Connecting neural and symbolic AI networks
Aug. 02, 2023.
2 min. read Interactions
US Office of Naval Research funding may help make AI more reliable and transparent
The U.S. Office of Naval Research has awarded Michigan State University researchers with a $1.8 million grant project to make artificial intelligence more reliable and transparent.
This could also enable people to entrust AI systems with more advanced jobs that rely on understanding language and visual information, including education, navigation and multimodal question-answering systems.
Connecting symbolic AI with deep neural networks
Led by Parisa Kordjamshidi, an assistant professor in the Department of Computer Science and Engineering, the team is working to connect symbolic AI with current deep neural networks to create a neuro-symbolic framework. This combined approach could create systems that have a vast wealth of learning from data along with explicit reasoning capabilities. Dan Roth, a professor at the University of Pennsylvania, is also a co-investigator.
“Every day, these AI models impress us, but we’re still not sure how trustworthy and reliable they are,” said Kordiamshidi. “Even when they provide the right answer, they might be right for the wrong reasons. We need to know what is their line of reasoning. That’s not very clear right now, and that’s the challenge.”
The researchers are also helping to better process a range of inputs — text, images and video — to make human interactions with computer systems more powerful and seamless.
Dr. Ben Geortzel: A comment on neural-symbolic AI
“Neural-symbolic AI approaches have been the subject of deep research and practical experimentation and commercial deployment for a couple decades now, and it’s great to see them finally start to get more of the attention, prominence and funding they deserve.
“When I built my first neural-symbolic AI system in Webmind Inc. in 1997, it was fairly cutting-edge, though not totally unprecedented. By now, there is a vast reserve of experience in such systems to draw on in doing neural-symbolic engineering, and I have no doubt this is one of the ingredients that will be valuable in integrating LLMs into broader AGI architectures during the coming months and years.” (Also see: https://people.cs.ksu.edu/~hitzler/nesy.)