What are the origins of intelligence. Is knowledge learned or already present at birth? How do our brains become shaped by our experiences? How does the brain really work and what breakthroughs await once researchers figure that out?
Justin Wood, associate professor of Informatics, and Samantha Wood, assistant professor of Informatics, are on the cutting edge of answering those questions and more through their Building a Mind Lab research at the Luddy School of Informatics, Computing, and Engineering.
“There are a lot of universities either interested in the brain or artificial intelligence,” Justin Wood said, “but very few that bring those areas together.
“That’s one reason why IU is so special -- it brings psychologists and computer scientists together to make a run at these really hard questions.”
Call it the intersection of biological intelligence and artificial intelligence, where researchers use vision models to understand the mind. The Woods use models from artificial intelligence to understand the brain and apply knowledge from psychology and neuroscience to build more efficient and more robust models of artificial intelligence.
“This is a symbiotic relationship and a promising direction for cognitive science, neuroscience and the engineering community,” Samantha Wood said.
The brain’s complexity creates formidable challenges for scientists seeking to replicate its abilities. The Woods tackle this complexity problem by leveraging tools from artificial intelligence, which can serve as models of how brains learn and develop. With this approach, the lab addresses classic nature-nurture questions at the heart of psychology, neuroscience and cognitive science.
Their research links studies of newborn chicks, computer vision models, virtual reality and digital twins, allowing the lab to study how learning works in biological and artificial systems.
“Our lab is interested in the origins of intelligence,” Samantha Wood said. “What kind of learning mechanisms do we have from birth that help us understand the world around us without any prior experiences or instruction?
“Our goal is to build an engineering level of understanding of the origins of intelligence.”
The Woods raise newborn chicks and newborn AI vision models in the same environments and test them on the same tasks.
“On the animal side,” Samantha Wood said, “we use virtual reality so we can raise animals in completely virtually created, digitally created environments. We can completely control the environments and stimuli they have access to from birth.
“On the AI side, we create virtual bodies that match the newborn animals and put brains into them. We use deep neural networks to create those brains.”
In other words, if chicks and vision models are raised in the same environment with the same experiences, do they develop the same kind of knowledge and behavior, and does that mean knowledge comes from experience and interaction with the environment?
So far, the answers are yes and yes.
Two thousand years ago, Greek philosophers Plato and Aristotle argued about whether humans are born with innate knowledge through evolution or whether every person learns everything from scratch. Justin Wood said that debate remains just as strong today.
“Nobody has been able to address this question before, because in order to do so, you need to control everything an animal sees from birth, then you need to give those same experiences to a model, and then you need to measure whether the model learned like the animal. This was a formidable technological challenge.”
Together, the Woods developed a solution to this challenge.
“We’re excited about this,” Justin Wood said. “We might be able to answer one of the great unsolved mysteries in science. We might be able to characterize the origins and development of knowledge. How do brains change and become shaped by the various experiences we get during our normal lifetime?”
Beyond that, Samantha Wood said, their research can boost understanding of how basic sensory circuits work, which is important for understanding things like visual processing problems, or for understanding the learning mechanisms in the brain, which has important implications for education and teaching.
“It also tells us about how we chose our actions and make decisions,” she said. “That has important applications for understanding decision making and mental health.”
The Woods use chickens because they are one of the few animal species that can be raised in a controlled environment without a caregiver.
“In order to study the origins of knowledge,” Justin Wood said, “we need to control everything in the animal’s world, including all the objects and agents and social partners.
“With a chicken, we can hatch them in darkness in an incubator. We can move them into a virtual reality chamber in darkness with night vision goggles, and then turn on the world to reveal an entirely controlled virtual experience for the animal without worrying about a mother hen or siblings providing uncontrolled experiences. The core goal is to control all the experiences the animal gets and give those same experiences to a model.”
With this approach, researchers can study the machinery of the mind, what it looks like and how it works.
The Woods believe that understanding development and learning can shed light on vision, learning processing, and even mental disorders.
“If we can take development seriously and understand how the mind develops as a function of experience,” Justin Wood said, “then we might have a chance to figure out how our brains work.”
Justin Wood guessed that, within five years, researchers will have a computational model that develops the same core mental skills as a newborn animal.
“If we have a complete understanding of the origins of knowledge, then we should be able to raise an embodied machine in the same world as an animal,” he said, “and just like the animal, the machine should develop object perception, navigation, social skills, motor skills, and numerical cognition. Like animals, all of these core skills should develop automatically in the machine. Once we have a runnable model that develops that same core mental skills as humans and animals, we’ll be in a position to tackle things like mental disorders.”
Samantha Wood said it might be 10 or more years before that computational model is developed.
“A lot depends on the huge changes we see in the field of artificial intelligence, which is hard to predict.
“Now that we’re seeing more interest in connecting biological intelligence to artificial intelligence, it seems like we’re getting these bursts of artificial intelligence inspiration faster.”
The Woods hope their research leads to a new and mature scientific field, where researchers use artificial intelligence to study the brain and understand how it works.
When that happens, imagine the breakthrough possibilities.