The coolest academic research paper of a century was published in 2018. In this paper, “Parachute use to prevent death and major trauma when jumping from aircraft: randomized controlled trial” seven US researchers established that jumping from planes without parachutes is as safe as with them.

There was, however, an important detail. The planes in their experiment were not flying high in the sky. They were parked on the ground. The participants in their study only had to jump one meter.

Obviously, nobody was injured – both in the “parachute group” and in the “no-parachute group”. The researchers then absurdly concluded that their study showed parachutes offer no benefit. This brilliant, sardonic paper is a perfect warning about how a study can look scientific but be completely misleading because of a deliberately flawed setup.

Every week, I read new "groundbreaking research" that claims to show that AI is “pure hype”, “not important” and “PR stunt”. These studies call out AI’s mistakes, like being unable to count the letter 'r' in the word 'strawberry' or failing at simple math. These “critical reports” often use the same intellectually dishonest logic as the parachute paper. They are designed to make AI look foolish by testing it in unfair or irrelevant ways.

For instance, skeptics create a pseudo-scientific test to prove AI has “no real intelligence”. They give the AI an impossibly difficult problem, like a complex mathematical puzzle that even the world's greatest geniuses have not solved. Unsurprisingly, the AI fails this ridiculous task. The skeptic then triumphantly announces that their "research" proves AI is not truly smart. This is a ridiculous conclusion, because the test was specifically designed to be impossible.

Another common tactic is to show that AI is dangerously unreliable. A researcher might spend weeks creating a very specific and bizarre situation to fool an AI. For example, they might design a strange sticker to put on a Stop sign, knowing its unique pattern will confuse a self-driving car's vision system. They then present this one, highly artificial failure as proof that the entire technology is flawed. This misleading approach completely ignores the millions of times the AI works perfectly in normal, real-world conditions.

Finally, we see dishonest reports about AI chat systems producing nonsense. In these cases, the critic will deliberately ask the AI strange or illogical questions. They keep trying until the AI gives a silly or factually incorrect answer. They then present these cherry-picked, bad answers as typical examples of the AI's performance. This method of showing only the worst results is designed to create a false impression, just like the joke parachute paper did. It is a warning to be critical of research that seems too simple or too shocking.