Details
- Title
- The Artificial Intelligence Quotient
- Creators
- Christopher Isaac Pereyda
- Contributors
- Lawrence B Holder (Advisor)Diane J Cook (Committee Member)Venkata Janardhan Rao Doppa (Committee Member)
- Awarding Institution
- Washington State University
- Academic Unit
- School of Electrical Engineering and Computer Science
- Theses and Dissertations
- Doctor of Philosophy (PhD), Washington State University
- Publisher
- Washington State University
- Number of pages
- 230
- Identifiers
- 99901087514301842
- Language
- English
- Resource Type
- Dissertation