When computational biologist Rhanor Gillette studies predatory behavior in animals, he doesn’t begin with lions, tigers, and bears. He focuses instead on a sea slug called Pleurobranchaea californica. Or, at least, a virtual version of one.
In a recent study published in the journal eNeuro, Gillette and his team from the University of Illinois showed that a computer-simulated sea slug, which they’ve named “Cyberslug,” can respond to stimuli much as its living counterparts do in the real world. This A.I. invertebrate reacts to food and other members of its species, according to the researchers, even displaying basic forms of self-awareness.
“We’ve designed Cyberslug to reproduce the relationships that we found in the brain of the real sea slug predator,” Gillette told Digital Trends.
In the simulation, the researchers track the Cyberslug’s “appetitive state,” or how full it is, to see how it responds to stimuli. When it meets a fellow slug it can either eat, mate, or flee, depending on its current disposition. In a low appetitive state, Cyberslug is satiated and steers away from potential food. But, when it’s appetitive state is high, “due to hunger, a good memory about it, and the tastiness of a stimulus … avoidance is simply switched to approach and attack,” Gillette said. “That is, the bot adds up sensation, motivation, and memory to make a cost-benefit decision of is this prey worth attacking?”
Gillette places this approach/avoidance decision-making at the heart of all behavior that deals with consuming or conserving resources. “So, like us, the bot and the animal make economic decisions not on the basis of simple information by itself, but on how the information makes it ‘feel.’”
In their virtual sea slug, Gillette and his team think they’ve reproduced a “primitive state of consciousness, where a simple animal’s experience is locked to immediate events, and its behavior is guided from moment to moment by reward learning and motivation.”
Cyberslug’s apparent self-awareness is still very primitive — it won’t be having an existential crisis any time soon. But Gillette and his colleagues hope that by further developing the model, they can find useful applications in fields like robotics and video games, granting machines and non-player characters more sophisticated behaviors and decision-making.
Moving forward, Gillette said his team will continue to empower Cyberslug to develop more complex levels of awareness, and to study its emergent traits. “We are now working to enhance the simple Cyberslug model for enhanced sociality and cognition with the aim of producing entities that may scheme for the future, and cooperate or deceive like the social mammals,” he said.
Let’s just hope Cyberslug remains in the simulation until it’s taught to play nice.