Doesn’t that suggest professional ethics are immoral? They block a temporary, one off experiment which could suggest how to avoid sadness.

The alternative is relying on intuition only and possibly making people permanently sadder. Or is there a way to achieve this knowledge consistent with the field of research ethics?

We saw this with challenge trials in the pandemic where ethicists fretted people would reject them and concluded challenge trials are immoral. Meanwhile opinion polls suggest people view challenge trials as virtuous and the current method of waiting for infections as immoral.

Yes, there is a way. You left them specifically opt in to it, you provide them disclosure documents of what's being tested, and you provide free access to mental health resources.

It would be interesting to find out if these experiments lead to any suicides etc. Although I suppose that Facebook is not interested in releasing that type of information.

Which drop of rain caused the dam to break? In the book Good Omens there is a demon who considered himself the best not because he was the best deceiver or torturer, but because of his ability to cause mass malcontent.

The real evil isn’t in the experiment, it’s that the researchers found that Facebook makes people’s lives worse. How many suicides are caused by Facebook’s investors? Whether they can mitigate it this way or that is splitting hairs.

Isn't every experiment temporary one off? Ethics are strict bc of the dark history of human experimentation.

> Doesn’t that suggest professional ethics are immoral? They block a temporary, one off experiment which could suggest how to avoid sadness.

NO, companies should not perform secret psychological experiments on people without their consent because they "could suggest how to avoid sadness."

> challenge trials in the pandemic

You mean the ones people consented to? Why is consent not an issue for you?

Which part of this conception of experimental ethics does this violate? Facebook's experiment did not directly cause the emotions, the posted content for that. Even if it did, this sort of thing is well within the bounds of normal experimentation with large systems.

This is similar to claiming that traffic experiments which change the flow of cars, in term causing some people to experience more traffic and become unhappy, are unethical. Do you think traffic experiments are unethical?

> Facebook's experiment did not directly cause the emotions, the posted content for that.

Facebook definitely caused the emotions - by deliberately choosing to show particular content.

> This is similar to claiming that traffic experiments which change the flow of cars, in term causing some people to experience more traffic and become unhappy, are unethical. Do you think traffic experiments are unethical?

If the objective of the traffic experiment had been to deliberately get a certain group of cars stuck in a traffic jam, this would absolutely be unethical.

> If the objective of the traffic experiment had been to deliberately get a certain group of cars stuck in a traffic jam, this would absolutely be unethical.

A closer analogue would be a traffic experiment designed to gauge the emotional effects of a particular route. That's an important difference.

The article itself says the experiment was designed to look for evidence of emotional contagion, which is quite different from "it was designed to make people sad".

Also, in another thread it's pointed out that the effect sizes from this study were extremely small - something like 0.3% more negative words were used by ~150k people. The effect is said to be on the same scale as any minor UI change, like a size/color change of the "like" button. So it's hard to see this as anything other than folks looking for a reason to get outraged.

> The article itself says the experiment was designed to look for evidence of emotional contagion, which is quite different from "it was designed to make people sad".

That was their research goal, yes. Their methodology was this:

> For some, that meant 90% of all "positive" posts were removed from their newsfeed for a week, rendering the social network a pit of despair.

Research can be unethical even if the intention of the research is not.

For the traffic analogy, if part of that traffic experiment would involve a routing change that researchers know will likely cause a traffic jam, then the experiment will be unethical, no matter how warranted the research may be.

That's the reason you cannot use placebos to test life-saving medication (without prior informed consent), even though it would certainly be beneficial for science if you could.

And in Facebook's case, their research goal wasn't even that ethical in the first place. Maye we can have this talk if they try to find a cure for depression through emotional manipulation, but this was literally just about preventing people from leaving Facebook:

> "At the same time, we were concerned that exposure to friends' negativity might lead people to avoid visiting Facebook."

Requirement for a clear research protocol statement from the outset, including clear timetable and detailed procedures.

Protection of clinical trial subjects

Clinical trials on minors

Clinical trials on incapacitated adults not able to give informed legal

Establishment of mandatory Ethics Committees

I disagree. If facebook is testing content like this then they explicitly know which content makes people sad. To manipulate someone's feed to push them stuff that knowingly makes them sad for the purpose of experimenting with someone's emotions, that is highly unethical to me.

There is a massive difference in studying and directing traffic to help make traffic flow better. The goal there is to improve driving for everyone. Facebook's goal is to intentionally make people unhappy for the sole purpose of making more money.

How would it have been ethical?

If their current process was already suspected to be the worse of the two, and their goal was to improve peoples' moods per se, what is the ethical thing to do? Keep people feeling more sad?

Or, perhaps, stop being algorithmic in what people see, or else put people in full control of simple sorting (prioritize my "starred" category and sort by date)?

Maybe it's unethical in academic context, but in public service context, you have an existing service, and you're in charge of improving it.

When you have an improved algorithm, would you change your algorithm all at once for everybody, or would you test your change on small % of users, before you go full-scale? You'd do the latter. There's no better option. Because if you're wrong, your impact is much more devastating.

In other words, there's always an algorithm filtering your feed. If that was unethical, it makes Facebook unethical by merely existing.

> in public service context

This is not a "public service" - it's a for-profit company.

> When you have an improved algorithm,

This is not an "improved algorithm". They deliberately removed almost all positive news from some people's feeds for a week.

I meant public service as in it's a service, and it's providing itself to the public. Not as in government service.

To be fair, the article states that the study was designed by a data scientist. Not that that implies that any additional ethical considerations were taken into account.

Data scientist means someone who's had classes in stats and knows how to use software to analyze data. They don't cover how to run experiment and don't cover ethics

To be fair, the profile of a data scientist at a F500 non tech company and at a company like FB is slightly different. Facebook's data science crew is known for being heavier on the stats lighter on the software (relatively speaking) and probably can design a good experiment.

That said, anyone who signs up for a job at Facebook doesn't give a shit about ethical design of experiments.

Hi, I used to work on ML and ran experiments like this one at FB. Just want to point out that the effect of experiments like this on an individual is smaller than the effect of making the app start 0.1 seconds faster, or retraining the network used to filter hate speech. The only reason the effect is significant is because N>650k, so tiny effect sizes can be detected.

whats the variance on that effect?? doesn't a large data set like that just lead to type ii errors?

Believe it or not, they didn't have the front end devs design it.

There was a specific task force at Facebook intended to better understand customer psychology. Came from them.