The Internet Observer Effect

Two weeks ago I offered some thoughts on Boaz Miller and Isaac Record’s argument about internet filter bubbles. Briefly, Miller and Record argue that, because the details of how Google adjusts your current search results based on your past search history are unknown, in order to justify a purported piece of knowledge learned through searching, you need to consult other sources of information. I had various objections, but overall I think their argument is correct, assuming that personalization really affects search results to the degree they suggest. In this post, though, I want to explore the problem from another angle: how does my search behaviour affect the search results of others—or more broadly, how does my behaviour on the internet affect the epistemic landscape of others?

Let me use an illustrative example. Back in November there was minor controversy on the University of Toronto campus when a “men’s rights” group invited The Myth of Male Power author Warren Farrell to speak. Predictably, campus feminists reacted strongly:


Anyways, the most vivid rallying cry was this quote from Farrell’s book:

We have forgotten that before we began calling this date rape and date fraud, we called it exciting.1

Yikes. But we all know that quotes can be taken out of context. Maybe what Farrell meant is that, in the past many women were forced to have sex against their will without their experience being acknowledged as rape. Farrell might mean that it’s a good thing that we finally call date rape what it is: rape. But I suspected otherwise. So I started thinking about how I could learn more. One way to learn more would be to buy the book. An obvious problem with this is that it gives support to someone I probably don’t want to encourage to write any more books. A subtler problem is that if, for instance, I buy the book from Amazon that purchase would be recorded, would affect the book’s sales rank, and ultimately make it more likely that others would purchase the book.

What if I just Google “Farrell rape” or something like that? Even that has the potential to affect others:

Google’s engineers have discovered that some of the most important signals can come from Google itself. PageRank has been celebrated as instituting a measure of populism into search engines: the democracy of millions of people deciding what to link to on the Web. But Singhal notes that the engineers in Building 43 are exploiting another democracy — the hundreds of millions who search on Google. The data people generate when they search — what results they click on, what words they replace in the query when they’re unsatisfied, how their queries match with their physical locations — turns out to be an invaluable resource in discovering new signals and improving the relevance of results. The most direct example of this process is what Google calls personalized search — a feature that uses someone’s search history and location as signals to determine what kind of results they’ll find useful. But more generally, Google has used its huge mass of collected data to bolster its algorithm with an amazingly deep knowledge base that helps interpret the complex intent of cryptic queries.2

My search behaviour on Google has the potential to affect not only my own future search results, but also those of everyone else who uses Google. So let’s say I Google “Farrell rape”. This could cause Google’s algorithm to associate “Farrell” more closely with “rape”. This could cause his website or book to be ranked higher when someone searches for “rape”. This could cause people to form beliefs about, for example, what “really” counts as rape.

As New York Times columnist Nicholas Kristof observes, North American attitudes towards rape are not as different from those in India as we might like to tell ourselves. If Farrell promotes mistaken views about women and rape, any activity that increases the prominence of Farrell or his views in Google searches will decrease the quality of information for those who use Google to learn about rape. And that, in turn, can only make the situation Kristof discusses worse.

The irony of my example is that this post is likely to have a far stronger effect than any search queries that I may have done (and in preparing this post I did many searches of this sort as well). I made the decision to not link to Farrell’s book, and not link to the website of the campus men’s rights group, but I did link to that Youtube video and the forum post containing Farrell’s quote. Even if I hadn’t, however, Google might have gained contextual clues from this post that would affect the search results of others. And just by clicking a link to read this post, even though you probably had no way of knowing that I would be discussing Farrell, you have increased the chances, however slightly, that someone else will.

I propose to dub this the “Internet Observer Effect”: any time you do anything on the internet, it affects the epistemic landscape of not just yourself, but everyone else who uses the internet.3 In physics, observer effects are (inevitable) changes we make to the universe by observing it. There is no way to observe the universe without changing it in some way; there is no such thing as truly passive observation. With the ubiquitous tracking and relentless data mining taking place today, it is just as impossible to be a passive observer on the internet.

