Graduate-degree viewers scored below a coin flip at telling a human video edit from an AI one

On a two-way choice, random guessing lands near 50%. Graduate-degree holders finished 15.2 points under that line, and bachelor's holders sat within a point of it ^[1][2]. Scoring well below chance takes a consistent rule aimed at the wrong cue. The column's authors say the higher earners and graduates used the same rule as everyone else, only with more confidence ^[5].

That explanation comes from the team that ran the study, writing as an Entrepreneur contributor; the publication notes that contributor opinions are their own ^[10]. The column reports accuracy by group but does not give how many of the 315 respondents fell into each group, or how confidence was measured ^[1]. The authors say the full methodology and dataset are published separately ^[6]. Until those subgroup counts are checked, the 16.2-point spread between bachelor's and graduate-degree holders comes from samples of unknown size ^[3].

The board-deck version of the study is that affluent, credentialed customers are easier to fool. That version leaves out the footage. The polish in this test came from the human editor ^[4]. If respondents were following a polish-means-AI rule, as the authors argue, a pair in which the AI added the graphics would have rewarded the same people, and the group applying the rule most consistently would have scored best ^[4]. The result describes one pair of edits of one piece of footage ^[1].

The authors carry the lesson past video, to any business assuming a customer segment will catch a shortcut or a quality drop ^[11]. The task, as described, was attribution: say which cut a human made ^[1]. The decision the authors say this assumption drives is which segment gets the polished version and which gets the rushed one ^[7]. That choice turns on whether customers notice or mind the cheaper version, and knowing who made a video answers a different question.

What holds up is the authors' account of their own habit. "We had never actually tested whether that assumption holds up," they wrote ^[8]. For an operator, the trade-off this quarter is a paid test on the specific content against a reset based on a rule of thumb. That rule could be the old one about discerning high earners or the reversed one this column invites. A reset made without a test this quarter leaves next quarter's results measured against a rule nobody checked on the company's own customers. The authors' recommended fix is to test the actual content or product against real people and watch what they do ^[9].