IonQ Quantum Cognition Research Aims to Make AI More Human

Quantum computing company IonQ recently published research demonstrating that quantum could better model the way humans make decisions. The study ran human cognition models on quantum hardware to better understand how the order of questions presented in research or a survey affects people's answers, uncovering insights into how our brains work.

This could be used in future AI models that more closely mirror human thought. Potential applications include reevaluating how survey questions are written, predicting better election results, detecting unconscious biases in AI models and producing better-tailored ads.

An example of human behavior which has been difficult for computers to predict is the Clinton-Gore order experiment. Test subjects’ perceptions of how trustworthy then U.S. President Bill Clinton and Vice President Al Gore were varied depending on whether they were considered individually or together, and on the order in which they were considered.

Presented with a poll asking: “Do you generally think Bill Clinton is honest and trustworthy?” and subsequently, the same question about Al Gore. Clinton received a 50% agreement rate and Gore received 68%. With the order reversed, the agreement rate for Clinton increased to 57% while for Gore, it decreased to 60%. Using quantum models to predict this kind of human behavior is known as quantum cognition.

“Quantum cognition is a collection of results that have been developed in cognitive psychology for good 25 years, using quantum mathematical structures to come up with better models for how humans evaluate information and make decisions,” explained IonQ senior principal scientist Dominic Widdows. “The embarrassing starting point is that classical probability is very bad at describing how people reason.”

Related:Top 6 Quantum Computing Companies as Ranked by Google

According to Widdows, quantum cognition is helpful because it makes it possible to explain some of the ways people reason more effectively than classical models.

“When someone asks you a question, if you're not quite sure of the answer, you come to a conclusion and give the answer,” he said. “If they ask you the same question again, you're not going to go through that same process again, you're going to give the same answer.” He added that the mathematics behind it is well-established and maps well onto the probabilistic nature of quantum modeling.

A key future application for quantum cognition could be informing how to order requests and surveys.

“For example, we're taught to give people a bit of good news before we give them bad news or a bit of praise before negative feedback. This is not well-modeled as a set of classical logical propositions,” said Widdows.