Artificial Conversations Spark Insights into the Evolution of Ideas
By Matt Windsor
Philosopher Marshall Abrams is designing a digital simulation of the flow of ideas among individuals that leads to cultural change.
In a small office in UAB’s Humanities Building, philosopher Marshall Abrams, Ph.D., is hosting a heated debate about the origins of life. The nine participants share their opinions rapid-fire, completing several hundred conversations every minute. Nevertheless, not a word is spoken; all the action is happening on Abrams’s computer screen. Welcome to philosophy’s digital era.
Above: a representation of the neural networks inside one "person" in the simulation.
Abrams is building a computer-based simulation of cultural change, the flow of ideas among individuals that exerts a powerful shaping force on a society’s guiding values. “Meteorologists want to understand the local changes that affect large-scale weather patterns,” Abrams says. “Social scientists want to do the same thing with cultural change. But just like the weather, it is very subtle and complicated. I’m trying to see what a digital model can add.”
Talk Amongst Yourselves
Researchers study the process of cultural change largely through surveys and by tracking books, speeches, films, blog posts, and other recorded artifacts of our collective thought processes. But Abrams, inspired by a growing number of simulations in social science fields, is taking a different approach. Building on software developed by Canadian philosopher Paul Thagard, Abrams has designed a system that lets individual software “agents” talk amongst themselves, sharing their opinions and then responding to the ideas of others by changing their own views.
Abrams is particularly interested in analogies—the way we view our relationships and place in the world through particular cognitive filters. “Your relationship with your boss might be influenced by your relationship with your father, because both are authority figures,” Abrams says. “In the political sphere, some people may think about the nation as a family, and what you think about your own family will affect how you think about the nation’s leaders. What I’m interested in is how these analogies about the world mutate over time.”
Abrams programs the initial set of views held by each agent, but these opinions change during each simulation as the agents encounter others. “The software enables the agents to notice new analogies and metaphors,” Abrams says. “They interact with each other and decide what to believe and what to reject.”
The system is not built for deception, Abrams points out. “No one is trying to convince others of an idea they themselves don’t believe in, as can happen in the real world,” he says. “Although I may go in that direction in the future.”
Easy for You to Say
The conversations—represented visually as dotted lines between agents, who are grouped in a circle—happen a little faster than in real life. “Right now I’ve got 20 agents, who can select from 40 to 50 beliefs,” says Abrams. In 10 minutes, those agents exchange ideas roughly 30,000 times.
Constantly shifting graphs within Abrams's program (left and right) reflect the changing views of participants (represented at center) as they interact
Abrams is building his software with the help of computer science student Kristen Hammack. He is no digital neophyte; he put himself through graduate school partly by helping to build e-commerce Web sites for businesses.
At the moment, Abrams’s simulation is still in the early stages. “I want to show that this strategy for modeling cultural change could be useful” in providing evidence for theories put forward by other researchers, he says. “Suppose a historian says that a change in 16th-century Europe came about because of a shift in analogies. You could build those ideas into the model and get some additional evidence about whether the theory seems plausible.”
A computer program that evolved to the point that it could actually predict coming cultural trends would be fascinating, but Abrams says his simulation is unlikely to reach that level of refinement. “I’m not sure this approach will ever reach the precision needed to predict the future,” he says. “But if that did happen, it would be pretty cool.”