The Senators demanding investigation and the article itself are both clearly implying Facebook has done something wrong. The accusation seems to be that Facebook used the data to target ads. I suspect that's true only in the sense that it went into the same generalized funnel all the URLs do and not that they're specifically parsing out this information, but I guess those details don't matter if your reputation stinks.

How do you know that they are not parsed out by Meta? The possibilities for this goes from collusion with the tax firms over somebody at Meta reverse engineering the coding to an automatic "AI" reverse engineering ("pattern discovery" Not sure if they do that but possibly?).

I don’t “know” that which is why I said I “suspect” it rather than “know” it. But it seems like such an obviously foolhardy move with such marginal gain to deliberately seek to collect and analyze that data from tax prep firms that I have a hard time imagining them signing off on it.

My point is that they might not have to sign off on it. The signal for the ads might have been discovered by an algorithm without making it explicit to the management.

I would not be surprised if google, meta and co don't have big ML codes that try to find signals to increase ad revenue, and they dump all data they have into it.

"It's an algorithm! Nobody knows how or what went into creating it!" Seems like the new version of "I'm in charge of making the rockets go up; where they come down is a different department."

Meaning, of course, people are in charge of the machine learning / algorithm, people manage the weights, they measure the outcomes, they exercise editorial control. It's not a mystical black box given to us by god or a super advanced alien race; it's people doing people things; if they get the profit they also get the liability / responsibility. Even if they're not sure how it works, they sure as hell still responsible for it.

I am not arguing against them having responsibility. They do. I am arguing that they might not have realized what the weights are doing. Because, yes, indeed, large neural networks ARE black boxes.

I am not sure how it would be workable for Facebook to be responsible for what information third parties choose to put into their tool. By that logic pretty much any application that allows users to upload text is equally guilty unless each post in manually reviewed.

I can easily imagine Facebook's machine learning algorithms picking a signal out of this data and using it to target ads. They may not have had to tailor a query to search for "size of tax deduction". The machine learning algorithms did that on their own. In a completely opaque manner.

I doubt it is anything anywhere near that fancy. Look at the primary source of this complaint, which is available in GitHub. One of the sites was putting sensitive data, such as income amounts, in the page title. The Facebook analytics thing sends the page title to Facebook. Google Analytics also takes the page title. You can easily see how any old targeting engine, machine inference or otherwise, would take the titles of visited pages as a signal.

Why on earth would you give Facebook the benefit of the doubt on anything?

Based on experience working in a similarly-sized company I have some guesses as to how they might operate. Is there a logical reason I should always assume the least favorable thing instead?

It's definitely on the tax company for sending extraordinarily sensitive data to google analytics.

However I'd say sending sensitive (or at least private) data to google analytics is more or less its intended purpose. I don't think the use and continued existence of google analytics is compatible with any reasonable privacy protection.