One of the most troubling aspects of this is the decontextualization that takes place as Google distills a mind-boggling amount of data into an ordinal ranking of webpages. I can link to a page with context, warning readers of the pitfalls of taking that page’s content at face value. But does Google know that my link is not meant to bolster that page as an authoritative source? It seems unlikely. Google search results don’t come with disclaimers like “70% of the pages that linked to this site did it in a critical manner” or “80% of the users who visited this page judged it to be bullshit”.

In conclusion, I don’t want to seriously suggest that you should stop searching the internet for sensitive topics, or avoid clicking links that may lead to pages you disagree with. The effect of any single search are almost certainly negligible, and such behaviour would surely exacerbate the filter bubble problems discussed by Boaz and Isaac. However, I don’t think the Internet Observer Effect is merely a philosophical curiosity. Links on pages surely have consequences worth considering. We might also want to think about the ways that Google could ameliorate this problem, for instance by implementing a standard whereby webpage authors or internet searchers may indicate the quality level of pages they link to or visit. It is also worth bearing in mind that we are all shaping the epistemic landscape of the internet, whether we want to or not. There is no way to be a passive user of the internet, any more than there is any way to be a passive observer of the universe.

  1. The quote in context, I have since learned, is:

    If a man ignoring a woman’s verbal ‘no’ is committing date rape, then a woman who says `no’ with her verbal language but ‘yes’ with her body language is committing date fraud. And a woman who continues to be sexual even after she says ‘no’ is committing date lying.

    Do women still do this? Two feminists found the answer is yes. Nearly 40 percent of college women acknowledged they had said “no” to sex even “when they meant yes.” In my own work with over 150,000 men and women – about half of whom are single – the answer is also yes. Almost all single women acknowledge they have agreed to go back to a guy’s place “just to talk” but were nevertheless responsive to his first kiss. Almost all acknowledge they’ve recently said something like “That’s far enough for now,” even as her lips are still kissing and her tongue is still touching his.

    We have forgotten that before we called this date rape and date fraud, we called it exciting. Somehow, women’s romance novels are not titled He Stopped When I Said “No”. They are, though, titled Sweet Savage Love, in which the woman rejects the hand of her gentler lover who saves her from the rapist and marries the man who repeatedly and savagely rapes her. It is this “marry the rapist” theme that not only turned Sweet Savage Love into a best-seller but also into one of women’s most enduring romance novels. And it is Rhett Butler, carrying the kicking and screaming Scarlett O’Hara to bed, who is a hero to females – not to males – in Gone With the Wind (the best selling romance novel of all time – to women). It is important that a woman’s “noes” be respected and her “yeses” be respected. And it is also important when her nonverbal “yeses” (tongues still touching) conflict with those verbal “noes” that the man not be put in jail for choosing the “yes” over the “no.”

    It seems clear that Farrell’s opponents were not misrepresenting the intent of the original quote.

  2. Stephen Levy, “Exclusive: How Google’s Algorithm Rules the Web” (Wired, Feb 22 2010)
  3. Casually searching, the closest previous concepts I can find are an entry in Urban Dictionary for the “Internet Uncertainty Principle” (which is more specific but similar to my idea) and a Wikipedia entry on the “Observer effect (information technology)“, which is about the effect of debugging and log files on application performance.


  • Pete Reply

    Amazing article, as usual, Mike. Keep up the good work.

  • Boaz Miller
    Boaz Miller Reply

    Wow. Great post! Google’s auto suggest is a manifestation of this problem. I remember searching for a certain academic, when the words “sexual harassment” appeared as a suggestion. And shamefully I used the suggestion to see what this was all about. Just like that, a person’s reputation was damaged by an automatic algorithm that uses the crowd’s “wisdom”. Consider making this into an academic paper – you have a strong point, and don’t forget to quote us 🙂

Leave a Reply

Your email address will not be published. Required fields are marked *