Should neuroscientists trust methods that can't explain how Donkey Kong works?

Many standard neuroscience methods failed when tested on a classic video game console.

We know so little about how neural systems work—even simple ones, like that of a fruit fly—it can be hard to test the experimental techniques and analytical methods neuroscientists use. In a newly published study, Eric Jonas and Konrad Kording set out to test these algorithms on a system they do understand: the microprocessor of the Atari 2600, an early home video gaming console that ran Donkey Kong, Pac-Man, and other classics.

“Since humans have designed this processor from the transistor all the way up to the software, we know how it works at every level, and we have an intuition for what it means to ‘understand’ the system,” the authors explain. When they applied these analytical tools to the console’s microprocessor, they found that they didn’t adequately explain how the technology works. “Neuroscience approaches fall short of producing a meaningful understanding,” says Kording. For Jonas and Kording, this calls the reliability of these methods into question.

Of course, a microprocessor is not a brain. The Atari 2600 has only 3000 transistors, while a mouse brain has 100 million neurons, each of which the authors estimate does the work of thousands of transistors. However, they say this is all the more reason for concern: “Many of these features, including a vastly smaller number of simpler units, should make processors much easier to reverse-engineer than biological systems.” In other words, if standard neuroscience methods can’t even understand something as simple as an early gaming console, how can they explain complex neural mechanisms?

While the study didn’t test every tool in a neuroscientist’s belt, Jonas and Kording say the approaches they did examine are well-established and widely used in the field. The authors say their results indicate that “big-data” approaches to neuroscience may not live up to expectations and highlight a need for analysis algorithms better at dealing with heterogeneous and non-linear data.

Reception from the scientific community has been mixed, Kording told us: “Some people love it and say that it helps them focus on the important questions. Some people misunderstand it. Some hate it, as it makes it clear that neuroscience is not as advanced as a field as people like to believe.”

Featured image courtesy of Jeff Saltzman.