twelve online experiments find our perception of faces is biased by preceding videos of the same face changing slowly. this interesting finding is hyped, first by the authors, second by the media, into time-warping nonsense.  

original article: Manassi & , Whitney, 2022 (Science Advances), reported in: The Daily Mail by Shivali Best on 1st February 2022

this story was in episode 24 #vision #face #illusion #time #memory

the error bar says

starting 2022 without a single CAPITALISED word, the Daily Mail tells us that our perception of the visual world is 15 seconds out of date; that our brain shows us images from the past rather than trying to see the unpredictable present.

in twelve experiments, 4132 people judged the age or gender of a face on a computer screen. before making their judgement, about half of these people were shown a video of a similar face, changing slowly from younger to older, or older to younger, or maler to femaler, or femaler to maler.

seeing these slowly-changing videos biased the following age- or gender-judgements of the people. a video of a face changing from 13 to 25 years old made people judge the 25 year old as being only 20. a video rejuvenating from 25 to 13 years, made the 13 year old appear 18. somehow.

a similar thing happened with a face changing slowly from male to female or female to male. because the videos were 30 seconds long, the researchers claim that, actually, we are seeing things as they were 15 seconds ago.


are we 15 seconds out of time?


much like the UK government's approach to breaking international law: our visual brain is only 15 seconds late in a "very limited and specific way". let me explain.

first, the 15 second claim relies on cherry-picking the very largest effect from the study. the average reported effect is around 11 'seconds'. i mean, if you insist on using that way of describing the effect.

second, you shouldn't. it's not at all about the number of seconds in the past. in one experiment, the researchers added an additional 15 seconds delay after the video & before the face judgement was made. this delay made no difference - people were still biased in judging the face's age. so it's not simply about time.

thirdly, nerdly, the experiments were done on-line, advertised as a 'very very short survey'. online experiments can be fine, but it's no substitute for proper in-person, controlled laboratory studies where the experimenter can, for example, check that the participant is watching the video.

fourth, all of the age judgements were biased in the same direction - the 25 & a half year old face was always rated younger & the 13 year old always older, even in the control conditions without the slowly-changing videos. these baseline biases account for about 3 'seconds' of that magic 15 seconds number. which, again, is not about the number of seconds, if you remember my second criticism, from about 30 seconds ago.

fifth, the gender version of the experiment did not work nearly as well as the age version, despite the authors' exaggerated claims. indeed, one of the two critical numbers was fudged - not 'statistically significant'.

finally, the faces used were very artificial - hair & ears were removed, they were presented within a blurred oval window. & worse, the cherry-picked '15 second' result was only found when the highest-level of noise was added to the images. it worked much less well with clear, sharp images.


if you ignore the hype, this is a solid study. it just emphatically does not show that we are living 15 seconds in the past. rather: our facial judgements are biased, in limited, specific & interesting ways, by what we've just seen