Can artificial intelligence solve brainteasers? Can it reason like an adult? Scientific American posed these questions in a new story examining AI’s ability to apply logic to a series of mathematical problems. 

The answer? AI systems surprise us in their human-like abilities to answer complex questions, but can also generate “spectacularly weird” failures, says Kentaro Toyama, an expert on human-computer interaction. 

Toyama and  Merve Hickok, both faculty members at the University of Michigan School of Information, discuss what AI has to offer and possibilities for growth in the burgeoning field. 

“AI does not have reasoning capabilities; it does not understand context; it doesn’t have anything that is independent of what is already built into its system,” Hickok says. “It might sound like it is reasoning; however, it is bound by its data set.”

The question of whether or not AI can gradually learn how to reason by feeding it more and more data has long been debated, Toyama says. 

“One thesis is that if you give it so many examples of logical thinking, eventually the neural network will itself learn what logical thinking looks like and then be able to apply it in the right instances,” Toyama says. “There are some [other] people who think, ‘No, logic is fundamentally different than the way that neural networks are currently learning, and so you need to build it in specifically.’”

RELATED

Read “You Can Probably Beat ChatGPT at These Math Brainteasers. Here’s Why” on Scientific American. 

Learn more about Merve Hickok’s research on AI bias and  regulation by visiting her UMSI faculty profile. 

Read more research on the limitations and potential of AI by Kentaro Toyama through his UMSI faculty profile.