The Facebook Research

The Facebook experiments are actually very clever.

The content stream is the presentation of everyone else's material to an individual user. So, in my content stream, I get stuff from Rod, stuff from my family, stuff from Moncton Free Press, stuff from the City, etc., including some sponsored or promoted content from magazines and advertisers.

Facebook has been tinkering with this content stream since day one. You can't show everything, because there's generally too much. So you show the most 'relevant' links (on some content streams you have the option to 'see most recent' and 'see most relevant'). They are experimenting with what counts as 'relevant'.

Facebook is an advertising company, and therefore the product it sells is the induction of beliefs in the users. Coke, for example, want people to think that Coke is good and good for you, and that they want a Coke now. The NRA wants you to believe that guns are harmless and that "they're trying to take away our freedoms."

So the experiments basically measure whether the presentation of posts created by friends and family, etc., rather than the creation and presenting of actual advertising, can produce the desired result. Intuiitively, it should. Present nothing but crime stories in the news feed and you'll end up thinking crime is everywhere. The experiments measure whether this intuition is correct.

From the perspective of ethics, they are blending two things which are, on the face of it, innocuous:

- they are altering stream results in an attempt to produce a 'better' stream - something every content vendor everywhere does, and has done since the days of FTP and UseNet

- they are accessing publicly available data to analyze it for affective and cognitive properties, something we do as well, and something that does not require permission from individual users

Does the combination of them create an ethical dilemma?

It's not clear to me that it does. Sure, it reinforces Facebook's image as a somewhat greasy operation that will manipulate results in order to satisfy the needs of its advertisers (hence making it no different from Dr. Oz and your local news broadcast). But that's not unethical, at least, not in the sense that you'd take them before the courts.

The question of ethics comes into the equation when there exists some possibility of harm, and where that harm is a predictable outcome of the experiment, and where that sort of harm would not normally be expected. The classic case, of course, is the testing of drugs that have harmful side-effects, where you have not disclosed the side-effects.

In the case of the manipulation of free digital content to stimulate emotional responses, and then measuring for those responses, the presence of actual harm is a lot more difficult to show. The mere production of emotional responses is not harm, otherwise most of what we do every day is ethically wrong. The mere measurement of emotional responses is not harm either.

If we don't actually harm someone, then how could it be ethically unsound? Doing all this in secret could be ethically wrong. But Facebook is not doing it in secret; it's all over ther news.

There is a hard line in research ethics to the effect that any interaction with a user needs to be declared beforehand, and conducted with the explicit consent of the user. I don't subscribe to that line. In one sense it is impractical. There are too many interactions and too many users to require consent in advance. In a second sense it's unnecessary. Research is not inherently evil, and studying people to find out how they work is not wrong. And third, it can be harmful. Creating conditions of consent alters research results; tell people their emotions are being monitored and they change their emotions.

This is the point of disagreement:

"They formulated a research hypothesis and tested it on human subjects. For this, explicit consent is required."

Here is a counterexample:

Engineers have theories regarding the length of left-tern lanes on the highway. To test this hypothesis, they construct a left-turn land, and then measure ho much it underfills or overfills. Based on this work they publish a paper. No research consent is obtained.

Should consent have been obtained? It fails the three tests. First, it is impractical. You can't have drivers fill out a consent form befor they enter the intersection. Second, it is unnecessary. No harm will be caused by the research. And third, requiring consent changes the outcome.

So it seems clear to me that this statement is false. The requirement for explicit consent must depend on different conditions. I argue actual harm must be cause, that it must be practical to obtain consent, and that obtaining consent can't change the results.

Facebook's experiments on users and Emotional Contagion (via Peter Turney)

https://gist.github.com/desilinguist/35e0a9f4abae47157104

The full paper -http://www.pnas.org/content/111/24/8788.full

http://www.telegraph.co.uk/technology/facebook/10932534/Facebook-conducted-secret-psychology-experiment-on-users-emotions.html

http://www.theguardian.com/technology/2014/jun/30/facebook-emotion-study-breached-ethical-guidelines-researchers-say

Listen to your customers, not to the HiPPO - http://videolectures.net/kdd07_kohavi_pctce/