[momeunier]

I don't agree with your proposed way of studying this. You're trying to study a symptom and then conclude on the cause of the symptom. That makes no sense. If you want to prove that Facebook is listening, then prove that Facebook is listening. And don't try to prove that Facebook is listening and then targeting you with ads based on what you said. There are way too many levels of indirections that can trigger false positive for a vast number of reasons you don't control at all. What you are trying to do is replicate in large-scale displays of anecdotes but with a slightly more controlled environment. That won't prove anything since it will just be anecdotes and will again not sustain rational explanation by experts of Facebook ads mechanisms.

Without getting into the ads delivery part and the anecdotes, how would you prove that Facebook is listening? How would you prove that there is a set of information taken from your speech or your audible environment transferred to Facebook.

[logicallee]

I think you didn't read my proposed methodology carefully.

    Box     Researcher    | Owner of phone (outside room)
  [     ]~       x        |      o

The researchers speaks about some subjects, but not others. The ~ represents that in some cases the researcher's conversations are being fed into the box, and in others they are not. The box is otherwise soundproof, and inside is the phone being tested.

We can pick subjects such as:

   - #1 Adult incontinence
   - #2 Cat food
   - #3 Last-minute trip
   - #4 ..
   -    ..
   - #10

The test group is that the researchers' voices are being fed into the box. The control group is that researchers voices are NOT being fed into the box.

It is important in order to maintain double-blind environment that the researchers not hear whether they are being amplified into the box.

The results might potentially look like this:

https://imgur.com/a/y7852

Of course, I just made this up. (I imagine the subjective 1-5 scores being whether the given subject reports seeing such an advertisement, from 0 definitely not to 5 definitely yes.) I even made subject 3 unsure about topics 1 and 3 to mimic that humans are fallible. Likewise subject 2 does not really report any advertisements. (This is likely in the real world - for example subject 2 could be explicitly excluded by advertisers for some reason.)

The attached is the kind of graphs that I would expect based on dozens of scientifically-minded people trying them.

If these are the two graphs that we got, and if the test and control groups were truly randomized, what other explanation could you offer?

Of course, my proposed experiment is orders of magnitude more scientific than what people are doing with their n=1, unblinded personal experiments. But theirs has some validity also.

[momeunier]

Thanks for explaining further. That would actually be an interesting experiment. Who's going to run it?

[logicallee]

Not sure who would run it. Nobody really cares that much.