What can the brain of an Atlantic salmon, frozen and thawed, tell us when shown pictures of human faces? In 2009, a team of researchers placed a lifeless specimen inside an MRI scanner for an experiment that looked like a laboratory joke. Yet, against all expectations, the screen lit up, revealing real brain activity in clearly identifiable regions. This seemingly impossible result became one of modern science’s most fundamental lessons, exposing a glaring flaw in the way we interpret complex data.
What you will learn
-
Why a scanner detected signals of “thought” in a fish that had been clinically dead for several days.
-
The danger of false positives in scientific studies dealing with thousands of data points simultaneously.
-
How this spoof study forced scientists around the world to change their calculation methods.
A Dead Salmon in Front of Vacation Photos
The story begins with a technical frustration. Craig Bennett and his team wanted to test the settings of their functional magnetic resonance imaging (fMRI) machine.
Rather than recruit a human volunteer, they bought a whole salmon from the local supermarket. They placed it in the magnet’s bore and, for fun, showed it a sequence of images depicting people in various social situations.
The goal was simple: to ensure the machine detected absolutely nothing on an inert object before moving on to real tests.
That’s where the unthinkable happened. While analyzing the fish’s brain images, researchers saw red and yellow spots appear, signs of intense metabolic activity. At that exact moment, by the numbers, the salmon seemed to be “evaluating” human emotions.
The Statistical Mirage of Thousands of Pixels
How could a decomposing brain process visual information? The answer isn’t biology but mathematics.
An MRI scanner divides the brain into more than 130,000 small data blocks called voxels. For each voxel, the machine looks for a correlation between a stimulus (the image shown) and a change in signal.
The problem is that across 130,000 trials, pure chance will inevitably create coincidences. This is what statisticians call “noise.”
Without a rigorous mathematical filter, random chance can conjure patterns where there is only emptiness. The salmon wasn’t a zombie; it was simply statistical proof that you can make data say almost anything if you don’t correct for it properly.
A Parody That Revolutionized Medicine Worldwide
Bennett’s team faced heavy resistance to publishing the study, with many journals fearing it would ridicule the discipline.
Yet the impact was massive. Before this salmon, many studies of the human brain published spectacular results based on similar calculation errors, often without realizing it.
The experiment highlighted the crucial importance of “correction for multiple comparisons.” Today, no imaging study can publish results without demonstrating that the background noise that made the salmon “think” has been eliminated.
This study, which even earned an Ig Nobel Prize, reminds us that science must doubt itself, even in the face of apparently indisputable visual evidence. It remains the best educational example to explain that seeing does not always equate to believing.
