Researchers Explore How to Make AI More Human
A multi-institution team of researchers are exploring how animals think to find new ways to improve artificial intelligence
By Stacy Kish
Artificial intelligence (AI) is polarizing. It has been touted as a saving grace and as a potential dark overload for the future, but how smart is this technology?
Alan Turing devised the Turing Test, a way of assessing if a technology is intelligent by seeing if it can successfully fool a human. While artificial intelligence systems have passed this test within certain domains, none have come close to touching the creative intelligence of humans, primates, and even honeybees
A team of researchers at Carnegie Mellon University, University of Pennsylvania and Johns Hopkins University received a $500,000 grant from the National Artificial Intelligence Research Institutes program to design a new class of Turing Tests that assess both biological intelligence and AI.
“The foundations of modern AI rest largely on discoveries in cognitive science and neuroscience,” said Timothy Verstynen, associate professor of psychology. “[As AI has progressed,] the goals have become more engineering focused, like how do I train a model to perform well in a very specific context.”
Currently, AI systems learn by trolling large amounts of labeled data that fits specific criteria for the task. This process is time consuming and limited. On the other hand, animals have evolved the ability to extrapolate beyond a few data points to solve problems. This flexibility and creativity is currently lacking in AI technology.
This project will build on the original Turing Test by expanding how the test evaluates the flexible thinking tasks, which are possessed by the most basic animals.
“Biological agents, like humans or other animals, learn from experiences in order postulate solutions or strategies for solving yet unseen problems. They predict, build hypotheses and continually learn across many different contexts,” said Verstynen. “We wanted to ask ‘How do we get AIs to act more like that?’”
This project aims to measure aspects of intelligence found in biological organisms lacking in AI. Once developed, they plan to administer this test to animals, humans and AI technology. By understanding AI performance from the biological vantagepoint, the team hopes to set new milestones for researchers to work toward to improve this technology.
The National Science Foundation is leading this effort by funding the agency’s National Artificial Intelligence Research Institutes program. The program has committed $120 million to establish six AI Research Institutes across the country. These centers are part of a long-term, multidisciplinary and multi-institutional effort to address the most fundamental questions surrounding artificial intelligence.
Verstynen joins team lead Konrad Kording at U Penn and Joshua Vogelstein and Leyla Isik of Johns Hopkins University on the project titled, “From Biological Intelligence to Human Intelligence to Artificial General Intelligence.