A few weeks ago, I saw in an article on The Guardian website that the organization UN Women was running an ad campaign aimed a demonstrating how badly women are regarded around the world. The campaign centered on entering phrases like “women should” into a Google search box and seeing what suggestions Google made to “autocomplete” the search string. The idea was that, because Google has a massive database of the ways in which people generally finish search strings, the Google suggestions would reflect the most popular completions. Google’s suggestions were not terribly complimentary toward women. “Women should” was autocompleted with phrases like, “stay at home,” “be slaves,” “be in the kitchen,” and “not speak in church.”
After recovering from my initial horror, I thought that this might be an interesting approach to finding out about trends in popular belief more generally, so I decided to try it out the phrase, “psychology is.” The suggestions I got were: “not a science,” “bullshit,” “the study of,” “empirical,” and “useless.” These completions were not exactly shocking to me, but they are rather disheartening if you think (as many psychologists do) that the discipline has, over the last century-and-a-third, achieved a relatively secure status among the sciences.
I sent off to three psychology e-mail lists to which I subscribe the suggestion that other people might try this out. I did not relay my exact results, but I did indicate that the outcomes would be less happy than they might expect. Soon afterwards, one person wrote back saying that they didn’t understand what I was on about. The autocompletes they had gotten to “psychology is” were “defined as,” “the study of,” “best defined as,” “not a science.” Not exactly favorable – three of the four are incomplete sentences – but not nearly as negative as I had gotten: no “bullshit,” no “useless.”
Only then did I remember that the Google search engine does not give the same results for everyone. It customizes its responses based on the search history of the person doing the search. So, my question immediately became, how much variability is there between people in this kind of search? Had this one other person and I covered pretty much the entire range, just by coincidence? Or, was everyone going to be wildly different from each other? I decided to ask a number of people to try it out to see what would happen.
I am not one of these people with hundreds or even thousands of Facebook “friends.” I have “only” 104. Many of these people are other historians of psychology, several from my own school. Quite a few are historians of science, psychologists, and baseball researchers, along with a number of old friends who toil in a random assortment of professions. Not exactly a random sample. Nevertheless…
A total of 18 tried out my search and sent me their results within about 12 hours of my posting my question, bringing my total up to 20 (including myself and the person whose reply to my original post had first made me wonder what was going on). Of those 20, 12 were men, 9 of whom were psychologists, and 8 were women, 4 of whom were psychologists.
All but 2 of those 20 received the autocompletion “not a science,” making it far and away the most frequent suggestion overall. One of those not offered this completion was male, the other one female, and both were non-psychologists. The second most frequent autocompletion was “the study of,” with 11 of 20. We might assume that most of these sentences would have been completed with “mind and behavior” or some such, but we cannot be certain. Similarly, the fourth most frequent autocompletion was the quite similar “defined as,” with 8 of 20. Tied for fourth most frequent was “empirical,” also with 8 of 20.
The third most frequent autocompletion was “bullshit.” Fully half of the people who completed my request received this suggestion. Exactly half of each sex drew it as well (6/12 men, 4/8 women). However, 8 of the 13 psychologists received this suggestion while only 2 of the 7 non-psychologists did. (For the statistically-minded, a two-way ?2(1)=1.4, NS. But the n is pretty low here.)
Then Alice White, a historian of science in the UK, sent me a Slate article by David Auerbach about the UN Women ad campaign and the use of autocomplete as a research method. According to him, “Google no longer returns many of the autocomplete results in the ads.” This is because “Google’s policies state that they will remove hate speech completions, among others…. In contrast to a few weeks ago, Google now refuses to complete ‘women shouldn’t have righ’ and ‘women need to be put in their plac’ up to the final letter.”
We can presume, I think, that Google does not adjust their autocomplete suggestions to protect psychology in the way they do to protect vulnerable groups of people. Still, the whole package of complications calls into question how seriously we can take the results of this sort of, admittedly appealing, approach to digital social research. The autocompletes are customized by the person doing the search, and Google intervenes to remove autocompletes they deem offensive.
Still, half of people responding to my question got the suggestion: psychology is… “bullshit.” I would like AHP’s readers to try this out as well. Please post your results as a comment below.