Flies may learn the same way we do

May 7, 2021
We've got more in common with flies than we might think. (Unsplash/Svetozar Cenisev)

We've got more in common with flies than we might think. (Unsplash/Svetozar Cenisev)

Neurological learning mechanisms shared by humans and other mammals may also be used in the brains of fruit flies, hinting at an elusive link between insect and mammalian learning. 

For the study, published Friday in Nature Communications, researchers incorporated data from recent experiments into a new computational model to demonstrate how the brains of fruit flies, much like our brains, may use dopamine signals to facilitate learning, supporting what's known as the reward prediction error hypothesis. 

The reward prediction error hypothesis holds that dopamine neurons signal an "error" between the expectation of a reward and actually experiencing one. The basic idea is that we learn from how right or wrong our predictions turn out to be, and the difference between the reward we received and the one we expected. This phenomenon has been observed across diverse species and is used to study reinforcement learning in artificial intelligence. 

The possibility that our brains use the same learning tool as an animal as basic as a fruit fly wasn't unexpected for James Bennett, an informatics researcher at the University of Sussex in England and the study's lead author. 

"When you step back and reflect on the ability of evolution to come up with very strong solutions in many different species, well, it's actually not that surprising that insects would use prediction errors because they have evolved over billions of years and this seems to be a rather successful learning algorithm in mammals," he told The Academic Times. "It would be silly to think that it wouldn't be successful for insects because it's such a general and simple algorithm for learning."

Previous studies have suggested that the circuitry of the mushroom body — a pair of structures in the insect brain that are associated with memory — is wired for learning, but none have shown that flies use prediction errors to learn, Bennett said. "That's the step we've taken, saying, 'Look, these connections look very much like prediction error learning.' We encountered quite a bit of resistance to the idea simply because of these previous experiments that have not managed to find evidence to support the reward prediction error hypothesis in flies. The view that we took is that there's not really any decent evidence refuting that hypothesis, either." 

Working with anatomical data from previous experiments, the researchers developed a computational model that simulated the mushroom body circuitry within the fruit fly's brain. "It's essentially a matter of seeing where there are connections between neurons, defining a set of equations that explain the activity between those neurons and solving those equations numerically on a computer," Bennett said of the methodology behind the study. 

Not only did the researchers find that the anatomy and physiology of a fruit fly's brain supports prediction error learning, they also showed that data from the previous insect experiments did not necessarily conflict with predictions from the reward prediction error hypothesis as previously thought. 

"A lot of experiments have failed to show that there's a prediction error coding going on in insects," Bennett said. "What we've attempted to do with our study is show how the behavior that is learned in some experiments matches very well with the prediction error hypothesis. The physiology in neurons — how they behave while learning — matches very well with our model."

The results are purely computational but could have important real-world implications. For example, discovering more about the way flies learn and make decisions could pave the way for more humane lab experiments involving insects instead of rodents.  

"If we're able to establish more of a link between insect and mammal brains, there would be much more reason to use insects, at least for the basic understanding of whatever your interest is — in my case, it's learning — before moving into higher organisms like rats and mice," Bennett said. "We can develop more refined hypotheses and get more meaningful data out of experiments with mammals, having done our preliminary investigation in insects." 

Studying the firing of neurons in the brains of flies could also provide insight into the brain circuitry underlying mental illness and drug addiction in humans, which are believed to be underpinned by prediction errors. But that idea could meet resistance within the research community, according to Bennett. 

"There's a bit of a taboo about thinking that insects have cognition — that they are conscious and they can experience things such as anxiety and depression, which are very complex phenomena in humans, just because [insect] brains are so tiny," he said. "But some people are beginning to look for those kinds of behaviors in flies that are indicative of anxiety — repetitive behavior such as walking around in circles — which has been observed in flies and rodents in finely controlled experimental setups." 

It's possible that a future study will refute the research team's finding that fruit fly brains support prediction error learning, Bennett said. But the real value of the study is in knowing more about how insects learn — possibly in ways that differ from our own. 

"If we find results different from what we predicted, that's kind of interesting in its own right," he said. "That means they're doing something else that is more sophisticated, and that will be invaluable for us to understand. The reward prediction error hypothesis is such a powerful tool for understanding learning in other animals, but also in artificial intelligence algorithms. If there were some other way of learning that is also very powerful, that would be great to know." 

The study, "Learning with reinforcement prediction errors in a model of the Drosophila mushroom body," published May 7 in Nature Communications, was authored by James E.M. Bennett, Andrew Philippides and Thomas Nowotny, University of Sussex. 

We use cookies to improve your experience on our site and to show you relevant advertising.