WILL MACHINES EVENTUALLY develop human-like intelligence?
Members of a multidisciplinary team of scientists at Michigan State University in East Lansing don’t believe that artificial intelligence will rival human intelligence any time soon. But as AI grows more advanced, it’s likely to learn in the same way as natural organisms—over generations, as a result of both environmental factors and trial and error.
“Understanding how learning behavior evolved … supplies clues to how our brains work and could even lead to robots that learn from experiences as effectively as humans do,” says lead author Anselmo Pontes, a doctoral student in computer science. Pontes completed the research with Robert B. Mobley, a doctoral candidate in zoology and ecology; Charles Ofria, a professor in the department of computer science and engineering; Christoph Adami, a professor of microbiology and molecular genetics; and Fred C. Dyer, a professor of integrative biology.
The team was inspired by the way animals such as honeybees learn to identify landmarks and navigate their environments. With that in mind, the researchers created a computer simulation to see if artificial organisms in a simulated environment could “evolve” to be able to use external signals to find food.
At the start of the experiment, organisms in the simulation had no ability to sense or move. But as the organisms reproduced, the simulation introduced forces such as genetic mutation, genetic inheritance, and competitive selection. Some mutations had neutral effects; others were lethal. But certain mutations led to the ability to collect more resources and reproduce more often. From generation to generation, a small number of organisms evolved from stumbling onto food accidentally to being able to learn from mistakes and find food with greater reliability.
“Evolution in nature might take too long to study, but evolution is just an algorithm, so it can be replicated in a computer,” says Pontes. In the simulation, he adds, “we saw populations evolve through the same behavioral phases that previous scientists speculated should happen but didn’t have the technology to see.”
Read “The Evolutionary Origin of Associative Learning,” published online August 23, 2019, in The American Naturalist.